Feb 3, 2017 - 4.6 Techniques for Preliminary Analyses of Data. 113-116. 4.7 Empirical Framework. 117-142. 4.7.1 Panel Data-Least Squares ( LS ) Models.
DETERMINANTS OF CAPITAL STRUCTURE : AN EMPIRICAL STUDY OF SELECTED INDIAN MANUFACTURING COMPANIES
A THESIS SUBMITTED FOR THE DEGREE OF DOCTOR OF PHILOSOPHY IN COMMERCE OF THE UNIVERSITY OF KALYANI WEST BENGAL
BY
SANDIP SINHA Research Scholar , Department of Commerce University of Kalyani , Nadia , West Bengal -741235 Registration No. : KU / Ph.D. – 0073 of 2015
UNDER THE SUPERVISION OF
DR. PRADIP KUMAR SAMANTA ASSOCIATE PROFESSOR , DEPARTMENT OF COMMERCE UNIVERSITY OF KALYANI KALYANI , NADIA , WEST BENGAL -741235
ii
iii
iv
PLAGIARISM
REPORT
v
v
ACKNOWLEDGEMENT First and foremost, I would like to express my deep and sincere gratitude to my supervisor Dr. Pradip Kumar Samanta, Associate Professor in Commerce, University of Kalyani for his continuous support and guidance during the course of my Ph.D. work. Apart from initiating serious academic discussions regarding my work, he has encouraged and coaxed me to attend numerous conferences and seminars to present papers, primarily, on the most important aspect of my work ( i.e. the application of the complex econometric technique of quantile regression ). The constructive criticisms and encouragement received, at these conferences and seminars, from distinguished fraternities from India and abroad, have immensely enhanced my confidence in proceeding with my doctoral work and broadened my academic outlook and horizon. I would also like to express my sincere gratitude to all the distinguished faculty members of the Department of Commerce, University of Kalyani for the support and guidance offered by them during the period of Ph.D. course work and also for the positive and constructive suggestions offered by them during my ( Ph.D. ) pre-submission seminar. I would like to thank all my fellow research scholars of the Department of Commerce and Business Management, University of Kalyani for maintaining a cordial and friendly atmosphere amongst us. I would like to thank all the office staff of the Department of Commerce, University of Kalyani for their support in official matters regarding my doctoral work. I would also like to sincerely acknowledge the service of Mr. Sibshankar Mistry , who has diligently supplied, with great care, numerous cups of tea during my visits to the department of Commerce, University of Kalyani. Last but not the least, I would like to express my deepest gratitude to my parents, betterhalf and little daughter for their continuous support, sacrifices and care which has enabled me to continue with my doctoral work peacefully.
Sandip Sinha
vi
Table of Contents Particulars
Page
Title of Thesis
i
Supervisor’s Certificate
ii
Declaration by Research Scholar
iii
Plagiarism Report
iv
Acknowledgement
v
Table of Contents
vi
Executive Summary
ix
List of Tables
xii
List of Figures
xiii
Chapter 1 : Introduction
1-9
1.1
Background
1
1.2
Concept of Capital Structure
2-5
1.3
Capital Structure Decision
5-7
1.3.1
Concept
5
1.3.2
Factors Affecting Capital Structure Decision
5
1.4
Plan and Chapterisation of Research Study
8-9
1.4.1
Plan of Research Study
8
1.4.2
Chapterisation of Research Study
9
1.5
Conclusion
Chapter 2 : Review of Theoretical Literature 2.1
Introduction
2.2
Capital Structure Theories
2.2.1
9
10 - 33 10 10 - 33
Traditional Theoretical Approaches
11 - 14
2.2.1.1
Net Income ( NI ) Approach
11
2.2.1.2
Net Operating Income ( NOI ) Approach
12
2.2.1.3
Intermediate ( or Traditional ) Approach
13
2.2.2
Modern Theories of Capital Structure
15 - 33
2.2.2.1 Modigliani and Miller ( M-M ) Hypothesis
15
2.2.2.2 Trade - Off Theory ( TOT )
16
2.2.2.3 Agency Cost Theory ( ACT )
21
2.2.2.4 Signaling Theory ( ST )
25
2.2.2.5 A Synthesis of Trade-Off Theory, Agency Cost Theory and /or Signaling Theory
28
2.2.2.6 Pecking Order Theory ( POT )
29
2.2.2.7 Market Timing Theory ( MTT )
33
2.3
33
Conclusion
vii
Table of Contents Particulars Chapter 3 : Review of Empirical Literature
Page 34 - 91
3.1
Introduction
3.2
International Studies
34 -73
3.3
Indian Studies
74 - 90
3.4
Conclusion
Chapter 4 : Research Methodology 4.1 Introduction 4.2 Research Problem(s) and Research Objective(s)
34
91
92 - 146 92 94 - 97
4.2.1
Identification of Research Problem(s)
94
4.2.2
Objectives of Research Study
97
4.3 4.4
Determinants of Capital Structure Formulation of Variables
99-107 107-110
4.4.1
Dependent Variable
107
4.4.2
Independent Variables
109
4.5
Data and Sample Selection
110-113
4.5.1
Type of Data
110
4.5.2
Source of Data
111
4.5.3
Selection of Sample
111
4.6 Techniques for Preliminary Analyses of Data 4.7 Empirical Framework 4.7.1 Panel Data-Least Squares ( LS ) Models
113-116 117-142 117-131
4.7.1.1
Formulation
117
4.7.1.2
General Assumptions
118
4.7.1.3
Specific Assumptions and Estimation of Random Effects ( RE ) , Fixed Effects ( FE ) and Pooled Ordinary Least Squares ( POLS ) Models
119
4.7.1.4
Tests for Choosing Appropriate Panel Data Model
123
4.7.1.5
Tests for Heteroscedasticity , Serial Correlation , Cross- sectional
125
Dependence and Multicollinearity in Panel Data Models
-
4.7.1.6
Robust Standard Errors
129
4.7.2
Quantile Regression ( QR ) Methodology
4.7.2.1
Conditional Quantile Regression ( CQR )
132
4.7.2.2
Unconditional Quantile Regression ( UQR )
134
131-140
4.7.2.2.1
Concept of UQR
134
4.7.2.2.2
Panel Data Recentered Influence Function ( RIF )-Least Squares ( LS ) Models
137
4.7.2.2.3
Estimation of RIF-LS Model
137
4.7.2.2.4
Bootstrapped Standard Errors
139
4.8 4.9
Statements of Hypotheses Conclusion
141 141
viii
Table of Contents Particulars Chapter 5 : Analyses and Findings 5.1 5.2.
Introduction Preliminary Analyses of Data
Page 143-189 143 143-153
5.2.1
Descriptive Summary of Statistical Measures
143
5.2.2
Correlation Analysis
148
5.2.3
Panel Data Unit Root Test
149
5.3 Panel Data Regression Analysis 5.4 Panel Data Unconditional Quantile Regression Analysis 5.5 Conclusion Chapter 6 : Concluding Remarks 6.1 Introduction 6.2 Overview of the Study 6.3 Limitations of the Study 6.3 Suggestions for Future Researches
150-156 157-187 188 189-199 189 189 198 199
References Appendix A : Names of Companies Constituting Final Sample Appendix B : Publication Appendix C : Papers Presented in Conferences and Seminars
200-213 214-220 221 222
ix
Executive Summary This research study deals with the primary research topic of ‘corporate capital structure’ and includes the following integral elements : ( 1 ) Research Problem , framed after a detailed review of theoretical and empirical literature on various aspects of corporate capital structure : To re-examine the issue of non-linearity in capital structure choices of firms through the application of the hitherto un-applied technique of ‘ panel data unconditional quantile regression’ with a view to indirectly ascertain the applicability of Trade-off Theory , Agency Cost theory and Pecking Order Theory through the analysis of the predictions of these theories on the firm-specific determinants of capital structure of Indian manufacturing firms. ( 2 ) Research Objectives : ( A ) To choose the appropriate panel data model from three mutually exclusive panel data models ( Pooled Ordinary Least Squares Model, Fixed Effects Model and Random Effects Model ) for analysing the impact of firm-specific determinants of capital structure on the mean of the variable measuring capital structure ( say, “ V ” ) in respect of Indian manufacturing companies listed on the Bombay Stock Exchange ( BSE ) , with a view to indirectly assess the applicability of Trade - Off Theory ( TOT )1 , Agency Cost Theory ( ACT ) and Pecking Order Theory ( POT ) for an average company with mean ( average ) level of leverage. ( B ) To apply ‘unconditional quantile regression technique’ on the chosen panel data model for analysing the differential impact ( that is , non-linear behaviour ) of the firm - specific determinants of capital structure over the entire unconditional distribution of the variable “ V ” in respect of Indian manufacturing companies listed on the Bombay Stock Exchange ( BSE ) , with a view to assess indirectly the applicability of Trade - Off Theory ( TOT ) , Agency Cost Theory ( ACT ) and Pecking Order Theory ( POT ) at different quantiles of the unconditional distribution of V, that is , for representative 2 companies having varying ( for example , ‘very low’ , ‘low’ , ‘moderate’ , ‘high’ or ‘very high’ ) levels of leverage.
1 2
Traditional Static Trade-Off Theory or Tax Shield-Bankruptcy Cost (TS-BC) Theory.
A representative company at a particular quantile of leverage represents a company with the corresponding level of leverage , for instance , a representative company at the 50th quantile represents a company with median ( or moderate ) level of leverage , or a representative company at the 10 th quantile may be said to represent a company with very low level of leverage.
x
( 3 ) Formulation of Variables :( a ) Dependent Variable : Total Debt to Capital Ratio in terms of market value and book value { that is Market Leverage Ratio ( MLEV ) and Book Leverage Ratio ( BLEV ) } ; ( b ) Explanatory Variables : Firm-specific determinants of capital structure, represented by : ( i ) firm size ( SIZE ) , ( ii ) tangibility ( TANG ) , ( iii ) non-debt tax shield ( NDTS ) , ( iv ) profitability ( PROF ) , ( v ) growth opportunities ( GROW ) , ( vi ) bankruptcy risk ( distance from bankruptcy ) { DFB }, and ( vii ) liquidity ( LIQ ) ; after controlling for the unobserved industry effects and the unobserved macro-economic and institutional factors through the inclusion of Industry Dummy Variables and Time Dummy Variables . ( 4 ) Sample : The final sample include a balanced panel data of 601 Indian manufacturing companies listed on the Bombay Stock Exchange ( BSE ) , comprising all the eleven broad industrial divisions ( as classified in the Prowess IQ database ) , over a period of 18 years from 1997-98 to 2014-15. ( 5 ) Econometric Models : The study seeks to apply Unconditional Quantile Regression ( UQR ) methodology to the sample data. However , as the basic structure of the data for this study is that of a panel data-type, the baseline econometric model will be a Panel Data -Least Squares ( LS ) regression model which will be chosen from the Pooled Ordinary LS model, Fixed Effects model and Random Effects model after performing the appropriate tests. Unconditional Quantile Regression , based on Recentered Influence Function ( RIF ) will then be applied to the chosen panel data model. ( 6 ) Research Hypotheses : The generalized Null Hypotheses ( H0 s ) to be tested in this study are stated as follows : ( A ) Panel Data Least Squares ( LS ) Regression :H0 : There is no statistically significant impact of firm - specific determinant ‘X’, industry division dummy variables ( IND2 , IND3 ,…, IND11 ) , and time period dummy variables ( D2 , D3, …, D18 ) on market leverage ratio ( MLEV ) or book leverage ratio ( BLEV ) in respect of the average Indian manufacturing company { that is , company with mean ( average ) level of leverage } listed on the Bombay Stock Exchange , where X = SIZE , TANG , NDTS, PROF , GROW , DFB and LIQ , and statistical significance refers to 1% , 5% and 10% levels of significance.
xi
( B ) Panel Data Unconditional Quantile Regression [ or Panel Data Recentered Influence Function - Least Squares ( RIF- LS ) Regression ] :H0 : There is no statistically significant impact of firm - specific determinant ‘X’, industry division dummy variables ( IND2 , IND3 ,…, IND11 ), and time period dummy variables ( D2 , D3, …, D18 ) on market leverage ratio ( MLEV ) or book leverage ratio ( BLEV ) in respect of the representative Indian manufacturing company listed on the Bombay Stock Exchange at the th unconditional quantile of MLEV and BLEV { that is , company with th quantile level of leverage } , where X = SIZE , TANG , NDTS PROF , GROW , DFB and LIQ ; = 10 , 25 , 40 , 50 , 60 , 75 and 90 ; and statistical significance refers to 1% , 5% and 10% levels of significance. ( 7 ) Findings of the Study : ( A ) Fixed Effects Regression : Firm size , tangibility , non-debt tax shield , profitability , growth opportunities , bankruptcy risk and liquidity appear to be the significant determinants of capital structure and no particular theory of capital structure but a mix of TOT , ACT and POT may be said to be applicable for the average manufacturing firm ( with mean level of leverage ) listed on the Bombay Stock Exchange ( BSE ). ( B ) Fixed Effects Unconditional Quantile Regression : The differential impacts of the explanatory variables on MLEV and BLEV over the unconditional quantiles of MLEV and BLEV present a holistic view of the relationships between the firm-specific determinants of capital structure and the leverage ratio and may be said to affirm the existence of non-linearity in the behaviour of such determinants in respect of Indian manufacturing companies listed on the Bombay Stock Exchange ( BSE ) . Moreover, the applicability of Trade - Off Theory , Agency Cost Theory and Pecking Order Theory at various unconditional quantiles of MLEV and BLEV suggest that capital structure decisions of Indian manufacturing corporate firms listed on the BSE are not determined by any one particular theory of capital structure but by a mix of various theories over the entire unconditional distribution of the leverage ratio .
xii
List of Tables Table No.
Particulars
Page No.
1.1
Composition of the Sources of Financing Liabilities
3
4.1
Theoretical Implications of Capital Structure Determinants
105
4.2
Formulation of Dependent Variable ( s )
108
4.3
Formulation of Independent Variables
109
4.4
Procedure of Sample Selection
111
4.5
Classification of Companies in Final Sample as per Industrial Division
112
4.6
Kernel Density Functions and Optimal Bandwidth Parameters
136
4.7
Minimum Bootstrap Replications for Some Applications
140
5.1
Descriptive Summary of Variables
143
5.2
Linear Trend Analysis of Variables
144
5.3
Pairwise Correlation Analysis of Variables
148
5.4
Panel Data Unit Root Test
149
5.5
Choice Between Pooled OLS ( POLS ), Random Effects ( RE ) and Fixed Effects ( FE ) Models
150
5.6
Tests for Heteroscedasticity and Serial Correlation in Fixed Effects ( FE ) Panel Data Model
151
5.7
Tests for Cross-sectional Dependence in Fixed Effects ( FE ) Model
151
5.8
Fixed Effects ( FE ) Regression ( Dependent Variable : MLEV )
154
5.9
Fixed Effects ( FE ) Regression ( Dependent Variable : BLEV )
154
5.10
Fixed Effects ( FE ) Regression ( Composite Results with Rogers or cluster - robust standard errors )
155
5.11
Test for Exogeneity of Explanatory Variables
156
5.12
Kernel Density Functions and Optimal Bandwidths for Estimation of MLEV and BLEV
157
5.13
Sensitivity Analysis of Unconditional Quantile Regression ( RIF - FE ) Model for Various Kernel Functions and Bandwidths [ Dependent Variable : MLEV ]
160
5.14
Sensitivity Analysis of Unconditional Quantile Regression ( RIF - FE ) Model for Various Kernel Functions and Bandwidths [ Dependent Variable : BLEV ]
167
5.15
Example Showing Impact of SIZE on MLEV
175
xiii
List of Tables Table No.
Particulars
Page No.
5.16
Analysis of Fixed Effects ( FE ) Regression Model and Unconditional Quantile Regression ( RIF - FE ) Model { based on Epanechnikov Kernel Density Function with Silverman Bandwidth } [ Dependent Variable : MLEV ]
181
5.17
Analysis of Fixed Effects ( FE ) Regression Model and Unconditional Quantile Regression ( RIF - FE ) Model { based on Epanechnikov Kernel Density Function with Silverman Bandwidth } [ Dependent Variable : BLEV ]
182
5.18
Applicability of Capital Structure Theories at Various Unconditional Quantiles of MLEV ( Dependent Variable )
187
5.19
Applicability of Capital Structure Theories at Various Unconditional Quantiles of BLEV ( Dependent Variable )
187
List of Figures Figure No.
Particulars
Page No.
1.1
Schematic Plan of Research Study
3
2.1
Net Income Approach
12
2.2
Net Operating Income Approach
12
2.3
Traditional Approach ( Durand )
13
2.4
Traditional Approach ( Solomon )
14
2.5
Static Trade - Off Theory
19
2.6
Agency Cost Theory
24
5.1
Charts showing Mean of Variables over Sample Period
145
5.2
Charts showing Histogram of Residuals from Fixed Effects Model
156
5.3
Charts Showing Kernel Density Estimation of MLEV
158
5.4
Charts Showing Kernel Density Estimation of BLEV
159
5.5
Graphs showing Estimated Coefficients of the Explanatory Variables in respect of Fixed Effects ( FE ) Regression Model and Unconditional Quantile Regression ( RIF - FE ) Model [ Dependent Variable : MLEV ]
183
xiv
List of Figures Figure No.
Particulars
Page No.
5.6
Graphs showing Magnitude of Estimated Coefficients of the Explanatory Variables in respect of Fixed Effects ( FE ) Regression Model and Unconditional Regression ( RIF - FE ) Model [ Dependent Variable : MLEV ]
184
5.7
Graphs showing Estimated Coefficients of the Explanatory Variables in respect of Fixed Effects ( FE ) Regression Model and Unconditional Quantile Regression ( RIF - FE ) Model ( Dependent Variable : BLEV )
185
5.8
Graphs showing Magnitude of Estimated Coefficients of the Explanatory Variables in respect of Fixed Effects ( FE ) Regression Model and Unconditional Regression ( RIF - FE ) Model [ Dependent Variable : BLEV ]
186
’
CHAPTER 1
1
`
Chapter 1 Introduction
1.1
Background
The study of capital structure endeavours to explain the combination of securities and sources of funds used by companies for financing real investment ( Myers , 2001 ). The capital structure of a firm1 may be said to directly determine its combined risk and cost of capital. The sources of capital having important consequences for the firm affect its value and hence shareholders’ wealth. For instance , while debt happens to be the cheapest form of external capital, the risk of default , volatility of earnings per share and return on equity of a firm increases with its borrowing. With increase in leverage , the benefits of a lower cost of debt decrease due to increasing financial risk and higher probability of financial distress and bankruptcy. A risk - return trade - off is thus involved in capital structure decisions ( Baker and Martin , 2011 ). Myers’ ( 1984 , p. 575 ) oft-quoted question, “ How do firms choose their capital structure?” and its subsequent answer “ We don’t know ” has given rise , perhaps , to the most controversial, debatable and puzzling issue in corporate finance theory. Though some pertinent answers have been provided to the above question, both theoretically and empirically, over the years, the issue has not yet been satisfactorily solved and extensive researches on this issue are being conducted till date , globally. The primary research topic for this thesis will thus be “ corporate capital structure ”. The second section of this chapter explains the concept of capital structure ; the third section discusses the concept of capital structure decision and the factors affecting capital structure decisions of a corporate firm ; the fourth section enumerates the plan of this study along with brief summaries of the ensuing chapters ; and the fifth section concludes this chapter.
The words “firm(s)” , “corporate firm(s)” and “company (or companies)” will be used interchangeably as done in corporate finance literature. 1
2
1.2
`
Concept of Capital Structure
The term “capital structure” is composed of two elements “capital” and “structure”. The concept of “capital” 2 may be explained from two points of view :( 1 ) Fund Concept. According to this concept, which recognizes the separate entity of a corporate firm and considers capital from the liability side of the balance sheet , capital may be defined as the funds provided by owners and lenders of a corporate firm in order to fulfill the objectives of its financing decision ( Banerjee, 1995 , p. 561 ). ( 2 ) Asset Concept. According to this concept , which considers capital from the asset side of the balance sheet , capital may be defined as the amount of investment made in the assets of a corporate firm in order to fulfill the objectives of its investment decision ( Ibid. , p. 561 ) . Considering the equality of the liability and the asset side of the balance sheet at any given point of time the two concepts may be integrated into the following definition of “capital” : Capital may be defined as funds provided by owners and outsiders ( sources of financing ) of a corporate firm for financing its investment activities ( employment of the sources of financing ). The term “structure” has been defined in : ( a ) Oxford English Dictionary as “ the arrangement of and relations between the parts or elements of something complex ” , and ( b ) Cambridge Advanced Learner’s Dictionary & Thesaurus as “ the way in which the parts of a system or object are arranged or organized ”. The concept of “structure” may be applied in accounting and finance to give rise to the concept of “account structure” , which may be defined as a well - defined group of elements, having some similar characteristics, which serves as a fundamental component of the accounting statements. Based on the four fundamental elements ( or accounts ) of the accounting statements of a corporate firm , namely, ( a ) capital ( or liability ) 3 , ( b ) asset , ( c ) revenue , and ( d ) cost ( or expense ) 4 , the basic account structures include the ‘asset
2
As used in accounting and finance.
3
Owned capital ( or owner’s liability ) and debt capital ( or outsiders’ liability ) .
The terms ‘cost’ and ‘expense’ are used here interchangeably in the literal sense, for e.g., operating cost may be termed as operating expense, even though in ( financial ) accounting parlance, expense refers to expired cost. 4
3
`
structure’ , ‘capital ( or liability ) structure’ , ‘revenue structure’ and ‘cost ( or expense ) structure’ ( Sinha , 2013 ) 5. Hence , capital structure may be defined as a composition ( or mix ) of the funds provided by owners and lenders of a firm for financing its investment activities. Now, the composition of the sources of financing or the liability side of an analytical balance sheet of a corporate firm may be presented as under : Table 1.1 : Composition of the Sources of Financing Liabilities ( 1 ) Owners’ Liabilities or Owned Capital :( a ) Variable Dividend - Bearing Owned Capital ( Equity Shareholders’ Fund ) ( b ) Fixed Dividend - Bearing Owned Capital ( Preference Share Capital )
( 2 ) Outsiders’ Liabilities or Debt Capital :( a ) Interest - Bearing Debt Capital :
Fixed CostBearing Capital
( i ) Long - term or Non - Current ( such as debentures, long - term bonds, long - term bank loans, etc. ) ( ii ) Short- term or Current ( such as short - term bonds, short - term bank loans, etc. )
Long - term Debt Capital
( b ) Non - Interest Bearing Debt Capital : ( i ) Long - term or Non - Current ( such as deferred tax liability, long - term provisions, etc. ) ( ii ) Short - term or Current ( such as account payables, accruals, short - term provisions, etc. )
Short - term Debt Capital
There are two views regarding the application of the concept of capital structure in corporate finance literature :( 1 ) Broader View. According to this view, all the components of the liability side of the balance sheet ( generally referred to as the ‘financial structure’ ) form the capital structure. ( 2 ) Restricted View. According to this view, owned capital and interest - bearing debt capital form the capital structure, excluding non - interest bearing liabilities which bear no explicit interest charges 6. However , the restricted view is generally followed because the terms ‘capital structure’ and ‘financial leverage’ ( or simply, ‘leverage’ ) are used synonymously, as explained below from two definitions of financial leverage as follows :
5
Present researcher.
“ unless the liability is seriously overdue and the creditor has started charging interest because of the delay in paying the liability ” Tracy ( 2016, p. 36 ) . 6
4
`
( 1 ) Financial leverage is “ the potential use fixed financial costs to magnify the effects of changes in earnings before interest and taxes on the firm’s earnings per share” ( Gitman , 2006 , p. 470 ). ( 2 ) Financial leverage is “ the extent to which fixed - income securities ( debt and preferred stock ) are used in a firm’s capital structure” ( Brigham and Houston , 2001, p. 610 ). “ Definition ( 1 ) considers both the cause [ presence of Fixed Financing Cost - Bearing Capital ( FFCBC ) in its capital structure which is said to give rise to Fixed Financing Cost ( FFC )7 in its cost structure ] and the effect [ magnification of percentage change in the initial value ( assumed to be not equal to zero ) of Earnings Per Share ( EPS ) from a given percentage change in the initial value ( assumed to be not equal to zero ) of Earnings Before Interest and Tax ( EBIT ) ] ; whereas definition ( 2 ) considers only the cause ” ( Sinha , 2013 , p. 68 ) 8. Thus, the concept of capital structure, in a restricted yet popular and logical sense, is associated with the cause of the financial leverage effect. This sub-section is concluded with the following definitions of capital structure cited in finance literature :1. Capital structure is “ the proportions of debt instruments and preferred and common stock on a company’s balance sheet” ( Van Horne , 2002 , p. 271 ). 2. “ Capital structure refers to the sources of financing employed by the firm. These sources include debt, equity, and hybrid securities9 that a firm uses to finance its assets, operations, and future growth ” ( Baker and Martin , 2011, p. 1 ).
7
Interest on debt and preference dividend.
8
Present researcher.
9
securities having the characteristics of both debt and equity shares, such as preference shares, convertible bonds etc.
5
1.3 1.3.1
`
Capital Structure Decision Concept of Capital Structure Decision
The objective of a firm is to maximise shareholders’ wealth. Capital structure decision should be focussed towards the attainment of this objective. In other words , capital structure decision should be evaluated on the basis of its impact on firm’s value. If the capital structure decision is expected to affect the total value of the firm then such a financing composition , that will maximise the shareholders’ wealth, should be selected ( Khan and Jain, 2011 ). The capital structure decision of a firm may be said to include its choice of an optimum or target capital structure, the period of average maturity of its debt and the specific sources of financing chosen at any particular time for raising new funding ( Ehrhardt and Brigham , 2009 ). The optimum capital structure refers to the most judicious combination of the sources of financing that maximises the value of the firm and hence minimises the weighted average cost of capital. The various factors affecting the capital structure decision are discussed below.
1.3.2
Factors Affecting Capital Structure Decision
The following factors [ referred to in Banerjee ( 1999 ) , Van Horne ( 2002 ) , Ehrhardt and Brigham ( 2008 ) , Chandra ( 2008 ) and Khan and Jain ( 2011 ) ] are generally considered by a firm when making capital decision :( 1 ) Stability of Sales : A firm having relative stability in sales indicating lower degree of operating leverage ( and hence lower level of business risk ), can safely afford to use more debt and incur higher fixed financing costs than a firm having unstability in sales. For instance, public utility companies having stable and predictable sales, generally employ more than debt and tend to have a higher degree of financial leverage than industrial firms. ( 2 ) Asset Structure : Firms having large proportions of tangible assets, which are suitable as collateral for loans, tend to become highly levered. For example , general - purpose assets which can be utilized by numerous businesses may be said to act as good collateral in comparison to special - purpose assets. Thus , real estate firms , mainly using general - purpose assets , are usually found to be highly levered than firms involved in technological research which employs special-purpose assets.
6
`
( 3 ) Size of Firm : Generally, small - sized firms rely to a considerable degree upon the shareholders’ funds for their financing as it is very difficult for them to obtain debt capital. Such firms are considered to be more risky than large-sized firms by prospective investors. ( 4 ) Operating Leverage and Operating Risk : Operating leverage is known as the first stage leverage and depends on the fixed operating costs of a firm. The degree of operating leverage is directly related to the percentage of fixed operating costs , ceteris paribus. Operating risk , that is , the variability of operating profit, is directly associated with operating leverage. A trade - off between operating and financial leverages should be considered by a firm so that the combined or total risk of the firm does not increase too much , as the multiplicative effect of the degrees of operating leverage and financial leverage give rise to combined leverage. A firm with a low degree of operating leverage and hence low operating risk is better able to employ greater degree of financial leverage, ceteris paribus, than a firm with high degree of operating leverage. ( 5 ) Growth Rate : Generally, firms growing at a faster rate tend to rely more heavily on external capital, ceteris paribus. Moreover, as floatation costs involved in issuing debt are much less than that in case of equity, rapidly growing firms are encouraged to employ more debt. However, these firms often face greater uncertainty, and so their willingness to use debt tend to reduce. ( 6 ) Profitability : It is often observed that highly profitable firms tend to use relatively lower amount of debt as their high rates of return enable them to finance most of their investments with internal sources of fund such as retained earnings. ( 7 ) Taxes : The interest on debt is a deductible expense while calculating taxable income and firms with higher tax rates are able to take comparatively more advantage of these deductions. Hence , the tax rate of a firm is directly related to the advantage of debt. ( 8 ) Control : Capital structure decision can be influenced by the attitude of the management towards control. The shareholders are the owners of the firm. Equity shareholders have voting rights and can exercise active control over the activities of the management. Preference shareholders do not enjoy voting rights except under special circumstances. The providers of debt capital have no direct involvement in the management of the firm. Accordingly, if the management’s prime objective is to maintain control, it is desirable to finance additional capital requirements with debt capital and preference shares. However, employment of higher levels of debt does not necessarily imply that management will enjoy complete control. This is because of the introduction of restrictive covenants ( certain restrictions on the activities of the management ) in the loan agreements by the debtholders to protect their interests. The
7
`
presence of such covenants would no doubt curtail management control to some extent. Moreover, use of excessive debt will also increase financial risk and may ultimately result in bankruptcy thus paving the way for complete loss of management control. ( 9 ) Flexibility : Flexibility refers to the ability of a firm to keep open future financing options while making present financing decision, thus providing manoeuvrability to the finance manager. It is neither desirable nor possible for a firm to issue debt continually without increasing its equity base due to the enhancement of default risk associated with employment of higher levels of debt. The present capital structure decision should be flexible enough to provide for future financing choices. For instance, if a firm undertakes an aggressive debt policy at present and in the future its prospects become bleak, then it might be forced to issue equity on unfavourable terms. It may be beneficial for the firm to issue equity now and preserve unused debt capacity ( or reserve borrowing power ) for unforeseen future needs which may arise because of the emergence of profitable investment opportunities, changes in government policies, recessionary conditions in the market, intensification of competition, disruption in supplies, etc. ( 10 ) Timing of Security Issues : The question of proper timing of issue of securities is another important factor to be considered. The firm has to decide whether to finance initially with equity issue and later with debt issue or vice versa, based on the economic and capital market conditions and expectations of the firm. ( 11 ) Industry Standards : The capital structures of peer firms belonging to the same industry and having similar business risk act as industry standards thus providing a benchmark for planning the capital structure of a firm. For instance, if a firm adopts a capital structure significantly different from industry standards, it may not be acceptable to the investors and it is imperative that the position of the firm in the capital market is justified. ( 12 ) Professional Advice : Capital structure decisions are often influenced by professional advice sought from credit rating agencies, investment bankers, financial analysts, etc. ( 13 ) Opinions of Prospective Investors and Lenders : The opinions of prospective investors and lenders regarding their preferences for the types of securities likely to be bought by them, may be sought by the firm while making capital structure decision.
8
1.4 1.4.1
`
Plan and Chapterisation of Research Study Plan of Research Study
This research study will traverse a logical and sequential path as outlined in the schematic plan below:
( 1 ) Identification of Primary Research Topic
( 2 ) Review of Literature ( Theoretical and Empirical )
( 3 ) Identification of Specific Research Problem(s) related to the Primary Research Topic
( 4 ) Formulation and Justification of Objective(s)
( 5 ) Formulation of related Variables
( 6 ) Procedure of Collection of Data and Selection of Sample
( 7 ) Discussions on the Methodological Techniques to be applied for analysis of data
( 8 ) Formulation of Statistical Hypotheses
( 9 ) Analysis of Data and Interpretations of Findings
( 10 ) Documentation of Research Report Figure 1.1 [ Schematic Plan of Research Study ]
9
1.4.2
`
Chapterisation of Research Study
This research study will comprise the following chapters , a brief summary of which are presented below : ( 1 ) Chapter 1 : Introduction In this present chapter , the primary research topic of “corporate capital structure” has been identified , and the concept of ‘capital structure’ and the factors affecting capital structure decisions of a corporate firm have been discussed. ( 2 ) Chapter 2 : Review of Theoretical Literature A review and discussion of the various theories of capital structure will be conducted in this chapter. ( 3 ) Chapter 3 : Review of Empirical Literature In this chapter , a review of the existing empirical studies on capital structure in the International and Indian contexts will be conducted in order to provide the basis for the identification of the specific research problem(s) to be addressed by this study. ( 4 ) Chapter 4 : Research Methodology The methodological aspects of the research comprising the identification of the specific research problem(s) , formulation of objectives , related variables and hypotheses , procedure of data collection and selection of sample , and discussions on the techniques to be applied for analysis of data , will be presented in this chapter. ( 5 ) Chapter 5 : Analysis and Findings In this penultimate chapter , the results of the analysis of data applying the techniques discussed in the previous chapter will be reported and the findings will be interpreted. ( 6 ) Chapter 6 : Conclusion This last chapter will be comprised of an overview of the study , the limitations of the study and suggestions for probable directions of future researches.
1.5
Conclusion
In this chapter , the primary research topic of this study has been identified as ‘ corporate capital structure ’ , the concepts of ‘capital structure’ and ‘capital structure decision’ have been explained , the factors affecting capital structure decision of a firm have been discussed and the plan and chapterisation of the study have been presented , thus paving the way for the review of the theoretical literature on corporate capital structure.
CHAPTER 2
10
Chapter 2 Review of Theoretical Literature 2.1
Introduction
In this previous chapter , the primary research topic of this study has been determined as “ corporate capital structure” , and the conceptual framework of capital structure has been discussed. This chapter discusses and reviews the various theories of capital structure in the second section with concluding remarks in the third section.
2.2
Capital Structure Theories
According to Myers ( 2003 , pp. 216-217 ) , “ There is no universal theory of capital structure, and no reason to expect one. There are useful conditional theories , however. The theories differ in their relative emphasis on the factors that could affect the choice between debt and equity...Each factor could be dominant for some firms or in some circumstances , yet unimportant elsewhere…At the end of the day some blend of all of the theories may be needed to explain capital structure.” Since the pioneering work of Modigliani and Miller ( 1958 ) , various issues relating to corporate capital structure are still being extensively studied. A detailed and careful perusal of the existing literature on the theoretical framework of corporate capital structure decisions enables one to categorise the capital structure theories under two broad heads :
( 1 ) Traditional Theoretical Approaches , which are purely definitional devoid of any economic or behavioural justification , and which include : ( a ) Net Income Approach ; ( b ) Net Operating Income Approach, and ( c ) Intermediate Approach.
( 2 ) Modern Theories , which are based on economic or behavioural justification , and the major and widely discussed among which include :( a ) Modigliani and Miller ( M-M ) Hypothesis ; ( b ) Trade- off Theory ( TOT ) [ or Tax Shield- Bankruptcy Cost Theory ( TS- BCT ) ] ; ( c ) Agency Cost Theory ( ACT ) ; ( d ) Signaling Theory ( ST ) ; ( e ) Pecking Order Theory ( POT ) ; and ( f ) Market Timing Theory ( MTT ).
11
These theoretical approaches and the theories are discussed [ based on the discussions in Banerjee ( 1999 ), Van Horne ( 2002 ), Ehrhardt and Brigham ( 2008 ) , Chandra ( 2008 ) , Khan and Jain ( 2011 ) and other specifically mentioned references ] in the following pages.
2.2.1
Traditional Theoretical Approaches
The traditional capital structure theories which are discussed below are based on the following general assumptions : ( i ) A firm finances from two sources only : risk - free debt and ( risky ) equity. ( ii ) There is no corporate tax. ( iii ) The firm’s assets and operating income remain constant over time. ( iv ) The dividend pay - out ratio is 100 % ( that is , no retentions of earnings ). ( v ) The firm can change its capital structure by selling debt to repurchase shares or by issuing shares to retire debt. ( vi ) The firm’s business risk is independent of its capital structure and remains constant. ( vii ) Expectations of investors about future operating incomes are homogeneous. ( viii ) The firm has a perpetual life.
2.2.1.1
Net Income ( NI ) Approach
According to NI Approach , propounded by Durand ( 1952 ), neither debtholders nor shareholders perceive that employment of additional debt adds to their risks ; so , irrespective of its level of debt , the firm’s cost of debt ( kd ) and cost of equity ( ke ) remain unchanged. The market value of the firm is determined by the sum of the capitalized values of equity ( capitalizing the firm’s net income to equity at constant ke ) and debt ( capitalizing interest on debt at constant kd ). Since kd is less than ke , the firm can continually reduce its weighted average cost of capital ( ko ) { and hence increase its market value } by choosing to finance its operations with cheaper debt instead of expensive equity, and may attain an optimal capital structure with 100% debt and no equity. The following diagram illustrates this approach.
Cost of Capital
12
ke ko kd
Degree of Leverage
Figure 2.1 [ Net Income Approach ]
2.2.1.2
Net Operating Income ( NOI ) Approach
The NOI Approach , also proposed by Durand ( 1952 ) , posits that additional levels of debt employed by a firm do not pose a problem for its debtholders , but shareholders do find higher leverage more risky. So , the cost of equity ( k e ) increases with increase in leverage ratio while the cost of debt ( kd ) remains constant. However , the increase in ke neutralizes the advantage of cheaper debt and renders the weighted average cost of capital ( ko ) constant irrespective of its debt-equity mix. Thus , regardless of the composition of debt and equity, the market value of the firm , determined by the capitalization of net operating income at constant ko , remains constant , thus rendering capital structure to be value - irrelevant. The following diagram illustrates this approach.
Cost of Capital
ke
ko kd
Degree of Leverage
Figure 2.2 [ Net Operating Income Approach ]
13
2.2.1.3
Intermediate ( or Traditional ) Approach
The propositions of this approach fall between the two extreme views of NI and NOI approaches. However, there are two slightly differing interpretations of this approach – one proposed by Durand ( 1952 ) and the other by Solomon ( 1963 ). Durand ( 1952 ) termed his approach as a “compromise ” between NI and NOI approaches and justified the same by explaining that restricted investors ( including banks and insurance companies ) who are deterred from buying stocks or low-grade bonds ( either by law, personal circumstance, income taxes or pure prejudice ) tend to create a “super premium” for high - grade bonds that provide an opportunity to maximize the value of a firm by effective bond financing. Initially, cost of debt ( kd ) and cost of equity ( ke ) remain constant and since kd is less than ke, the weighted average cost of capital ( ko ) declines resulting in an increase in the value of the firm ( V ). However, use of additional debt beyond a “certain level” 1 make both debtholders and stockholders perceive an increase in their risks, causing kd and ke ( and hence ko ) to increase and V to decrease . That “ certain level ” occurring ( somewhere ) at the middle ( 50 % debt and 50 % equity ) represents the optimal capital structure. Thus , a U - shape curve is assumed by ko . The following diagram illustrates this interpretation of the traditional approach.
Cost of Capital
ke
ko P kd
Degree of Leverage
Figure 2.3 [ Traditional Approach ( Durand ) ]
Solomon ( 1963 ) defines this level [ denoted by ‘P’ in Fig. ( 3 ) ] as that “ determinate point ” where the marginal cost of more debt is equal to the ( weighted ) average cost of capital. 1
14
Solomon ( 1963 ), in his interpretation of the traditional theory, divides the impact of degree of leverage on firm value ( V ) into three stages : ( i ) Stage I ( Increasing Value ) : Up to a certain level of leverage, k d remains constant, and ke either remains constant or rises slightly with debt but not fast enough to offset the advantage of cheaper debt, resulting in decrease in ko and increase in V. ( ii ) Stage II ( Optimum Value ) : Once a certain degree of leverage has been reached , the effect of additional debt on ko and hence on V will be negligible. This is because kd remains constant while ke increases due to increase in financial risk , but the advantage of cheaper debt is just neutralized by the increase in ke . There is , in fact , a range of leverage within which ko is minimum and V is maximum. ( iii ) Stage III ( Declining Value ) : Beyond the optimal range of leverage, both debtholders and equityholders perceive an increase in their risks with additional debt, causing k d and ke to increase, but the increase in ke exceeds the advantage of cheaper debt. As a result, ko increases and V decreases. The ko curve will be saucer - shaped implying an optimum range of leverage. The following diagram illustrates this interpretation of the traditional approach.
ke
Cost of Capital
ko
Stage I Stage II Stage III kd
Degree of Leverage
Figure 2.4 [ Traditional Approach ( Solomon ) ]
15
2.2.2 Modern Theories of Capital Structure The modern theoretical research on corporate debt - equity choice began with the pioneering work of Modigliani and Miller ( 1958 ) and is still going on in the absence of a conclusive and unifying theoretical framework. Several theories have been proposed in between.
2.2.2.1
Modigliani and Miller ( M-M ) Hypothesis
M-M ( 1958 ) , applying arbitrage process , presented a formal proof of the value - irrelevance conclusion of the Net Operating Income ( NOI ) Approach based on the following additional assumptions [ to that mentioned on page ( 11 ) , excluding assumption ( v ) ] : ( i ) Capital markets are perfect ( absence of information, transaction and bankruptcy costs ). ( ii ) Firms and individuals can borrow at the same rate of interest ; ( iii ) Firms can be classified into homogeneous risk classes with all firms within a particular risk class having the same business risk. They summarized their results related to value of firm, cost of capital and leverage in the following two propositions : ( a ) Proposition I : “ The market value of any firm is independent of it’s capital structure and is given by capitalizing it’s expected return at the rate…appropriate to it’s ( risk ) class ” M-M ( 1958 , p. 268 ). Alternatively, “ the ( weighted ) average cost of capital to any firm is completely independent of its capital structure and is equal to the capitalization rate of a pure equity stream of its ( risk ) class ” M-M ( 1958 , pp. 268-269 ). Thus the value of a levered firm will be equal to the value of an unlevered firm within the same risk class. ( b ) Proposition II : “ The expected yield of a share of stock is equal to the appropriate capitalization rate…for a pure equity stream in the ( risk ) class, plus a premium related to financial risk equal to the debt-to-equity ratio times the spread between ( the capitalization rate and the cost of debt ) .” M-M ( 1958 , p. 271 ) The behavioural justification for the M-M Hypothesis is the arbitrage process which refers to the act of purchasing an asset ( or security ) in a market where its price is low and selling it another market where its price is high. The M-M Hypothesis explains that the market value of otherwise homogeneous firms differing only in respect of leverage cannot be different due to operation of the arbitrage process. The investors of the higher - valued firm will sell their shares and buy the shares of the lower-valued by borrowing the required funds on their personal accounts in proportion to their shares in the debt capital
16
of the levered firm ; thus being able to substitute ‘personal leverage’ or ‘home-made leverage’ for corporate leverage. This will enable them to be better off as they would earn the same rate of return at lower investment outlay with similar or lower risk. The behaviour of the investors will consequently result in : ( a ) increasing the share prices ( and hence the total value ) of the firm whose shares are being purchased, and ( b ) lowering the share prices ( and hence the total value ) of the firm whose shares are being sold. This will continue until the market value of the two firms become identical. The M-M Hypothesis has been criticized, supported and extended since its inception. Durand ( 1959 ) and Barges ( 1963 ) criticized the hypothesis due to its unrealistic assumptions. Solomon ( 1963 ) cast doubt on the invariable constancy of the cost of debt , arguing that when the accepted level of leverage is exceeded , the consequent increase in the probability of default in interest payments leads to an increase in the cost of debt. Stiglitz ( 1969 ) using a state - preference framework , and Rubinstein ( 1973 ) using a mean variance approach in a Capital Asset Pricing Model ( CAPM ) - framework , proved the M-M ( 1958 ) propositions considering risky debt in the absence of bankruptcy costs. The relaxation of some of the assumptions of M-M hypothesis by considering factors of market imperfections such as taxation, bankruptcy costs, agency costs, transaction
costs
and
asymmetric
information, has given rise to the following five major and widely discussed theories of capital structure: trade - off theory, agency cost theory, pecking order theory, signaling theory and market timing theory.
2.2.2.2
Trade - Off Theory ( TOT )
The traditional Trade - Off Theory or the Tax Shield - Bankruptcy Cost Theory ( TS-BCT ) postulates that the value of a firm may be maximized at an optimal level of capital structure where the marginal benefits of debt ( that is , the present value of tax savings from tax - deductibility of interest payments ) and the marginal costs of debt ( that is , the present value of potential bankruptcy
2
costs , holding its assets and investment plans
constant ) are equalized ( Chen and Kim , 1979 ).
2
Bankruptcy occurs when the fixed obligations to creditors cannot be met and there is a transfer of ownership and a formal reorganization of the capital structure of the firm. The costs related to this transfer may be classified as direct ( for instance , legal , accounting and trustee fees as well as the possible refusal of income-tax loss carry-overs and carry-backs) or indirect ( for instance, opportunity costs arising from disruptions in relationships between suppliers or customers of the firm and related to the transfer of ownership or control ( Huagen and Senbet ,1978, p. 384-385). However, direct bankruptcy costs having economies , can be substantial for small firms but less important for large firms. Indirect bankruptcy costs can be significant for both large and small firms ( Warner, 1977 ).
17
There are two versions of this theory : ( a ) static, and ( b ) dynamic. The static version implicitly assumes the existence of target leverage and postulates that firms are already at their target levels of debt. The dynamic version explicitly assumes the existence of target leverage and accounts for the adjustment of observed leverage towards the target level.
2.2.2.2.1
Static Trade - Off Theory ( STOT )
This theory posits that the optimal leverage for a firm is determined as a function of the distribution of its future earnings , business risk , bankruptcy costs and taxes. So, a shift in the probability distribution of earnings, implying an increase in the probability of bankruptcy ( relative to the level of leverage before the shift in the earnings probability distribution ), will simultaneously raise the expected marginal bankruptcy costs and lower the expected marginal tax savings. As a result , leverage will become less attractive on the margin and optimal leverage will be reduced until marginal expected tax savings and marginal expected bankruptcy costs equalize ( Castanias , 1983 ). Durand ( 1952 ) examined NOI and NI approaches under corporate taxes and concluded that the value of a firm increases with leverage under both approaches but the increase is much greater in case of NI approach. M-M ( 1963 ) , incorporating corporate tax in the analysis, showed that the market value of a firm will increase with increase in leverage due to the tax - deductibility of interest charges , and the value of the levered firm will be equal to the sum of the value of the unlevered firm and the present value of interest tax shield. In effect , M-M ( 1963 ) revised their propositions ( 1958 ) from the improbable conclusion of the value irrelevance NOI Approach to the more improbable value - relevance conclusion of the NI Approach. Robicheck and Myers ( 1965 ) , Baxter ( 1970 ) , Hirshleifer ( 1970 ) , Kraus and Litzenberger ( 1973 ) and Scott ( 1976 ) opine that the concept of an optimal capital structure may be supported through the incorporation of bankruptcy costs within the foundations laid by M-M, but their results have not provided directly usable formulae for the determination of optimal capital structure ( Kim , 1978 ; Leland , 1994 ). Kraus and Litzenberger ( 1973 ), based on a state - preference model with wealth taxes and bankruptcy costs, propose a stochastic dynamic programming approach to search for an optimal capital structure. It is shown by Scott ( 1976 ) that the existence of an optimal capital structure is implied by imperfect markets for physical assets along with a constant liquidation value of the firm’s assets in bankruptcy , if investors are indifferent to risk ( Kim , op. cit. ). Miller ( 1977 ),
18
introducing personal taxes 3 in the analysis , derived an expression for the gain from leverage when different corporate and personal tax rates are applied , arguing that the tax disadvantage at the personal level may fully or partially neutralise the tax advantage of debt over equity , and that an optimum debt - equity ratio may not exist for any individual firm but may exist for the corporate sector as a whole. It was further suggested by him that the leverage clientele hypothesis whereby “companies following a no-leverage or low leverage strategy…would find a market among investors in the high tax brackets ; those opting for a high leverage strategy…would find the natural clientele for their securities at the other end of the scale” Miller ( 1977, p. 269 ) , and based on that hypothesis opined that “since one clientele is as good as the other…it would still be true that the value of any firm, in equi-librium, would be independent of its capital structure, despite the deductibility of interest payments in computing corporate income taxes” ( Ibid., p. 269 ). Brennan and Schwartz ( 1978 ) , based on the option pricing framework, model the value of the levered firm as a function of unlevered firm, the amount of outstanding debt and its time to maturity. Numerical techniques are utilized by them for the determination of optimal leverage, assuming the value of the unlevered firm follow a diffusion process with constant volatility ( Leland , op. cit.). Kim ( 1978 ) , based on a mean - variance model incorporating taxes and bankruptcy costs , presents numerical examples of optimal capital structure showing that the optimal capital structure occurs before the debt capacity ( the maximum amount a firm can borrow in the capital market ) such that a firm , whose objective is to maximize shareholders’ wealth, will always search for the optimal capital structure rather than simply maximize its borrowing. This model is , however , less parsimonious without the knowledge of the joint distribution of market and firm returns ( Chen and Kim , op. cit. ; Leland, op. cit.) . DeAngelo and Masulis ( 1980 ) propose a two-date state-preference model of optimal capital structure wherein the impact of non - debt tax shields 4 are shown to substitute and diminish the benefits from interest tax shields. It is postulated that at higher levels of leverage, the marginal savings from an additional unit of debt declines with increase in the quantum of non - debt tax shields due to the increased probability that the potential interest tax shields ( or debt tax shields ) will be fully or partially lost through bankruptcy. Thus , leverage ratio and non-debt tax shields should be inversely related.
3 4
Taxes on incomes from stocks ( dividends and capital gains ) and income from debt ( interest income ).
such as accelerated depreciation allowances on investments, tax loss carry - forwards or backwards, preliminary and preoperative expenditure , research and development expenditures, etc.
19
The following diagram illustrates the static trade-off theory :
[ Source : Myers,S.C. ( 1984 ). The Journal of Finance, p. 577 ]
Figure 2.5 [ Static Trade - Off Theory ]
2.2.2.2.2
Dynamic Trade- off Theory ( DTOT )
Dynamic trade-off theory propounds that firms have a value-maximising optimal or target leverage deviations from which , being costly, will be gradually removed over time. Brennan and Schwartz ( 1984 ) develop a contingent claims dynamic equilibrium valuation model wherein a firm is assumed to choose its investment policy and capital structure policy from a set of feasible policies determined by investment opportunities, capital market equilibrium and provisions of bond indentures. The analysis highlights three main facets of optimal financing decision , namely, ( a ) design of an optimal bond indenture, ( b ) choice of an optimal initial capital structure, and ( c ) choice of an optimal capital structure given the existing capital structure. Kane et al. ( 1984 ) develop a contingent claims model in an options valuation framework for a levered firm incorporating bankruptcy costs along with corporate tax and differential personal taxes on capital gains and ordinary income , with a view to deduce the consistency of the magnitude of the tax advantage to debt over a range of observed leverage ratios. It is concluded that simultaneous existence of levered and unlevered firms cannot be solely explained by differential bankruptcy costs across firms. Moreover , the results of simulation analysis indicate that in case of the quantum of tax advantage to debt being low , the cost of substantial deviation from the optimal leverage ratio is also low. Fischer et al. ( 1989 ) develop a dynamic model of capital structure choice of a hypothetical firm , deriving closed-form solutions for the value of the firm’s debt and
20
equity as a function of its dynamic recapitalization decisions. The resulting optimal dynamic capital structure policy is shown to depend upon the benefit of debt financing, the probable costs of debt financing , the riskless interest rate , variability of underlying asset , and the quantum of recapitalization costs. Hennessy and Whited ( 2005 ) , incorporating flotation costs for equity issue, analyze a model showing the observance of a negative correlation between profitability and debt under some possible values of parameters ( Miglo, 2013). Ju et al.( 2005) develop a calibrated contingent - claims model wherein bankruptcy can be forced by long - term in the event of the value of a firm being very low , providing estimates of optimal capital structures and showing that based on the model’s predictions firms are not underlevered ( Ibid.). Strebulaev ( 2007 ) develops and analyzes a model wherein financially distressed firms are required to sell their assets at a discount showing that the level of debt is relatively lower than the predictions of the static models ( Ibid.). Titman and Tsyplakov ( 2007 ) develop a model wherein the value of equity or the value of debt is maximized depending on the costless writing of contracts , thus offering an explanation for the slow adjustment towards the target debt level ( Ibid.). Tserlukevich ( 2008 ) presents a model in which investments are assumed to be irreversible and fixed investment cost depends on the existing level of capital , thus providing a replication of the inverse relationship between profitability and leverage ( Ibid.). De Angelo et al. ( 2011 ) develop a dynamic model of capi-tal structure wherein firms employ transitory debt5 to fund investment opportunities . The model is shown to : ( i ) replicate industry leverage very well , ( ii ) explain issue or repayment of debt better than existing trade-off models , ( iii ) account for changes in leverage accompanying investment ‘spikes’ , and ( iv ) generate leverage ratios with slow average speeds of adjustment to target , with rebalancing towards target largely occurring when the investment needs of the firms are moderate. The developments in dynamic capital structure theory have resulted in empirical research being increasingly focusing on firm’s speed of adjustment to target capital structure.
5
The difference between a firm’s actual and target levels of debt.
21
2.2.2.3
Agency Cost Theory ( ACT )
Jensen and Meckling ( J - M, hereafter ) [ 1976 ] propound the agency6 cost aspect to explain the existence of optimal capital structure, arguing that the probability distribution of cash flow provided by the firm is not independent of its ownership structure. Assuming that firm size and outside financing are constant, they posit that the value of a firm is a function of the agency costs incurred. Agency costs includes the costs for both debt and equity. The costs related to equity include : ( i ) the monitoring expenditures by the equity shareholders ( principal ) to control the activities of the manager ( agent ) through budget restrictions, compensation policies, operating rules, etc. ; ( ii ) the bonding expenditures by the manager to guarantee that certain actions, which would harm the shareholder, will not be taken or to ensure that the shareholder will be compensated if such actions are taken ; and ( iii ) the residual loss , that is the monetary equivalent of the reduction in welfare of the shareholders due to the divergence of the manager’s decisions from those decisions which would maximize the welfare of the shareholder ( J-M, 1976 ). The agency costs of debt include ( i ) the opportunity wealth loss ( or opportunity cost ) caused by the impact of debt on the investment decisions of the firm ; ( ii ) the monitoring and bonding expenditures by both the bondholders and the owner- manager ( that is , the firm ) ; and ( iii ) the costs associated with bankruptcy and reorganization ( Ibid.) . The determination of optimal debtequity ratio involves a trade-off between these two types of agency cost. J - M ( 1976 ) identify two types of conflicts because of the incentive problem associated with the issuance of new debt and new external equity : ( i ) conflict between managers and shareholders, and ( ii ) conflict between shareholders and debt holders.
( A ) Conflicts between equity shareholders and managers Conflicts between shareholders and managers arise because less than the entire residual claim on a firm are held by managers, and as a result the entire gains from their profit enhancing activities are not captured by them but the entire costs of such activities are borne by them ( Harris and Raviv, 1991 ). J-M ( 1976 ) analyze the effect of outside equity on agency costs by comparing the behaviour of a manager when the entire residual claims
Jensen and Meckling ( 1976 , p. 308 ) define an agency relationship “ as a contract under which one or more persons ( the principal (s)) engage another person ( the agent ) to perform some service on their behalf which involves delegating some decision making authority to the agent. If both parties to the relationship are utility maximizers there is good reason to believe that the agent will not always act in the best interests of the principal.” 6
22
on a firm is owned by him to his behaviour when a portion of those claims is sold by him to outsiders. In the former case, he will make operating decisions [ involving the benefits derived by him from monetary returns and the utility generated by various nonmonetary aspects ( corporate perks, fringe benefits, etc. ) of his entrepreneurial activities ] which maximize his utility without generation of any agency costs of equity. In the latter case, agency costs will be generated by the divergence between his interest and those of the outside shareholders, since he will then bear only a fraction of the costs of any nonmonetary benefits he takes out in maximizing his own utility. As the owner-manager’s fraction of the equity falls, his fractional claim on the outcomes falls and this will tend to encourage him to misappropriate larger amounts of the corporate resources in the form of perquisites or to devote insignificant effort in searching out new profitable ventures. Consequently, the minority shareholders will expend more resources in monitoring the behaviour of the owner - manager. Prospective minority shareholders will realise about the divergence of the owner-manager’s interests from theirs, and so the price to be paid by them for shares will reflect the monitoring costs and the effect of the divergence between the manager’s interest and theirs. As the owner-manager’s proportional ownership falls , the opportunity costs for obtaining additional cash in the equity markets rise ( J-M , op. cit.). Holding constant the manager’s absolute investment in the firm, increases in the fraction of the firm financed by debt increase the manager’s share of the equity, lower the agency cost of equity and mitigate the conflicts arising between shareholders and managers ( Harris and Raviv , op. cit.). Easterbrook ( 1984 ) and Jensen ( 1986 ) argue that debt can act as an effective substitute for dividends and help in disciplining managerial behaviour for firms with substantial free cash flow
7
as managers of these firms are able to expend current
cash by increasing dividend payments or by repurchasing stock without investing it in suboptimal projects or wasting it in other organizational inefficiencies. Thus managers , having control over the use of future free cash flows , can promise to announce a permanent increase in dividend payments. However, such promises are weak because of the possibility of the reduction in dividend payments in the future. The fact that capital markets punish reductions in dividend payments with large cuts in share price is consistent with the agency costs of free cash flow. The managerial promises of paying out future cash flows in a manner which cannot be accomplished by increase in dividend payments are bonded through
7
Free cash flow is cash flow in excess of that required to fund all projects that have positive net present values when discounted at the relevant cost of capital ( Jensen, 1986 ).
23
issue of debt in exchange for equity share , leading to the debt-holders being given the right to take the firm into bankruptcy court if the promise to make the interest and principal payments is not fulfilled ( Jensen , 1986 ).
( B ) Conflicts between debtholders and equity shareholders The conflicts between debtholders and equity shareholders arise because of : ( 1 ) Over-investment ( or asset substitution or risk shifting ) problem. Equity shareholders are allowed , by the debt contract , an incentive to invest in value-decreasing projects ( Harris and Raviv , op. cit. ) ; the cost of incentive for investing in sub-optimal projects financed by debt being borne by the equity shareholders issuing the debt ( Chen and Kim , op. cit.; Harris and Raviv , op. cit.). ( 2 ) Under-investment ( or debt overhang ) problem. Myers ( 1977 ) argues that , when firms are in financial distress , the existence of risky debt (maturing after an investment decision) may induce equity shareholders to reject an investment even though it has a positive Net Present Value ( NPV ). This is because the entire cost of the investment are borne by the equity shareholders whereas the debtholders mainly enjoy the returns from the investment. So, employment of larger levels of debt may result in rejection of more positive NPV projects ( Chen and Kim , op. cit.). The agency cost of debt could be reduced if managers choose safer projects with higher probability of success. Diamond ( 1989 ) and Hirshleifer and Thakor ( 1992 ) show how reputational considerations induce firms and managers to invest in relatively safe projects ( Harris and Raviv , op. cit.). According to Diamond ( 1989 ), older and more established firms find it optimal to choose safe projects without engaging in asset substitution, and adopt longer debt repayments with lower borrowing costs that eventually lead to better and valuable reputation ; on the other hand, younger firms with little reputation tend to invest in risky projects and eventually switch to a safe project if they survive without any default ( Ibid. ). Hirshleifer and Thakor ( 1992 ) show that the investment policies of firms are distorted in favour of comparatively safe projects because of managerial incentives for building their reputations, whereby their interests are aligned with those of debtholders even though their terms of employment depend on the shareholders. This effect contradicts the popular agency problem of risky debt which is imperfectly protected by covenants which tempt the equity shareholders in favouring excessively risky projects for the purpose of expropriating wealth from debt-holders. Thus , shareholders can actually be made better off ex ante when the firm is allowed to issue more debt because of conservatism resulting from managerial concern for reputation.
24
The agency cost theories imply that the process of choosing the level of corporate leverage is rather complex with the objective of reducing the capacity of the manager to act in a fashion contrary to the interest of the shareholders and of reducing the shareholders’ capacity to act in a fashion contrary to bondholders’ welfare ( Sheutrim et al. , 1993 ). The following figure illustrates the agency cost theory . The total agency costs, AT ( E ) is represented as a function of the ratio of outside equity to total outside financing , E = So / ( B + So ) , for a given firm size, V*, and given total amounts of outside financing ( B + So ) , where So and B respectively refer to outside equity and debt. A𝑆𝑜 ( E ) , AB ( E ) and AT ( E* ) are respectively the agency costs associated with outside equity , agency costs associated with debt , and the minimum total agency costs at optimal fraction of outside financing E* ( J-M , 1976 , p. 344 ).
[ Source : Jensen and Meckling ( 1976 ). Journal of Financial Economics, 344 ]
Figure 2.6 [ Agency Cost Theory ]
25
2.2.2.4
Signaling Theory ( ST )
Ross ( 1977 ) introduced signaling theory based on the concept of information asymmetry. He opines that “ implicit in the irrelevancy proposition is the assumption that the market knows the ( random ) return stream of the firm and values this stream to set the value of the firm. What is valued in the marketplace, however, is the perceived stream of returns for the firm. Putting the issue this way raises the possibility that changes in the financial structure can alter the market’s perception. In the old terminology of Modigliani and Miller, by changing its financial structure the firm alters its perceived risk class, even though the actual risk class remains unchanged ” ( Ross , 1977, p. 25 ). He suggests that real financial variables such as financial leverage or dividend policy may be chosen by managers ( who, as insiders, have monopolistic access to information about the firm’s expected future cash flows) for the purpose of sending clear signals to the public about the future performance of the firm if induced by proper incentive. These signals cannot be mimicked by unsuccessful firms who have insufficient cash flow to back them up ; moreover managers have incentives to tell the truth. There would be no signaling equilibrium without management incentives to signal truthfully. A firm that increases the level of debt or the quantum of dividend payout is signaling that its expected future cash flows are likely to be sufficiently large for meeting debt obligations or dividend payments without enhancing the likelihood of bankruptcy. The following empirical implications may be derived from this model : ( a ) the cost of capital is independent of the financing decision of the firm, despite each firm having its own unique level of debt ; and ( b ) the value of a firm and its profitability are positively related to its debt-equity ratio ( Copeland et al., 2005 ). Leland and Pyle ( 1977 ) present a model focussing on owners instead of managers. They assume that entrepreneurs have better information about the expected value of their venture projects than do outsiders. The inside information held by an entrepreneur can be transferred to suppliers of capital because it is in the owner’s interest to invest a greater fraction of his or her wealth in successful projects. Thus the owner’s willingness to invest in his or her own projects can serve as a signal of project quality, and the value of the firm increases with the percentage of the equity held by the entrepreneur relative to what otherwise would have been held given a lower-quality project. The empirical implications of this signalling argument are that : ( a ) if the original founders of a company going public decide to keep a large fraction of the stock, then these firms should experience greater price earnings multiples ; and ( b ) if the firm’s value is positively related to the fraction
26
of the owner’s wealth held as equity in the firm, then the firm will have greater debt capacity and will use greater amounts of debt ; and ( c ) debt is not a signal in this model but its use will be positively correlated with the firm’s value ( Ibid.). Heinkel ( 1982 ) develops a costless signaling equilibrium model involving capital structure relevance in a perfectly competitive but asymmetrically informed capital market . Assuming a positive correlation between firm value and credit risk across firms , the model refutes the capital structure irrelevance theorem of Modigliani-Miller , and it is shown that in equilibrium , larger amounts of debt financing are employed by riskier and more valuable firms. Lee et al.(1983) develop an equilibrium model in which a firm is said to resolve the informational asymmetries about the qualities of products offered by it for sale through the application of a mechanism combining signaling and costly screening arguments. The model considers the capital market scenario in which costly information about a firm’s a priori imperfectly known earnings distribution are produced by bondholders who are supposed to utilize this information for the specification of a schedule for bond ; and based on this schedule , the relevant parameters of the earnings distribution of the firm are subsequently signaled to potential shareholders through it’s optimal choices of debt-equity ratio and debt maturity structure. Brennan and Kraus ( 1987 ) develop a costless signaling equilibrium model for deriving the conditions under which the adverse-selection problem , arising out of information asymmetry and preventing a firm from issuing
securities for financing profitable investment
opportunites, may be overcome by an appropriate choice of financing instruments that helps in revealing private information regarding its future prospects to outside investors and may depend upon its prior capital structure. The requirement that the economic value of the net claim issued by the firm takes on its minimum value when issued by that firm imposes conditions on the function relating the payoff on the net claim to the firm’s future earnings, for instance , the payoff function must be V-shaped if the information asymmetry lies about the mean of the earnings distribution. The conditions on the net payoff function required for a revealing equilibrium may not be sufficient for yielding a unique financing strategy ; moreover , no strategy satisfying the conditions may exist depending on the firm’s preexisting capital structure and the nature of the information asymmetry. Issues of equity shares along with debt retirements or issues of junior convertible bonds are also considered as examples of financings for the resolution of distinct types of information asymmetry.
27
Noe ( 1988 ) models the financing decisions of a firm as a sequential signaling game showing that : ( a ) when the future cash flows of the firm are observed perfectly by corporate insiders , dominance of debt over equity financing results through the application of refinements to the set of admissible equilibria ; and ( b ) when the insiders have imperfect information about the firm’s future cash flows , sequential equilibria may exist , in which this pecking order breaks down and equity financing is preferred to debt financing by some firms ; however , the announcement effect of equity financing will be negative relative to debt financing , despite this breakdown of the pecking order. Brick et al. ( 1989 ), differentiating higher ( lower ) valued firm on the basis of lower ( higher) variance of the distribution of cash flow, develop a separating and sequential signaling equilibrium model in which information about the variance of cash flow is conveyed to the market through the levels of debt and dividends used by the firms and costs of signaling are minimised. It is shown that if optimal level of debt and risk are positively related then issue of new equity ( by decreasing leverage ) and simultaneous payment of cash dividends enable the higher valued firm to signal its quality. Constantinides and Grundy ( 1989 ) develop a fully revealing or separating signaling equilibrium model wherein : ( a ) if investment is assumed to be fixed , all types of firms under-take positive net present value investment and repurchase equity shares financed by the issue of a new security in the form of convertible debt whose covenants act as signals ; and ( b ) if investment is assumed to be endogenous , different types of firm undertake different optimal levels of investment and repurchase equity shares financed by straight debt with the par value of straight debt and the quantum of investment jointly acting as signals. Ravid and Sarig ( 1991 ) generalize dividend signaling and debt signaling models in an intuitive fashion, suggesting that the two financial policies be considered together as parts of a commitment package. They propose that managements of firms take the “total-cashflows” view, that is , they regard both dividends and interest as part of a financial commitment package. According to their propositions, better firms are expected to be more highly leveraged and to pay higher dividends than lower quality firms. As better firms issue more debt and pay higher dividends, announcements of increases in either leverage or dividend payout should result in an increase in the market’s valuation of the firm. They further show that firms that face higher effective corporate tax rates, ceteris paribus, will use more debt financing and pay higher dividends.
28
2.2.2.5
A Synthesis of Trade-Off Theory, Agency Cost Theory and /or Signaling Theory
Bradley et al. ( 1984 ) , incorporating positive personal taxes on equity and on bond income, expected costs of financial distress 8 , and positive non-debt tax shields , develop a singleperiod model which synthesizes the modern balancing theory of optimal capital structure . It is shown that optimal leverage is negatively related to expected costs of financial distress and to the amount of non-debt tax shields. Moreover , a simulation analysis demonstrates that optimal leverage and variability of earnings are inversely related if costs of financial distress are significant. Leland and Toft ( 1996 ) present a model of an optimal capital structure of a firm that can choose both the amount and maturity of its debt , whereby bankruptcy is determined endogenously and not by the imposition of a positive net worth condition or by a cash flow constraint. Optimal leverage ratios , credit spreads, default rates, and write-downs, which are consistent with historical averages are predicted by the model. Short term debt cannot take advantage of tax benefits as completely as long term debt ; but it is more prone to provide incentive compatibility between debt-holders and equity-shareholders and also to reduce or eliminate the agency costs associated with the “asset substitution” problem. The optimal maturity of the capital structure is determined by balancing the tax advantage of debt against bankruptcy and agency costs. John ( 1987 ) , with a view to examine optimal corporate financing arrangements under asymmetric information in respect of different patterns of temporal resolution of uncertainty in the underlying technology , characterize and compare the associated informational equilibria and the optimal financing arrangements arising from the special cases of an agency problem, a signaling problem and an agency-signaling problem. It is shown that in the agencysignaling equilibrium , capital structure choices , deviating optimally from agency-cost minimizing financing
arrangements and thus inducing risk-shifting incentives in the
investment policy , signal the private information of corporate insiders at the time of financing.
The term “costs of financial distress” is used in a general context to represent both bankruptcy costs and agency costs of debt which become economically significant only when the firm is in financial distress.In the agency costs framework, costs associated with financial distress include the costs of renegotiating the debt contracts of a firm and the opportunity costs of non-optimal production or investment decisions arsing in the event of the firm being in financial distress. In the bankruptcy cost framework, these costs include the direct and indirect costs of bankruptcy. ( Bradley et al. 1984, pp. 859-860 ). 8
29
2.2.2.6
Pecking Order Theory ( POT )
( A ) Overview The pecking order theory or the theory of hierarchical financing posits that capital structure decisions of firms will be driven by their desire to finance new investments , first internally with retained earnings , then with low-risk debt , and finally with equity only as a last resort ( Harris and Raviv , 1991 ). The hierarchy of financing choices have been explained from two point of views - one proposed by Donaldson ( 1961 ) , and the other by Myers ( 1984 ) and Myers and Majluf ( 1984 ). Donaldson ( 1961 ) observes that managers prefer the funding of new investments with retained earnings rather than debt, but also prefer debt to equity financing if retained earnings are not adequate. Accordingly, firms : ( a ) passively accumulate retained earnings and thus become less leveraged when they are profitable, and ( b ) accumulate debt and thus become more leveraged when they are unprofitable. Donaldson ( 1961 ) propounds an explanation based on costs : ( a ) firms adopt a hierarchical financing order to avoid expenses incurred during an issuance of equity shares ; and ( b ) in the case of financing deficit, firms prefer debt to equity shares as the issuance costs of debt are generally lower than those of equity shares. Myers ( 1984 ) and Myers and Majluf ( 1984 ), challenging this view, posit pecking order proposals based on the objective of the firms to maximize shareholders’ wealth. The POT originates from the concept of information asymmetry9 which leads to the problem of adverse selection originally identified by Akerlof ( 1970 )10 . The problem of adverse selection is similarly present in the capital markets manifested in the relationship between individual firms and their potential investors. Information asymmetry occurs when potential investors, having less information the quality of a firm and its investments than the managers of the firm who have full information, may face difficulties in segregating good quality firm from
9
A situation in which one party in a transaction has relevant and superior information compared to another.
In the well - known ‘ Market for Lemons’ article , Akerlof ( 1970 ) discusses the problem in relation to the used car market and the asymmetric information between buyers and sellers about the quality of a given car.The idea behind the example is that there are good quality cars and bad quality cars. If the price of the good quality cars is $15.000 and the price of bad quality cars is $5.000, then the problem occurs as the buyers is unable to distinguish between the quality of cars, and hence is not willing to pay the $15.000 for a car. If there are equal amounts of bad and good cars, the buyer will pay $10.000 for a car, however, the seller of a good car is unwilling to sell at this price in which case the buyer only attracts the bad quality cars. 10
30
bad quality firm. As a compensation for this uncertainty in the ascertainment of quality, the investors would require a higher rate of return, thereby making the funding more expensive for firms. So, if possible , ‘good quality’ firms will choose different funding alternatives. Myers and Majluf ( 1984 ) showed that , if due to information asymmetries , new investors are less well-informed about the value of a firm’s assets than present firm insiders , then equity may be mispriced by the market. If the firm is required to finance new projects by issuing equity, severe underpricing of equity may allow the new investors to capture more than the NPV of the new project resulting in a net loss to existing shareholders. The project , even being NPV-positive , will be rejected , resulting in an under-investment problem which may be avoided if the new project can be financed through internal funds and/or riskless ( or not too risky ) debt which involve no undervaluation , and hence will be preferred to equity ( Harris and Raviv , 1991 ). Myers ( 1984 ) has outlined the hierarchies of business financing as follows : “ ( 1 ) Firms prefer internal finance. ( 2 ) They adapt their target dividend payout ratios to their investment opportunities, although dividends are sticky and target payout ratios are only gradually adjusted to shifts in the extent of valuable investment opportunities. ( 3 ) Sticky dividend policies plus unpredictable fluctuations in profitability and investment opportunities, mean that internally generated cash flow may be more or less than investment outlays. If it is less, the firm first draws down its cash balance or marketable securities portfolio. ( 4 ) If external finance is required, firms issue the safest security first. That is, they start with debt, then possibly hybrid securities such as convertible bonds, then perhaps equity as a last resort. In this story, there is no well - defined target debt-equity mix, because there are two kinds of equity, internal and external, one at the top of the pecking order and one at the bottom. Each firm’s observed debt- equity ratio reflects its cumulative requirements for external finance.” Myers ( 1984 , p. 581). So firms prioritize their sources of financing preferring internal financing over external financing ; and if, after utilizing internal funds, external financing is at all required then debt financing is preferred over equity financing. In general, it will be the cheapest for a firm to use from the least to the most expensive source of finance in the following hierarchical order : internal financing, bank debt, bond market debt, convertible bonds, preference capital, and common equity.
31
( B ) Implications of Pecking Order Theory ( POT ) Ross et al. ( 2003, p. 440 ) discuss the following implications associated with the peckingorder theory that contradicts the trade-off theory. 1. There is no target amount of leverage. According to the trade-off model , each firm balances the benefits of debt, such as interest tax shield, with the costs of debt, such as costs of financial distress. The optimum quantum of leverage occurs at that level where the marginal benefits of debt and the marginal costs of debt become equal. The peckingorder theory , by contrast , does not imply a target amount of leverage. The leverage ratio is rather chosen by each firm based on its financing needs. Projects are first funded out of retained earnings , thus lowering the percentage of debt in the capital structure, because both the book value and the market value of equity are raised by profitable, internally funded projects. Additional cash needs are met with debt, raising the debt level. However , exhaustion of the firm’s debt capacity at some point results in issue of equity shares. Thus, probable availability of projects may be said to determine the amount of leverage. 2. Profitable firms use less debt. Profitable firms depend less on external sources of finance due to their ability of generating internal cash. Since firms desiring the need for external capital take resort to debt first , profitable firms end up relying on less debt. The trade-off theory does not predict this implication. More profitable firms with greater cash flow are able to generate greater debt capacity to capture the debt tax shield and the other benefits of leverage. 3. Companies like financial slack. The pecking-order theory is based on the constraints of obtaining funds for investments at a reasonable cost. If the managers of a firm try to issue more equity shares , a skeptical potential investor thinks that the share is overvalued , thereby leading to decline in share-price. Since this happens with debt only to a lesser extent , managers tend to rely first on financing through debt. However , firms can only issue that much debt that would be able to counter the expected costs of financial distress. It will , hence , be easier to have the cash ahead of the time when the actual need for funding investments will arise. This is the idea behind financial slack. Firms will thus accumulate cash today as they know that profitable projects have to be funded at various times in the future and they will not be forced to go to the capital markets. However, there is a limit to the amount of cash a firm will want to accumulate , so that managers are not tempted to pursue wasteful activities with excessive free cash.
32
( C ) Extensions to the Pecking Order Theory ( POT ) Myers ( 1984 ) introduces a modified pecking order theory, which posits that firms face the limitation of “debt capacity” or maximum attainable leverage and frictions associated with raising of capital and thus cannot strictly adhere to the pecking order predictions. Lemmon and Zender ( 2004 ) suggest a modified version of the pecking order theory, in which each firm has a debt capacity. They classify firms as under - leveraged or over - leveraged relative to its estimated debt capacity, and argue that over - leveraged firms have no choice but to reduce leverage because they cannot borrow further in the market ; whereas under-leveraged firms would “stockpile” debt capacity if the financing deficit is negative, by using their excess cash to retire outstanding debt ( Flannery and Rangan , 2006 ). Halov ( 2006 ) proposes a model considering a firm without internal funds where the choice of security depends on the current adverse selection cost of the security, the future information environment and the future needs of financing of the firm. Current debt issues make future security issues more sensitive to the degree of asymmetric information in the issuance period. It is observed that future adverse selection costs affect negatively the debt component of new external financing and positively the cash reserves of the firm, thus explaining the relative preference of companies to equity over debt, and providing an idea about why the incentive for issuing equity depends not only on the extent of asymmetric information in current period but also in future periods ( Miglo , 2013 ). Halov and Heider ( 2011 ) present a model showing that more equity and less debt should be employed by a firm if risk plays a major role in the adverse selection problem of external financing. This helps to explain why large mature firms issue debt and young small firms issue equity. An outside investor presumably knows less about the risk of an investment if he faces a young small nondividend paying firm than if he faces a large mature dividend paying firm ( Ibid. ). Miglo (2012) , considering firms with a two-stage investment project in which asymmetric information exists regarding both the firms’ quality and their growth potential, show that if the extent of asymmetric information regarding quality is high compared to that about growth, then an equilibrium where high-quality firms issue equity does not exist that is consistent with POT. If the extent of asymmetric information regarding quality is small enough while that regarding growth is high enough, the firms’ behavior will differ from what is predicted by POT. These results may be said to explain why firms in growing industries , characterized by the high degree of uncertainty about the rates of growth , do not follow the predictions of POT ( Ibid.) .
33
2.2.2.7
Market Timing Theory ( MTT )
According to this theory, firms are more likely to issue equity when the market value of the firm’s equity is high and to repurchase equity when the market value is low. The current capital structure of a firm is the cumulative effect of past attempts to time the equity market ( Baker and Wurgler , 2002 ). Market timing theory predicts that an equity offering will be preceded by a period of positive abnormal returns and that the stock price will drop after the announcement of the offering ( Lucas and McDonald , 1990 ). This theory posits that there is no optimal capital structure because the debt to equity ratio will change whenever there is a market timing behavior. The first appearance of such explanations is detected by analyzing the historical studies of stock returns, namely, Taggart ( 1977 ), Marsh ( 1982 ), Asquith and Mullins ( 1986 ). More recent studies
11
use the Market-to-Book ratio
to detect possibilities of timing. Hovakimian ( 2006 ) and Kayhan and Titman ( 2007 ) point out that the main drawback of Baker and Wurgler’s ( 2002 ) model is the proxy used to capture market timing. The historical Market-to-Book ratio may capture not only past equity market timing but also the firm’s growth opportunities. Welch ( 2004 ) and Alti ( 2006 ) have attempted to explain the timing of the market by using other indicators such as market appreciation of share price. As noted by Miglo ( 2013, pp. 15-16 ) , “ Only a few theoretical models exist on market timing…To be comparable with trade-off theory or pecking order theory, market timing models should be able to explain a broader set of phenomena about capital structure than currently exists.”
2.3
Conclusion
In this chapter , the traditional theoretical approaches and the major modern theories related to corporate capital structure decisions have been discussed , paving the way for the review of the prior empirical studies based directly or indirectly on these theoretical frameworks and conducted in the contexts of International and Indian scenarios , in the following chapter.
11
Rajan and Zingales ( 1995 ), Hovakimian, Opler and Titman ( 2001 ) and Baker and Wurgler ( 2002 ).
CHAPTER 3
34
Chapter 3 Review of Empirical Literature 3.1 Introduction The theories of capital structure have been discussed in the previous chapter. However, a review of prior empirical studies, based directly or indirectly on these theories, is necessary for the identification of the research problem(s) to be addressed by this study . The second and third sections of this chapter presents such a review , in chronological order , of the important 1 empirical studies on capital structure in the contexts of International and Indian scenarios respectively. The last section concludes this chapter.
3.2 International Studies 1)
Chudson ( 1945 ) [ USA ]2 studied patterns of financial structure on a cross section
of manufacturing, mining, trading and construction companies mainly for 1937. It was observed that companies with high proportion of fixed assets tended to use more longterm debt ; and that the ratio of long-term debt to total assets varied widely across industry, and varied irregularly and inversely with size and profitability respectively. 2)
Modigliani and Miller ( 1958 ) [ USA ] based on cross-sectional samples of 43
electric utilities during the period 1947-1948 and 42 oil companies during 1953, regressed the weighted average cost of capital ( estimated as net operating cash flows after taxes scaled by the firm’s market value ) against financial leverage ( measured as the ratio of the market value of debt to the market value of firm ). The results ( slope coefficients of leverage for both electric utilities and oil companies being 0.006 were not significantly different from zero ) suggested that capital structure does not affect cost of capital, lending support to their propositions ( M-M, 1958 ).
1
It is practically impossible to review each and every study on capital structure owing to the enormous volume of such studies over a period of more than five decades. 2
The country / region under study is mentioned within parenthesis.
35
3)
Barges ( 1963 ) [ USA ] tested the M-M propositions by conducting linear and non-
linear 3 regression analyses on cross-sectional samples ( and sub-samples ) of 61 railroads companies, 63 department store companies, and 34 cement companies for the period from 1954 to 1956. The empirical results contradicted the M-M propositions and mostly validated the intermediate approach. 4)
Weston ( 1963 ) [ USA ] criticized M-M ( 1958 ) study arguing that : ( i ) the data for
electric utility sample included insufficient number of observations over certain ranges of capital structure
4
to justify drawing inferences about the shape of the relations over the
entire range ; ( ii ) the oil industry could not be said to be homogeneous in terms of business risk ; and ( iii ) the assumption of perpetual cash flows without growth was not feasible. By incorporating growth of cash flows and firm size as additional explanatory variables in his cross sectional regression study of 59 electrical utilities for 1959, he observed that weighted average cost of capital decreases with leverage which is consistent with the existence of a gain to leverage, that is , tax shield on debt has value. 5)
Miller and Modigliani ( 1966 ) [ USA ] conducted an empirical study on the effects
of leverage on cost of capital and on market value of the firm, based on a cross-sectional sample of 63 electric utility firms for the periods 1954, 1956 and 1957. The market value of the firm was attributed to the present value of operating cash flows generated by: assets-in-place, tax subsidy on debt, growth potential and firm size. The results, which indicated that tax subsidy on debt contributed a significant amount of about 26% on average to the value of the firm, were consistent with notion that debt-tax shelter has value. 6)
Schwartz and Aronson ( 1967 ) [ USA ] were the first to look for empirical
verification of the optimal financial structures after the Modigliani-Miller controversy began. They investigated the effect of ‘industry’ on the financial structure ( measured by the ratio of common equity to total assets , both at book values ) based on cross-sectional data for four groups of industries ( railroad , electric and gas utility , mining and industrial companies) for two periods, 1928 and 1961. Applying one-way analysis of variance technique which used the F-ratio or variance ratio test of statistical significance, they find that, although financial structures within industries does not vary significantly, there are significant differences in the financial structures among different industries for a given
3 4
Second-order term for leverage.
Only eight observations having debt - to - total market value ratios between 0% and 50%, and the bulk of the observations falling within the range of 50% and 80% .
36
year. They also find that there are insignificant differences in equity-ratio or other important financial ratios over time for the same industry and suggest that a relationship between leverage and business risk, as proxied by industry classification, may exist. The analysis indicates the existence of optimal capital structure for each industry and stability of financial structure over time. 7)
Gupta ( 1969 ) [ USA ] conducted a comprehensive study focusing on the analysis
of various financial ratios ( namely, profitability ratios, turnover ratios, leverage ratios and liquidity ratios ) with respect to three exogenous variables : industry, size ( measured by the amount of total assets ) and growth ( measured by the annual average compounded growth rate in sales ). A cross-sectional sample of 1,73,000 corporations covering 21 industries, classified into 13 size categories ( ranging from total assets of less than $50,000 to assets of $250 million and more ) were examined for the year 1961-62. The study confirmed that leverage ( measured as the ratio of total debt to total asset ) were positively related to growth and negatively related to size at the firm level. Significant industry effect on debt ratio was also found. It was further observed from the individual analysis of selected industries that “family pattern of ownership “appeared to be an important determinant of leverage in the paper and allied product industry. 8)
Baxter and Cragg ( 1970 ) [ USA ], in their study on the choice of long term
financing instruments, analyzed 230 security issues ( 131 debt, 33 equity, 38 convertible, 5 preference, and 23 combined issues ) made during 1950-65, applying Logit and Probit Regression Analyses. They concluded that : ( a ) companies raising relatively large sums favoured debt, possibly due to concern over control ; and ( b ) companies with high ratios of market capitalization to total assets favoured equity possibly reflecting ‘timing’ consideration. 9)
Remmers, Stonehill, Wright and Beekhuisen ( 1974 ) [ International ] conducted a
study to analyse whether industry-class and firm size are the determinants of leverageratio ( total debt to total assets , both measured at book values ) in the USA and in the international scenario. The results of a one-way analysis of variance test run on firms from nine American industries did not support the hypothesis that industry-class is a determinant of leverage-ratio. Further examination of a sample of four manufacturing industries in five developed countries, namely, France, Japan, Netherlands, Norway and the USA led to the conclusion that industry did not appear to be determinant of corporate debt ratio in Netherlands, Norway and the USA, but it did appear to be a determinant in France and Japan. They also observed that : ( a ) size did not appear to be a determinant
37
of debt ratio ; and ( b ) institutional variables , such as earning rate and growth rate , seemed to be important determinants of leverage-ratio internationally. 10)
Toy, Stonehill, Remmers, Wright and Beekhuisen ( 1974 ) [ International ] conducted
a study to test the hypothesis that growth, profitability, and risk are determinants of corporate debt ratios based on a sample of 816 firms in four manufacturing industries in five industrialized countries ( U.S.A, France, Japan, Norway and Netherlands ) during the period 1966 -1972. Application of linear Ordinary Least Squares ( OLS ) regression methodology showed : ( a ) significant relationships in all countries except France ; ( b ) higher debt levels were found to be associated with higher earnings risks ; ( c ) varying cultural perceptions of debt ratios were attributed to varying debt ratios between countries , that is , Japan was expected to have higher debt ratios than the U.S.A . 11)
Scott and Martin ( 1975 ) [ USA ] analysed the influence of industry and firm size
on financial structure ( that is , equity ratio measured by the ratio of common equity to total assets, both at book values ) based on a cross-sectional sample of firms from 12 industries during the period 1967 to 1972. The parametric one-way Analysis of Variance ( ANOVA ) and the non-parametric Kruskal-Wallis one-way ANOVA by ranks techniques were applied. The results of the analysis led to the conclusion that industry class and firm size may be said to determine the financial structure. 12)
Taub ( 1975 ) [ USA ] attempted to examine the relationship between debt-equity ratio
of the firm and its choice of new financing using logit analysis to examine 172 issues of equity and bonds made during the period 1960-69 , with the help of the following explanatory variables : ( a ) the difference between the expected future return on firm’s capital and pure rate of interest, ( b ) the uncertainty of the firm’s future earning , ( c ) firm size, ( d ) tax rate , and ( e ) the firm’s period of solvency. The empirical results showed that differences between return to the firm and long term rate of interest and size were directly related to debt-equity ratio , the uncertainty of the firm’s earning was indirectly related to debt-equity ratio and the results for the remaining variables were not significant. 13)
Taggart ( 1977 ) [ USA ] presented an integrated model of corporate financing pattern
of non-financial firms during 1957-1972 using stock-adjustment model and observed that the level of sales had positive effect on liquid assets and that timing considerations appeared to exert a significant influence on corporate financing decision. It was stated that when the debt-equity ratio is below target, firms issue more bonds and less stock and when permanent capital is below the target, firms issue more of both bonds and stock. The study concluded that : ( a ) bonds are substituted for equity issue when the stock
38
market is depressed and market value of debt-equity ratio is a determinant of long-term debt capacity ; and ( b ) firms base their stock and bond issue decision on the need of permanent capital and their long-term debt capacity. 14)
Ferri and Jones ( 1979 ) [ USA ] examined the determinants of financial structure
with the objective to investigate the relationships between the financial structure of a firm and its industry-class, size, variability of income and operating leverage. The results of the study indicated that : ( a ) industry-class was linked to a firm’s leverage, but not directly as indicated in previous studies ; ( b ) a firm’s use of debt is related to its size, but the relationship is not linear and positive as indicated in previous studies ; ( c ) variation in income is not associated with a firm’s leverage ; and ( d ) operating leverage does influence financial leverage and these two types of leverage are more or less linearly and negatively related as suggested by financial theory. 15)
Errunza ( 1979 ) [ Central America ] conducted a study on the determinants of
financial structure in the Central American Common Market ( CACM ) Countries ( namely, Costa Rica, Guatemala, Honduras and Nicaragua ) using a sample of 15 large domestic private sector companies for each of the countries. Applying two-way analysis of variance by ranks it was observed that : ( a ) firm’s financial leverage was not associated with firm size ; ( b ) the negative relationship between financial leverage and coefficient of variation in EBIT showed that risky firms were more likely to employ low percentage of debt in their financial structure ; ( c ) firm’s growth rate did not seem to be associated with leverage, and the relationship did not turn out to be positive as indicated in other works; ( d ) there was a negative relationship between dividend payout and leverage ratio, though cause-and-effect relationship between them was not clear ; ( e ) the earning rate was negatively related with leverage ; ( f ) the degree of operating leverage does not influence the use of debt ; and ( g ) financial leverage and debt service capacity was found to be negatively related. 16)
Masulis ( 1980 ) [ USA ] conducted a cross-sectional study on the effect of stock
prices on leverage based on exchange offers, or swaps 5 . For a sample containing 106 leverage-increasing and 57 leverage-decreasing exchange offers for the period 1962-1976, highly significant announcement effects were found. For the Wall Street Journal announcement date and the following day, the announcement period return was 7.6% for leverage-
5
In an exchange offer or swap, one class of securities is exchanged for another and it does not simultaneously effect the assets structure because of no cash involvements.
39
increasing exchange offers and the return was-5.4% for leverage-decreasing exchange offer. The author directly examined a sample of 18 non-convertible debt issues without any covenants to protect against the issuance of new debt with equal seniority. The announcement period return was observed to be -0.84% and it was statistically significant. Also, two-day announcement returns of 3.3% and 3.6% were observed for a sample of 43 preferred-for-common stock exchange offers and a sample of 43 debt-for-preferred exchange offers respectively. It was concluded that : ( a ) stock prices are positively related to leverage changes because a gain in value induced by debt tax shield and a positive signaling effect ; ( b ) leverage increases induced wealth transfers across security classes with the greatest effect on unprotected convertible debt ; ( c ) debt holders’ wealth is expropriated by shareholders in leverage-increasing offers ; and ( d ) higher leverage is a signal of the confidence of the mangament in the future of the firm. 17)
Marsh ( 1982 ) [ UK ], in his study, focused on firm’s choice of financing instruments
at a given period in time. He developed the descriptive model of the choice between long-term debt and equity and the coefficients of the models were estimated by using logit analysis of sample of 748 issues of equity and debt made by companies over the period 1959-74. The following proxy variables were used as determinants of target debt ratio: ( i ) size ( logarithm of capital employed ), ( ii ) risk ( standard deviation of EBIT ), and ( iii ) asset structure ( ratio of fixed to total assets ). The empirical results indicated positive relations between firm size and debt ratio and fixed assets and debt ratio, and negative relation between risk and debt ratio. It was concluded that the timing and market condition are different for debt issue and equity issue and that the firm’s past history and market condition heavily influence the choice between debt and equity financing, lending support to the Market Timing Theory of capital structure. 18)
Bowen, Daley and Huber ( 1982 ) [ USA ] conducted a study based on a random
sample of 90 non-regulated firms ( 10 each from 9 industries ) for the periods from 1951 to 1969 , with a view to examine the relationship between leverage and industry classification both cross-sectionally and across time, and also to test empirically the DeAngeloMasulis ( 1980 ) propositions regarding the role of non-debt tax shields in determining an optimal capital structure. Applying various parametric and non-parametric one-way ANOVA and multiple analysis of variance ( MANOVA ) tests, the following conclu-sions were drawn from the study : ( a ) a statistically significant difference was found to be present between mean industry financial structures ; ( b ) statistically significant stability over the entire time period studied was demostarted by the ranking of mean industry financial structures;
40
( c ) firms showed a tendency to move towards their industry mean over both five and ten years period ; and ( d ) in case of non-regulated firms at the industry level , a significant role was played by non-debt tax shields in determining the optimal use of debt in capital structure. 19)
Castanias ( 1983 ) [ USA ] examined the relationship between the probability of
bankruptcy ( failure rates ) and leverage ratios using cross-sectional data related to a sample of 36 lines of business ( broadly under manufacturing, retail and commercial services ) for the periods 1940, 1950, 1960, 1970 and 1972-77. Kendall ( rank-order ) and Pearson ( productmoment ) correlation coefficients were calculated. The empirical results were not consistent with Miller ( 1977 )’ s capital structure irrelevance model but were consistent with a variant of the tax shelter-bankruptcy cost model wherein firms in lines of business having high failure rates were found to have less debt in their capital structures. 20)
Bradley, Jarrell and Kim ( 1984 ) [ USA ] used cross-sectional data relating to a
sample of 851 firms ( from 25 industries ) during the period 1962-81 to test the impact of the following three firm specific attributes : ( a ) volatility , ( b ) non-debt tax shield , and ( c ) intensity of Research & Development ( R & D ) and advertising expenses, on leverage ratio. Correlation analysis, standard Analysis Of Variance ( ANOVA ) test and linear Ordinary Least Squares ( OLS ) regression analysis were conducted. It was observed that volatility and intensity of R & D and advertisement expenditures were negatively related to leverage ratio ; non-debt tax shield was positively related to leverage ; and industry class was found very significant factor for debt-equity choice. The findings for volatility and financial distress costs were consistent with static trade-off theory. But the finding of non-debt tax shield was somewhat puzzling and the authors noted non-debt tax shields act as instrumental variable for the securability of the firm’s assets, with more securable assets leading to higher leverage ratios. 21)
Kim and Sorensen ( 1986 ) [ USA ] conducted a study to examine the presence of
agency costs and their relation to the corporate debt policy and also to analyse the relationship between corporate leverage ratios and other variables such as operating risk, growth rate, and firm size , based on a sample of 168 firms listed on Value Line Investment Survey database for the years 1978, 1979 and 1980. The sample was divided into two equal groups : one with comparatively high level of insider ownership and the other with comparatively low level of insider ownership. The results of regression analysis showed that : ( a ) higher level of insider ownership is associated with higher leverage ratio , wich may be explained by agency costs of debt and/or equity ; ( b ) less debt ( instead of more
41
debt ) are used by firms with high growth ; ( c ) more debt ( instead of less debt ) are used by firms with high operating risk ; and ( d ) firm size and debt level are uncorrelated. 22)
Kester ( 1986 ) [ USA and Japan ] tested the hypothesis whether Japanese manufac-
turing firms were more highly leveraged than manufacturing firms in the U.S.A . The determinants of capital structure taken into consideration were growth, profitability, risk, size and industry classification. The sample included 344 Japanese firms and 452 U.S.A firms from 27 different industries for the period 1982-83. The debt-equity ratio was measured based on both on market value and book value. The results of the study showed that when leverage was measured on : ( a ) market value basis, no significant country differences between U.S.A and Japanese manufacturing firms were observed , after controlling for growth, profitability, risk, size and industry classification ; ( b ) book value basis, Japan was found to have significantly higher level of leverage than U.S.A , although this result was concentrated among the mature and heavy Japanese industries. 23)
Titman and Wessels ( 1988 ) [ USA ] introduced a factor-analytic technique for
examining the impact of the following eight firm-specific determinanats of capital structure: collateral value of asset, non-debt tax shield, growth, product uniqueness, industry classification, size, volatility, and profitability. Based on a sample of 469 firms over the period from 1974 to 1982, and using the maximum-likelihood method of estimation, the results of the study indicated that : ( a ) product uniqueness was statistically significant and inversely related to debt ratio , consistent with the view that firms which are able to potentially impose high cost on their customers, workers and suppliers in the event of liquidation tend to have lower debt ratios ; ( b ) firm size was statistically significant and negatively related to short-term debt ratios , as small firms may be said to incur relatively high transaction costs when issuing long-term securities ; ( c ) proftability was significant and negatively related to debt ratio lending support to the pecking order theory. 24)
Sekely and Collins ( 1988 ) [ International ] , based on sample of 677 firms belonging
to 9 industries and 23 countries , conducted a study for examining the possible existence of cultural influence ( beyond industry and country influences ) on capital structure. Applying Kruskal-Wallis test for differences of ranks between multiple samples , it was observed that : ( a ) significant country effects and minimum industry effects on capital structure due to cultural differences may be said to exist ; ( b ) some inter-country influences on capital structure resulting from similar cultural patterns among group of countries may also be said to exist.
42
25)
Friend and Lang ( 1988 ) [ USA ] examined the impact of managerial self-interest
on capital structure decision based on a sample of 984 firms listed on the New York Stock Exchange ( NYSE ) from 1979 to 1983. They hypothesized that ceteris paribus, management in closely held corporations would have higher unique risk than in publicly held firms and would have less constraints on its behaviour so that a more negatively significant impact of its investment on debt should be obtained. To test this hypothesis, they classified the sample into two equal-size groups : one being ‘closely held’ and another being ‘publicly held’ corporations depending upon the fraction of stock owned by managerial insiders. They further divided these groups into two subgroups : one with nonmanagerial principal investors and another without non-managerial principal investors. In their econometric model, they incorporated the following explanatory variables : asset structure, profitability, size, volatility, market value of equity held by dominant managerial insider, fraction of equity held by dominant managerial insider having more than 10% share, and fraction of equity held by dominant non-managerial stockholder who holds more than 10% share but not the officer or director. Based on their empirical evidences, the authors concluded that : ( a ) management in ‘closely held’ corporations have higher ability and desire to adjust debt ratio according to its own interest despite the existence of nonmanagerial investors ; ( b ) the level of debt decreases as the level of management investment in the firm increases, reflecting greater non-diversifiable risk of debt to management than to public investors for maintaining a low debt ratio ; and ( c ) for corporations having large non-managerial investors, the average debt ratio is significantly higher than in those with no principal stockholders, thereby suggesting that the existence of large non-managerial stockholders might make the interests of managers and public stockholders coincide. 26)
Fischer, Heinkel, and Zechner ( 1989 ) [ USA ] developed a model of dynamic capital
structure choice considering recapitalization costs. The empirical analysis , based on a sample of 999 firms over a 34 - quarter period from the third quarter of 1977 to the fourth quarter of 1985 , tested the hypothesis that firms with large ( low ) debt ratio ranges, have a low ( high ) effective corporate tax rate , a high ( low ) variance of underlying asset value, a small ( large ) asset base , and low ( high ) bankruptcy costs , applying Weighted Least Squares technique. A firm’s debt ratio range{ measured as the difference between the maximum and the minimum debt ratio over the 34-quarter sample period , with debt ratio defined as total liabilities ( or long term debt ) divided by the sum of total liabilities ( or long -term debt ) and equity market value } was used as an empirical measure of capital structure relevance. The results indicated that larger ( smaller ) , less ( more ) risky
43
firms with higher ( lower ) effective tax rates have narrower ( wider ) observed debt ratio ranges ; and firms in the manufacturing industry , where bankruptcy costs are more likely to be high ( low ), also have narrower ( wider ) debt ratio ranges. 27)
Pinegar and Wilbricht ( 1989 ) [ USA ] in a survey study of capital structure choice
analyzed responses received from the managers of 176 firms chosen out of the list of fortune 500 firms for 1986. Managers of 121 firms indicated that they follow a financing hierarchy ( pecking order ), while that of 47 firms indicated that they seek to maintain a target capital structure. The financing hierarchy showed that the managers first prefer internal equity ( retained earnings ) for financing new projects. The next priority goes to straight debt, convertible debt, external common equity, straight preferred stock and convertible preferred stock in a sequence. So the projected cash flow from the asset is the major determinant of the choice of the managers among various sources of capital, leading to the conclusion that pecking order theory may be said to be more likely to be followed by corporate managers than maintaining a target debt-equity ratio. 28)
MacKie‐Mason ( 1990 ) [ USA ] , based on a sample of 1,747 observations of regis-
tered public offerings over the period 1977 to 1987 , primarily analysed the effect of taxes on corporate financing decisions after controlling for other firm-specific determinants ( probability of costly bankruptcy, the potential for investment inefficiencies due to moral hazard, and signaling costs of equity ) , unobserved industry fixed effects, such as interindustry variations in the degree of asymmetric information between managers and investors ( through industry dummy variables ) and unobserved macro-economic effects ( through time dummy variables ). The results of the application of Probit regression technique showed significant and substantial tax effects on financing choices. 29)
Chiarella, Pham, Sim and Tan ( 1992 ) [ Australia ] examined the importance of
capital structure determinants based on a sample of 226 firms over the perod from 1977 to 1985. It is observed that : ( a ) non-debt tax shields are significantly and negatively related to debt ratios, consistent with De Angelo and Masulis ( 1980 )’s proposition that non-debt tax shields can be substituted for debt-tax shields ( that is, tax shields from interest on debt ); ( b ) profitability is significantly and negatively related to debt ratios lending some support to the pecking order theory ; ( c ) some size effect is present indicating that more debt may be employed by larger firms ; ( d ) growth opportunities and collateral value cannot be said to be significant determinats of capital structure.
44
30)
Hamid and Singh ( 1992 ) [ International ] analysed the corporate finance charac-
teristics of the top 50 manufacturing firms from nine developing countries ( India, Thailand, Jordan, Malaysia, Taiwan, Mexico, Pakistan, Zimbabwe and South Korea ) over the period 1980-1987. The observation of lesser usage of internal finance by firms in developing countries compared to firms in developed countries was attributed to lower retention ratios and varying growth rates. 31)
Mehran ( 1992 ) [ USA ] examined the relationship between a firm’s capital structure
and its executive incentive plans, managerial equity investment, and monitoring by the board of directors and major shareholders based on a sample of 124 randomly selected manufacturing firms during the period 1979‐1980. The results of regression analysis, which showed positive relationship ( s ) between the firm’s leverage ratio and : ( a ) percentage of executives’ total compensation in incentive plans, ( b ) percentage of equity owned by managers, ( c ) percentage of investment bankers on the board of directors, ( d ) percentage of equity owned by large individual investors, and ( e ) percentage of ownership by individual investors, indicated that agency costs between managers and shareholders and firm’s capital structure may be said to be related. 32)
Shuetrim, Lowe and Morling ( 1993 ) [ Australia ] explored cross-sectional and time
variations in capital structure applying fixed effects panel data regression technique to a sample of 209 firms between 1973 and 1991, considering variables that vary : ( a ) across both firms and time ( firm size , growth , collateral or tangibility , cash flow , potential debt tax shield and tax exhaustion dummy variable ) ; ( b ) only over time , that is macroeconomic variables [ real asset prices s, consumer price inflation and fund cost differential ( the differential between real cost of debt and real cost of equity ) ] ; ( c ) only across firms [ unobserved industry effect ( through the inclusion of industry dummy variables ) and listing category dummy variable to distinguish between listed and unlisted firms ]. It was observed that the most important among these variables to influence leverage were firm size, growth, collateral, cash flow and real asset prices. 33)
Balakrishnan and Fox ( 1993 ) [ USA ] conducted an empirical investigation of the
importance of specialized assets and other unique characteristics of a firm in order to explain the variance in capital structure across firms based on a sample of 295 firms belonging to 30 industries over the period 1978-1987. The independent variables used to determine the leverage of the firm included risk, depreciation, research and development expenses ( R and D ), advertising expenses, and growth. The relationship between leverage and growth was found to be insignificantly negative. There existed a negative relation
45
between R and D and leverage , while the relation between advertising expenses and leverage was significantly positive. It was concluded that unique firm-specific assets and skills were the most important determinants of capital structure. The firm-specific effects contributed most to the variance in leverage, suggesting a strong link between strategy and capital structure. 34)
Klein and Belt ( 1993 ) [ USA ] , considering a sample of all non-financial and non-
regulated firms for the period 1983-1988 , attempt to test the likelihood that a firm will choose internal over external sources of finance, and to model the probability of choosing between debt and equity. Application of contingency table analysis provides evidence that faster growing firms which have already employed internal sources finance to a greater extent must employ external sources of finance. Moreover , application of logit analysis suggests that firms with lower asymmetric information tend to employ external sources to finance the majority of their required funds with debt being the most preferred choice. These observations effectively provide more support for the pecking order hypothesis. 35)
Hatfield , Cheng and Davidson ( 1994 ) [ USA ] , considering a sample of 183 firms
( belonging to 55 industries ) which announced a new debt issue for the period January 1, 1982 to December 31, 1986 , examine Masulis (1983)’s hypothesis that “ when firms which issue debt are moving toward the industry average from below, the market will react more positively than when the firm is moving away from the industry average” , by classifying firms’ leverage ratios as being above or below their industry average prior to announcing a new debt issue , and testing whether this has an effect on market returns for shareholders. A single post-event interval of day 2 to 90 depicted a slow, negative effect ( - 3.2% ) following the debt issue. Firms with high debt had significant negative market reactions for several intervals ; however, the difference between these firms and firms with low debt was not statistically significant. It is concluded that the market does not seem to be concerned with the relationship between a firm’s debt level and that of its industry. 36)
Rajan and Zingales ( 1995 ) [ International ] attempted to examine the determinants
of capital structure of public firms in the G-7 countries, namely, United States of America ( USA ) , Japan , Germany , France , Italy , United Kingdom ( UK ) , and Canada. The crosssectional study , based on a sample of 4557 non-financial firms from 1987 to 1991, focused on the following four determinants of capital structure : tangibility of assets, investment opportunities ( growth ) , firm size and profitability. It was observed that Germany and UK were the lowest levered on an average. Application of regression analysis using maximum likelihood method based on censored Tobit model showed that tangibility of assets and
46
the size were significantly positively related to leverage and growth opportunities and profitability were significantly negatively related to leverage for all the countries except Itay. It was also observed that firms having majority state ownership appeared to employ higher levels of leverage. The study concluded that : ( a ) firm leverage, at an aggregate level , were reasonably similar across G-7 countries ; and ( b ) though the determinants identified to be related to leverage in the United States by previous cross-sectional studies seem similarly related in other countries as well , a thorough investigation of the evidence conveyed the irresoluteness of the theoretical foundations of the observed relations. 37)
Johnson ( 1997 ) [ USA ] examined the relations between corporate debt ownership
structure and several firm’s characteristics such as age, size, volatility, market-to-book ratio, collateral value of assets and fixed asset ratio. The sample size of 847 firms for 1989 was taken for analysis. It was concluded that firms used more public debt if they face lower information and monitoring cost, have a lower likelihood and costs of inefficient liquidation and have fewer incentives to take actions harmful to lenders. Bank debt use and private non-bank debt use were both statistically related to leverage, the fixed asset ratio and the market-to-book ratio, but the signs of relationships were opposite across the sources. The only similarity found between the determinants of the two sources was that both were negatively related to age. 38)
Fama and French ( 1998 ) [ USA ] use Fama and MacBeth ( 1973 )-type cross-
sectional regressions
6
of market value of firm on earnings, investment, and financing
variables to measure tax effects in the pricing of dividends and debt. The study is based on a sample of 2400 firms for the period 1965 to 1992. The explanatory variables include past, current, and future values of dividends, interest, earnings, investment, and R and D expenditures. Controlling for profitability ( pretax expected net cash flows ), the marginal relation between leverage and value is typically negative, rather than positive ; thus producing no indication that debt has net tax benefits. However, the results could be interpreted in terms of Miller’s ( 1977 ) hypothesis that leverage has no net tax benefits because personal taxes on interest offset the corporate tax savings. It seems more likely that leverage conveys information about profitability that is missed by the control variables, although the regressions fail to measure how ( or whether ) the tax effects of financing decisions affect firm value.
6
where time - series standard deviations of the slopes in the year - by - year cross - sectional regressions are used to construct standard errors for the average slopes.
47
39)
Shyam-Sunder and Myers ( 1999 ) [ USA ] conducts tests for the Target Adjustment
Models ( TAM ) relating to trade-off theory against the Pecking Order Model ( PEM ) based on a sample of 157 firms over the period from 1971 to 1989. The results of regression and subsequent simulation analyses suggest that : ( a ) PEM provides an excellent first-order description of corporate financing behaviour ; ( b ) TAM seems to perform well when tested independently; ( c ) in case of joint testing of the two models , the coefficients and significance of the PEM hardly vary whereas TAM’s performance declines although coefficients are statistically significant ; and ( d ) TAM is not rejected even when false , but PEM can be easily rejected when false indicating the tests have power regarding PEM. It is thus concluded that greater confidence is displayed in PEM than in TAM. 40)
Hirota ( 1999 ) [ Japan ] explores the determinants of capital structure using four
cross-sectional samples of 407 to 546 Japanese firms for 1977, 1982, 1987 and 1992. He seeks to explain the leverage of these firms by a combination of conventional capital structure variables ( non-debt tax shields , asset tangibility , growth opportunities , business risk , profitability , and size ) and Japanese institutional variables, including : bank relationships ( measured by the proportion of debt due to the largest bank lender ), keiretsu membership, regulation of new equity issues ( measured by a dummy representing firms who satisfy the voluntary code enforced by major Japanese security companies between 1973 and 1996 ), and a variable representing a firm’s incentive to exploit free cash flows ( a firmspecific debt-equity yield differential ). Almost all the variables in both groups entered the cross-sectional regressions with the expected sign for each sample, and most were significant. This suggests that conventional capital structure theory can help understand the behaviour of firms in a country that is usually thought to be either “non-Anglo-Saxon” or at least bank-based. But the results for the institutional variables also show that there is more to firm financial behaviour in Japan than is captured by the conventional variables. 41)
Wiwattanakantang ( 1999 ) [ Thailand ] analysed the determinants of the capital
structure based on a cross-sectional sample of 207 non-financial firms listed on the Securities Exchange of Thailand for the calendar year 1996. OLS regression technique was applied to linear models with the dependent variable ( s ) being book value and market value measures of leverage and various combinations of independent variables such as size, profitability , tangibility , business risk , non-debt tax shields , tax exhaustion dummy variables, firm age , industry dummy variables , and seven proxy variables for ownership structure ( namely, family-owned firms, conglomerate groups, foreign-owned firms, stateowned firms, size of the board of directors, managerial ownership, and the degree of
48
ownership concentration ). The results of the study suggest that : ( i ) tax effect , signaling effect and agency costs play a significant role in financing decisions ; (ii ) firm owned by single family have significantly higher level of debt and managerial shareholdings influence debt ratio consistently and positively for these firms ; and ( iii ) debt ratio is affected negatively by large shareholders implying their potential role of monitoring the management. 42)
Hovakimian, Opler and Titman ( 2001 ) [ USA ] , based on a sample of 11,136 firm
years covering security issuance behavior and 7,400 firm years covering security repurchase behavior from 1979 to 1997, test the hypothesis that firms tend to move toward a target debt ratio either during raising of new capital ( equity or debt ) or during retiring debt and repurchasing existing equity, through the application of a two stage procedure ( the first stage based on a doubly - censored Tobit regression model and the second stage based on a Logit regression model ). It is observed that movement of a firm towards a target debt ratio seems to more important when choosing between repurchasing of equity and retiring of debt than when choosing between issuances of equity and debt. 43)
Booth, Aivazian, Kunt and Maksimovic ( 2001 ) [ International ] conducted a study
to analyze capital structure choice of firms in 10 developing countries ( India, Pakistan, Thailand, Malaysia, Turkey, Zimbabwe, Mexico, Brazil, Jordan and Korea ) by using firmspecific attributes ( namely , tax rate, business risk, tangibility of asset , size , profitability and market-to-book ratio ) and macroeconomic indicators [ namely, ( stock market value / GDP), ( liquid liabilities / GDP ), real GDP growth rate , inflation rate, and Miller ( 1977 ) tax term]. Pooled and fixed effects models were applied to the data consisting of the largest companies in each country over the period 1980 to 1990. It was observed that : ( a ) debt ratios in developing countries may be said to be affected similarly by the same independent variables that are significant in developed countries ; and ( b ) there are systematic differences across countries in the manner by which the debt ratios are affected by macroeconomic factors. 44)
Graham and Harvey ( 2001 ) [ USA ] conducted an anonymous survey , consisting
of 14 main questions ( which along with the sub-parts totaled over 100 questions ) , on 392 Chief Financial Officers ( CFOs ) about the cost of capital, capital budgeting, and capital structure. Information on 12 characteristics of the firms and management ( firm size , foreign sales, industry, CEO’s education, CEO’s age, CEO’s tenure, ownership, dividend payment, whether they were regulated, and the proportion of common stock that the top three executives
49
owned if all their options were exercised , debt-equity ratios and debt ratings ). The analysis showed that many survey responses differed by firm and management characteristics. 45)
Ozkan ( 2001 ) [ UK ] extended the empirical research on the dynamic trade-off
theory of capital structure using an unbalanced panel data of 390 non-financial and nonregulated firms from 1984 -1996. The explanatory variables included size, growth, non-debt tax shield, profitability and liquidity. Three types of regression techniques, namely, pooled OLS, Anderson and Hsiao ( AH ) type instrumental variable estimators and Generalized Method of Moments ( GMM ). Comparing the statistics, GMM was found as strong estimation technique whereas the AH type estimation was found poor and yielded larger variance then GMM. Profitability, liquidity, non-debt tax shield and growth opportunities were found to be negatively related to leverage whereas size was found to be positively related to leverage. The speed of adjustment toward target level was observed as 0.5 from all three methods of estimation. It was concluded that firms have long-term target leverage ratios and they adjust to the target ratio relatively fast, implying that the costs of being away from their target ratios and the cost of adjustment are equally important for firms. 46)
Pandey ( 2001 ) [ Malaysia ] examined the determinants of capital structure by
utilizing a sample of 106 non-financial firms for the period 1984 to 1999 , and classifying the data into four sub-periods of four years each ( namely , 1984-87 , 1988-91 , 1992-95 and 1996-99 ) corresponding to different stages of Malaysian capital market. Book value and market value debt ratios ( decomposing debt into total debt , long term debt and short trem debt ) were used as the dependent variable. Cross-sectional OLS, Pooled OLS and Fixed Effects regression techniques were applied. The results of pooled OLS regressions showing that profitability , size , growth , risk and tangibility significantly affected all types of debt ratios were found to be consistent with the results of fixed effect estimation with the exception that risk was found to loose its significance. Investment opportunity or Marketto-book value ratio was found to have no significant impact on the debt ratios. Profitability was found to be negatively related to all the debt ratios in all time periods and under all estimation methods , lending support to the pecking order theory. 47)
Bevan and Danbolt ( 2002 ) [ United Kingdom ] with a view to examine the
determinants of capital structure extended Rajan and Zingales’ ( 1995 ) analysis of the UK by : ( i ) analysing the robustness of their conclusions to variations in the gearing measure ( using book values and market values of four different measures of gearing or leverage), and , ( ii ) decomposing the analysis into long-term debt and short-term debt and their subelements, considering a cross-sectional sample of 822 non-financial firms over a period of
50
four years from 1987 to 1991. The explanatory variables included size, tangibility, profitability and growth opportunities. OLS regession technique was applied. In case of the first analysis, gearing was found to be significantly positively correlated with tangibility and size, and significantly negatively correlated with profitability and growth opportunities. The results of decompositional analysis showed that : ( a ) all the components of short-term and long-term debt were significantly negatively correlated with profitability ; ( b ) almost all the components ( except short-term securitized debt ) of short-term debt were significantly negatively related to tangibility, whereas all the components of long-term debt demonstrated significant positive correlation ; ( c ) size was found to have significant negative correlation with all short-term bank borrowings and significant positive correlation with all long-term debt forms and short-term paper debt ; and ( d ) growth opportunities were found to be significantly positively correlated with total current liabilities and trade credit. It was concluded that a detailed analysis of all forms of corporate debt was required for a fuller understanding of the determinants of capital structure. 48)
Baker and Wurgler ( 2002 ) [ USA ] attempted to examine the hypothesis that equity
market timing has large and persistent effects on capital structure , considering a sample including 2,839 observations on firms at the first fiscal year end after Initial Public Offering ( IPO ) , 2,652 observations on firms in the next fiscal year ( IPO + 1 ) , and so on, down to 715 observations on firms at 10 years after the IPO ( IPO + 10 ). Ordinary Least Squares ( OLS ) and Fama-Macbeth regresssions are applied. The results indicated that : ( a ) past market timing opportunities, proxied by an external-finance-weighted average of past market-to-book ratios, was statistically significant and negatively related to debt ratios ; ( b ) low-leverage ( high-leverage ) firms tend to raise funds when their equity valuations were high ( low ) ; and ( c ) equity market fluctuations have large effects on capital structure that persist for at least a decade. This study lended much support to the Market Timing Theory of capital structure. 49)
Chui, Lloyd and Kwok ( 2002 ) [ International ] , based on a sample of 5591 firms
across 22 countries and classified into four industries ( namely, primary , manufacturing, advanced manufacturing and services ) for the years 1991 , 1994 and 1996 , examined whether corporate capital structure is affected by national culture. The results of empirical analyses showed that firms in countries with high scores on the cultural factors of “conservatism” and “mastery” tend to use less debt in their capital structures, even after controlling for the industry effect , differences in economic performance , legal systems, development of financial institutions, and other firm-specific determinants of debt ratios in each country.
51
50)
Frank and Goyal ( 2003 ) [ USA ] , adopting a modified version Shyam-Sunder and
Myers’ ( 1999 ) pecking order model , test the theory based on a sample of 768 firms over the period 1971 to 1998 . The results of regression analysis are presented separately for two time periods - the first 19 years ( 1971 to 1989 ) corresponding to the period considered by Shyam-Sunder and Myers’ ( 1999 ) , and the subsequent 9 years ( 1990-1998 ) - each period being segregated into two groups , namely , ‘data with no gaps permitted’ and ‘data with gaps permitted’ in the reporting of funds flow. It is observed that for the period 1971-1989 , the financing deficit coefficient declines sharply from 0.75 to 0.28 when the sample changes from ‘data with no gaps permitted’ to ‘data with gaps permitted’ . Since firms belonging to the group ‘data with no gaps permitted’ tend to be large in size , additional test is applied by segregating this group into quartiles based on total assets ; the financing deficit coefficient is observed to grow strongly with increase in firm sizes. The results for period 1990-1998 indicate that the financing deficit coefficient decline from 0.33 to 0.15 when the sample changes from ‘data with no gaps permitted’ to ‘data with gaps permitted’ . It is conclude that contrary to the pecking order theory, the financing deficit may be said to tracked more closely by net equity issues than by net debt issues. Moreover , some aspects of pecking order behaviour are exhibited by large firms for the period 1971-1989 but the evidence is not robust for the period 1990-1998. 51)
Pao, Pikas and Lee ( 2003 ) [ Taiwan ] , based on a sample of 21 firms belonging
to high-tech industries ( electronics, telecommunications, computer hardware, software, networking, information systems, and other related industries ) and 21 firms belonging to traditional industries ( clothing, textile, trading, agriculture, manufacturing, etc.) over the period from 1996 to 1999 , attempted to analyze the important determinants of capital structure, considering total debt ratio as the dependent variable along with ten independent variables ( namely, firm size , growth opportunities, profitability , tangibility of assets , non-debt tax shields, dividend payments , and business risk ). Four linear models , namely , multiple regression variance-component model, first-order autoregressive model, and variance-component moving average model were applied. It is observed from the results that the determinants of capital structure of the two industries are different. Among the four linear models, the variance component model has the lowest Root Mean Squared Error ( RMSE ) for both industries, indicating that time series and cross-sectional variations play a significant role in analyzing the determinants of capital structure in Taiwanese industry.
52
52)
Welch ( 2004 ) [ USA ], using a dataset of publicly traded firm from the period 1962
to 2000, and based on a dynamic model, demonstates that debt and equity issuing and repurchasing activities are not employed by U.S. corporations for counteracting the mechanistic effects of stock returns on their debt equity ratios. It is observed that over 1-5 years horizons, 40% of debt ratio dynamics can be explained by stock returns. Even though 60% to 70 % of debt ratio dynamics ( long-term debt issuing activity being the most relevant to capital structure decision explaining about 30% of the variation in corporate debt ratio changes) can be explained by corporate ( net ) issuing activity , such issuing motives cannot be fully explained. The study concludes that stock returns : ( i ) act as the primary determinant of capital structure ; ( b ) are probably the only well understood influence of debt ratio dynamics ; and ( c ) when accounted for , render taxes, bankruptcy costs, and other proxies utilized in prior studies to play a very modest role in explaining capital structure. 53)
Voulgaris, Asteriou and Agiomirgianakis ( 2004 ) [ Greece ] , with a view to formu-
late some policy implications for possible improvement in the financial performance of the Greek manufacturing sector , investigate the determinants of capital structure of manufacturing firms by considering panel data of two random samples belonging to 16 industries - one consisting of 143 small and medium sized enterprises ( SMEs ) and another comprising 75 large sized enterprises ( LSEs ) , both covering the period 1988 to 1996. It is observed that : ( i ) profitability is a major determinant of capital structure for both size groups ; ( ii ) efficient assets management and assets growth are found to be essential for the debt structure of LSEs ; and ( iii ) efficiency of current assets, size, sales growth and high fixed assets are found to affect substantially the credibility of SMEs. The observatons imply that the efforts of the SMEs should be focussed on : ( a ) increasing their cash flow capacity through better management of assets and achievement of higher exports , ( b ) ensuring good banking relations , and ( c ) exploring alternative sources of financing ; whereas the LSEs should adopt strategies for improving their competitiveness and securing new sources of financing. 54)
Bancel and Mittoo ( 2004 ) [ USA and Europe ] , with a view to examine the
following hypotheses : ( i ) European and U.S. managers make their capital structure decisions using similar factors, ceteris paribus, ( ii ) Cross-country differences in managerial views on the determinants of capital structure are influenced primarily by the legal system of the home country, ceteris paribus, and ( iii ) Cross-sectional differences in managerial rankings of major determinants of debt and equity policies differ systematically according
53
to the quality of legal systems, ceteris paribus , surveyed managers of 720 firms in sixteen European countries ( namely, Austria , Belgium , Greece , Denmark , Finland , Ireland , Italy , France , Germany , Netherlands , Norway , Portugal , Spain , Switzerland, Sweden , and the U.K. ) on their capital structure choice and its determinants. It is observed that : ( a ) European and U.S. managers use similar factors in making their financing decisions ; ( b ) managers are primarily concerned by financial flexibility and earnings per share dilution in issuing debt and common stock, respectively ; ( c ) since hedging considerations are valued by managers “windows of opportunity” are used by the when raising capital ; ( d ) cross-country variations in the rankings of several major factors and differences in debt policy factors are explained by the quality of the country’s legal system ; ( e ) common stock policy factors are strongly influenced by firm-specific factors , such as growth opportunities ; and ( e ) institutional environment and international operations influence firms’ financing policies. The results suggest that optimal capital structure is determined by most firms by trading off factors such as tax advantage of debt, or bankruptcy costs, agency costs, and accessibility to external financing. 55)
Deesomsak, Paudyal and Pescetto ( 2004 ) [ Asia Pacific ] examined the determinants
of capital structure of firms operating in four countries ( namely , Thailand , Malaysia, Singapore and Australia ) with different legal , financial and institutional environments. The sample consisted of 294 Thai , 669 Malaysian , 345 Singaporean, and 219 Australian firms ( non-financial firms listed on the relevant national stock exchanges ) for the period 1993 to 2001. The explanatory variables included firm - specific variables ( namely , tangibility, profitability , firm size, growth opportunities , non-debt tax shield , liquidity , earnings volatility , and share price performance ) and country - specific variables ( namely , degree of stock market’s activity, level of interest rates, legal protection of creditor’s rights, ownership concentration, and three country dummies ). Fixed effect and pooled OLS regression techniques were applied. The results suggest that the capital structure decision of firms is influenced by the environment in which they operate, as well as firm-specific factors identified in previous studies. The financial crisis of 1997 was also found to have had a significant but diverse impact on firm’s capital structure decision across the region. 56)
Fattouh, Scaramozzino and Harris ( 2005 ) [ South Korea ] present a model of
firm value-maximization in which the firm’s capital structure is a non-linear function of a vector of costs including asymmetric information costs and is subject to a debt ceiling. The study is based on an unbalanced panel data of 4,256 firm-year observations over the period 1992-1999. Applying conditional quantile regression methodology, the authors find
54
clear evidence of heterogeneity in the determinants of capital structure choice. The size of the firm and its rate of growth have a positive impact on debt at low values of the debt ratios, but a negative impact at high values of the debt ratios. By contrast, the proportion of net fixed assets has a negligible impact at low values of the debt ratios, but a significantly positive impact at medium or high values of the debt ratios. It is inferred that observed non-linearities in the determinants of capital structure are consistent with the proposed model. 57)
Eldomiaty, Choi and Cheng ( 2005 ) [ Egypt ] examine the signaling effects of the
determinants covering the basics of tradeoff model , pecking order hypothesis ) of capital structure that are relevant to a transitional market, based on sample of 99 firms from 14 non-financial industries over the period 1997 to 2003. The methodology begins with the determination of the relevant determinants of debt in a transitional economy followed by the examination of the potential signaling effects of the relevant determinants of debt. The robustness of the signaling effect is examined using the ‘Extreme Bound Analysis.’ The overall results indicate that the worthiness of the investment ( market value of firm ) is determined by interest rates ( macro-economic factor ) and financial flexibility ( firm-specific factor ) which have robust and significant signaling effect. 58)
Lee and Yu ( 2005 ) [ Taiwan ] , based on a balanced panel data comprising a sample
of 20 selected electronic companies listed on the Taiwan Stock Exchange during the period 1993 to 2002 , investigate whether firm value is affected by the application of financial leverage. Panel threshold regression model 7 , which may result in threshold effects and asymmetrical relationships between the debt ratio and firm value , is employed to test for the existence of an optimal debt ratio. Four proxy variables for firm value - Return On Assets ( ROA ) , Return On Equity ( ROE ), Earnings Per Share ( EPS ) and Tobin’s q - are considered. A single threshold effect ( the threshold value being 37.48% ) is found to exist between debt ratio and firm value only when the firm value is proxied by Tobin’s q. The estimated value of the two coefficients ( 1 and 2 ) are positive , with 1 in the low debt level being significant, while 2 in the high debt level being insignificant, indicating that in order to maximize the firm’s value , financial leverage should be utilized judiciously by managers . 7
At a threshold debt ratio capital structure decisions can be explained or split by two different linear functions. If the current debt ratio is below this optimal debt ratio, the firm will increase its debt ratio and vice versa.
55
59)
Gaud, Jani, Hoesli and Bender ( 2005 ) [ Sweden ] examined the determinants of
capita structure based on a panel data sample of 104 non-financial firm listed in Swiss stock exchange over the period 1991-2000 and following the same methodology of Ozkan ( 2001 ). Size and assets structure appeared to be significantly positively related to leverage and profitability and growth were found to be significantly negatively related to leverage, lending more support for trade-off theory than the pecking order theory. The speed of adjustment to target capital structure observed as being less than 20% appeared to be very slow in comparison to other European countries and the USA. 60)
Leary and Roberts ( 2005 ) [ USA ] , considering an unbalanced panel data of 127,308
firm-quarter observations ( 3,494 non-financial firms each with a time series of observations ranging in length from 16 to 71 quarters over the period 1984 to 2001 ) , examine whether firms engage in a dynamic rebalancing of their capital structure while allowing for costly adjustment. The results are consistent with the predictions of the modified pecking order theory whereby firms are less ( more ) likely to obtain funds from external capital markets when they have sufficient internal funds ( large investment needs ). Hence , a dynamic rebalancing strategy is followed by firms but adverse selection costs may be an important determinant in their financing decision. 61)
Mackay and Phillips ( 2005 ) [ USA ] examine the importance of industry, firm-
specific and real factors on capital strucuture decisions , based on a sample forming an unbalanced panel of 3074 firms ( 17,140 firm-years ) operating in 315 competitive industries with a HHI ( Herfindahl - Hirschman Index ) measuring industry concentration under 1000, and 309 firms ( 1630 firm-years ) operating in 46 concentrated industries with a HHI over 1800 8 . Applying Ordinary Least Squares ( OLS ) and Generalised Method of Moments ( GMM ) regressions, it is observed that : ( a ) financial structure depends on a firm’s position within its industry in addition to standard industry fixed effects ; ( b ) in competitive industries, a firm’s financial leverage depends on its natural hedge ( that is , its proximity to the median industry capital-labor ratio ), the actions of other firms in the industry, and its status as entrant, incumbent , or exiting firm ; and ( c ) in concentrated industries where strategic debt interactions are stronger, financial leverage is higher and less dispersed , but a firm’s natural hedge is not significant.
8
Following the industry classification scheme used by the U.S. Department of Justice and Federal Trade Commission ( 1997): Unconcentrated industries ( HHI under 1000 ), moderately concentrated industries ( HHI from 1000 to 1800), and highly concentrated industries ( HHI over 1800 ) .
56
62)
Flannery and Rangan ( 2006 ) [ USA ], using an unbalanced panel data of non-
financial and non-regulated 1,11,106 firm-year observations over the period 1965 to 2001, estimate a general partial-adjustment model of firm leverage through the application of panel data fixed effects regression methodology, the analysis of whose results indicate the exisence of target capital structure with a typical firm being able to adjust about onethird of the difference between its actual and its target debt ratio within one year. It is concluded that the observed changes in capital structure may be said to be explained to a greater extent by “ targeting” or “mean - reverting” behaviour than market timing or pecking order considerations. 63)
Huang and Song ( 2006 ) [ China ] examine the determinants of capital structure
based on a sample of 1086 Chinese listed companies over the period 1994 to 2003. The results of cross-sectional OLS regression analysis indicate that leverage : ( a ) increases with firm size, sales growth , tax , volatility and tangibility ; ( b ) decreases with profitability, non-debt tax shields and growth opportunities ; ( c ) correlates with industries ; ( d ) is significantly affected by ownership structure ; and ( e ) is not significantly affected by state ownership or institutional ownership. It is concluded that capital structure choices for the sample of Chinese listed companies may be better explained by static trade-off model rather than pecking order hypothesis. 64)
Dhaliwal, Heitzman and Zhen Li ( 2006 ) [ USA ] , based on a pooled data set of
22,874 firm-year observations covering the period 1982 -2004 , examine the associations between leverage, corporate tax , personal tax , and the implied cost of equity capital. The results of pooled and cross-sectional Ordinary Least Squares ( OLS ) regressions suggest that the equity risk premium associated with leverage is decreasing in the corporate tax benefit from debt and increasing in the personal tax disadvantage of debt , thus lending empirical support to the theoretical proposition of Modigliani and Miller (1963) and De Angelo and Masulis ( 1980 ) that the association between capital structure and cost of equity is affected by taxes. 65)
Sayılgan, Karabacak and Küçükkocaoğlu ( 2006 ) [ Turkey ] , applying fixed effects
panel data methodology on a sample of 123 manufacturing firms listed on the Istanbul Stock Exchange ( ISE ) over the period from 1993-2002 , analyzed the impact of firmspecific characteristics on the corporate capital structure decisions of Turkish firms. The dependent variable ( financial leverage ) was measured as the ratio of book value of total debt to book value of total equity ; and the independent variables included size, profitability and ‘growth opportunities in plant, property and equipment’, ‘growth opportunities in total
57
assets’, non-debt tax shields and tangibility. It was observed that all the independent variables were significant determinants for the capital structure decisions of Turkish firms, with size and ‘growth opportunity in total assets’ being positively related to leverage ratio, and profitability , ‘growth opportunities in plant , property and equipment’ , non-debt tax shields and tangibility being negativey related to leverage ratio. 66)
Delcoure ( 2007 ) [ Central and Eastern Europe ] conducts a study to investigate
whether traditional capital structure theories developed to explain western economies are portable to emerging Central and Eastern European ( CEE ) countries ( namely , Czech Republic , Poland , Russia and Slovakia ). Applications of panel data regresson techinques to an unbalanced panel data comprising of 22 Czech , 61 Polish , 33 Russian , and 13 Slovak publicly traded companies over the period 1996 to 2002 , indicate that : ( a ) financial leverage is positively related to firm size, asset structure, non-debt tax shield, and corporate tax rate, and negatively related to earnings volatility except for short-term leverage ; ( b ) the capital structure choices for firms in CEE countries may be said to be different from those in developed countries , with firms in CEE countries tending to rely more heavily on short-term debt than long-term debt in constrast to firms in developed markets ; ( c ) corporate capital structure choices in the CEE countries may be said to be partially explained by pecking order , trade-off and agency cost theories ; ( d ) the capital structure choices for firms in CEE countries may be explained by the modified pecking order theory wherein the hierarchical choice of financing is retained earnings, equity, bank and possibly market debt. 67)
Eldomiaty ( 2007 ) [ Egypt ] , based on a sample of 99 non-financial firms from 14
industries over the period 1997 to 2003 , attempts to examine the determinants of corporate leverage considering the theoretical predictions of trade-off theory , pecking order theory , and agency cost ( free cash flow ) theory. The results of ordinary least squares regression applied for each of the above capital structure theories indicate that : ( a ) the significant trade-off-related determinants of capital structure are effective corporate tax rate, tangibility and bankruptcy risk ; ( b ) the significant pecking order-related determinants of capital structure are growth and profitability ; and ( c ) borrowing decisions are not affected by the assumptions of free cash flow. It is concluded that the determinants of capital structure in emerging and developed markets appear to converge.
58
68)
DeJong, Kabir and Nguyen ( 2008 ) [ International ] analyze the importance of firm-
specific factors ( namely, firm size, asset tangibility, profitability, firm risk and growth opportunities ) and country-specific factors [ namely, legal enforcement, shareholder / creditor right protection, market / bank-based financial system, stock / bond market development and growth rate in a country’s gross domestic product ( GDP ) ] in the leverage choice, based on a balanced panel data comprising 11,845 non-financial and non-regulated firms from 42 countries around the world over the period from 1997 to 2001. Application of Seemingly Unrelated Regression ( SUR ) estimation method indicates that : ( a ) firm-specific determinants of leverage vary across countries in constrast to previous studies implicitly assuming fairly similar impact of these factors across developed and developing countries ; and ( b ) country-specific factors directly influencing the firm-specific factors appear to have an indirect effect on leverage. 69)
Mahajan and Tartaroglu ( 2008 ) [ International ] investigate the market timing
hypothesis of capital structure in major industrialized ( G-7 ) countries namely, USA, Canada, France , Germany , United Kingdom , Italy and Japan. Applying Fama-MacBeth regressions on a sample comprised of selected non-financial and non-regulated firms from G-7 countries included in Standard and Poor’s Compustat Global ( Global Vantage ) files over the period from 1993 to 2005 it is observed that : ( a ) leverage is negatively related to the historical market-to-book ratio for all countries ; ( b ) firms from all countries ( except Japan ) rebalance their leverage following equity issuances and the impact of equity market timing attempts on leverage is short lived ; and ( c ) current market-to-book ratio is negatively related to book leverage for Canada and US and to market leverage for all countries. It is thus concluded that the results support the dynamic trade-off theory instead of the market timing hypothesis of capital structure. 70)
Antoniou, Guney and Paudyal ( 2008 ) [ International ] analysed the determinants
of capital structure in capital market-oriented economies ( the United Kingdom and the United States of America ) and bank-oriented economies ( Germany , France and Japan ). The analysis of the results of the application of two-step system-GMM procedure to a panel data sample of 4,854 firms ( 244 French, 479 German, 1,442 Japanese, 1,562 British and 1,127 American ) over the period 1987 to 2000, indicated that :- ( a ) the leverage ratio is : ( i ) positively related to tangibility of assets and firm size , ( ii ) negatively related to profitability , growth opportunities, and share price performance , and ( iii ) affected by the market conditions, legal and financial traditions prevalent in the respective countries ; and ( b ) target leverage ratios exist for firms, with French and Japanese firms respectively
59
being the fastest and the slowest in adjusting their capital structures toward their target level. It was concluded that a firm’s capital structure appears to be greatly influenced by the economic environment and institutional features , relationship between borrower and lender , tax systems, capital market exposure, investor protection levels, and corporate governance practices, in the country in which it operates. 71)
Fattouh, Harris and Scaramozzino ( 2008 ) [ UK ] examined the existence of non-
linearity in the determinants of capital structure based the firm value-maximization model of Fattouh et. al ( 2005 ). Conditional Quantile Regression methodology was applied on an unbalanced panel data of 6416 firm9 - year observations over the period 1988 -1998. The results indicate that debt-equity ratio is : ( a ) positively ( negatively ) related to firm size for low ( high )-leverage firms ; ( b ) positively related to asset tangibility for low-leverage firms ; ( c ) negatively related to profitability and non-debt tax shield for high-leverage firms. It is thus concluded that if non-linear relationship exists between a firm’s capital structure and its determining costs ( plus the non-debt tax shield ), then the results obtained from conditional quantile regression appear to be more informative and statistically more robust than those obtained from ordinary least squares. 72)
Lemmon, Roberts and Zender ( 2008 ) [ USA ] , applying pooled ordinary least
squares, fixed effects, and Generalised Method of Moments ( GMM ) regression methodologies to an unbalanced panel data of non-financial firm-year observations ranging from 54,963 to 225,839 over the period 1965 to 2003, find that an unobserved time-invariant effect accounts for the majority of variation in leverage ratios generating stable capital structures with high or low levered firms maintaining their level of leverage for over two decades. It is also noted that this characteristic of the data - generating process of leverage is robust to firm entry and exit , is present before the initial public offering and is unaccounted for by the determinants of capital structure identified in prior studies, suggesting that factors remaining stable over long horizons tend to determine the variation in capital structures. 73)
Frank and Goyal ( 2009a ) [ USA ] studied publicly traded non-financial American
firms over the period 1950 to 2003 to determine the factors which may be reliably important for capital structure decisions. The most reliable factors explaining market leverage along with their observed effect were as follows : median industry leverage ( positive ) , market - to - book ratio ( negative ) , tangibility ( positive ) , profits ( negative ) , log of assets
9
Non-financial firms listed on the London Stock Exchange.
60
( positive ) , and expected inflation ( positive ) , dividend payout ( negative ).Though more or less similar effects were observed when considering book leverage , firm size, market-tobook ratio, and expected inflation were not found to be reliable determinants of book leverage. It is concluded that some versions of the trade-off theory of capital structure may be said to be reasonably supported by the empirical evidence. 74)
Frank and Goyal ( 2009b ) [ USA ] , based on the Frank and Goyal ( 2008 )’s static
trade-off agency-based model of capital structure and using an unbalanced panel data of 2,26,355 firm-year observations from 1971 to 2006 , show that : ( a ) the evidence on the negative relationship between profits and leverage has been misinterpreted by previous literature due to the wide-spread application of empirical methods which are familiar but inappropriate ; ( b ) an increase in both book value and market value of equity are experienced by more profitable firms ; ( c ) highly profitable ( least profitable ) firms tend to issue ( repay ) debt and repurchase ( issue ) equity ; ( d ) large ( small ) firms tend to be more active in the public debt ( equity ) markets ; ( e ) financing decisions depend on market conditions ( that is , timing of market ) with reduction in employment of external sources of funds emanating from poor market conditions , the firms having small and low profits being affected quite severely. 75)
Psillaki and Daskalakis ( 2009 ) [ Europe ] , with a view to consider whether capital
structure choices are affected by firm-specific factors ( such as, asset structure, size, profitability , risk and growth ) and country-specific factors ( such as, financial development and institutional features ), investigate the determinants of capital structure of small and medium sized enterprises ( SMEs) of four European countries ( namely, Greece, France, Italy and Portugal ) , considering a balanced panel data comprising 1252 Greek firms , 2006 French firms , 320 Italian firms , and 52 Portuguese firms over the period 1997 to 2002. Application of ‘Period Seemingly Unrelated Regression -pooled Estimated Generalized Least Squares’ methodology indicate that : ( a ) capital structure choices are determined by the SMEs in these countries quite similarly ; ( b ) size is significantly positively related to leverage, while asset structure, profitability and risk are significantly negatively related to leverage ; and ( c ) growth does not appear to be a statistically significant determinant of leverage for any of these countries. The similarity in the institutional and financial characteristics of these countries and the commonality of their civil law systems may be attributed for the above results. However, differences in the size of the coefficients in the country regressions due to firm-specific effects result in structural differences. It is thus concluded that
61
differences in capital structure choices of SMEs in these countries may be better explained by firm-specific rather than country-spcific factors. 76)
Rauh and Sufi ( 2010 ) [ USA ] , applying panel data fixed effects regression meth-
odology to a unique sample of 133 fallen angels10 over the period 1996 to 2006, demonstrate that : ( a ) substantial variation in capital structure is unexplained by traditional capital structure studies ignoring heterogeneous nature of debt ; ( b ) firms with low credit quality, in comparison to firms with high credit quality , tend to possess a multi-level capital structure comprising both subordinated non-bank debt with flexible covenants and secured bank debt with stringent covenants ; and ( c ) firms with low credit quality rely on strictly monitored bank revolving credit facilities for liquidity , while firms with high credit quality access to various forms of unrestricted and flexible sources of financing such as medium term notes and shelf registrations. 77)
Galvao and Montes-Rojas ( 2010 ) [ USA ] , with the objective of re-estimating the
partial adjustment model of Flannery and Rangan ( 2006 ) , applied the econometric technique of ‘penalized conditional quantile regression for dynamic panel data with fixed effects’ to an unbalanced panel data sample of 19,140 firm-year observations covering a period ranging from 19 to 30 years within the period 1971-2005 11 . The estimated adjustment speeds lying within the range of 3% to 44% across the quantiles suggested that substantial heterogeneity may probably be present in the speed of adjustment among firms. 78)
Korteweg ( 2010 ) [ USA ] , extending the Modigliani-Miller (1958 ) results through a
new relationship between the market value , systematic risk ( beta ) and net benefits to leverage ( NBL ) of a firm , estimates NBL considering two panel data samples - SIC212 sample comprising 232 firms across 22 industries , and FF 13 sample comprising 290 firms across 30 industries , both over the period 1994 to 2004. The results of pooled ordinary least squares regression indicate that net benefits are increasing ( decreasing ) in leverage for low ( high) - debt firms , implying the existence of an optimal structure. The NBL captured by the median firm is found to be upto 5.5.% of the firm’s market value.
10
Non-financial U.S. public firms that are downgraded from investment grade ( BAA3 or better ) to speculative grade ( BA1 or worse ) by Moody’s Investors Services. 11
firm - years with a negative book value of equity or missing data for long-term debt, debt in current liabilities, or any of the leverage factors, are omitted. 12
Two - digit Standard Industrial Classification ( SIC ).
13
Fama and French ( 1997 ) classification of industries.
62
79)
Fan, Titman and Twite ( 2010 ) [ International ] investigate the influence of institu-
tional environment on capital structure and debt maturity choices by analysing a sample of 36,767 firms from 39 countries over the period 1991-2006. The results of pooled ordinary least squares regression , fixed effects regression and Fama-Macbeth regression indicate that : ( a ) a significant portion of the variation in leverage and debt maturity ratios is explained by a country’s legal and tax system, the level of corruption and the preferences of capital suppliers ; ( b ) firms in countries that are relatively more corrupt tend to empoy more debt ( especially , short-term debt ) and less equity, while firms operating within legal systems providing better protection for financial claimants tend to employ more equity, and comparatively more long-term debt ; ( c ) higher leverage and more longterm debt is associated with the existence of an explicit bankruptcy code and / or deposit insurance ; ( d ) in countries where there is a greater tax gain from leverage , more debts are employed by the firms , while firms in countries with larger government bond markets have lower leverage , suggesting that government bonds tend to replace corporate debt ; ( e ) countries with more extensively defined benefit pension funds have higher debt ratios and longer debt maturities, whereas those with more extensively defined contribution fund activities have lower debt ratios ; ( f ) debt ratios are lower in countries that limit the bond holdings of pension funds ; and ( g ) in insurance industry, a significant association between financing choices and firm size exist. 80)
Sheik and Wang ( 2011 ) [ Pakistan ] conduct a study to explore the factors affecting
the capital structure of manufacturing firms and to investigate whether convincing explanations for capital structure decisions of Pakstani firms are provided by the capital structure models derived from Western settings. The investigation is performed using panel data procedures for a sample of 160 firms listed on the Karachi Stock Exchange during 2003-2007. The results indicate that debt ratio appears to be : ( a ) significantly negatively related to profitability, liquidity, earnings volatility, and tangibility ; ( b ) significantly positively related to firm size ; and ( c ) not significantly related to non-debt tax shields and growth opportunities. These findings , being consistent with the predictions of the trade-off theory, pecking order theory, and agency cost theory , suggest that the financing behaviour of Pakistani firms may be explained by capital structure models derived from Western settings.
63
81)
Kayo and Kimura ( 2011 ) [ International ] , based on a sample of 17,061 non-
financial companies from 40 countries over the period 1997 -2007 , analyze the influence of the various determinants of capital structure at three hierarchical levels : ( a ) firm-level ( size, tangibility, profitability, growth opportunities and bankruptcy risk ) ; ( b ) industry-level ( munificence
14
, dynamism
15
and industry concentration ) ; and ( c ) country-level ( stock
market development, bond market development, financial system and GDP growth ). Initial application of hierarchical linear modeling for assessing the relative importance of the above -mentioned levels indicate that 78% of firm leverage is explained by time and firm levels. Subsequent application of random intercepts and random coefficients model for analyzing the direct and indirect influences of firm-, industry-and country-level determinants on firm leverage lead to the observations that variables at industry-and country-levels appear to have indirect influences on firm-specific determinants of leverage and several structural differences may be said to exist in the financial behaviour of firms in developed and emerging countries. 82)
Icke and Ivgen ( 2011 ) [ Turkey ] examine the firm-specific factors which are
influen-tial on capital structure decisions of 212 industrial firms listed in Istanbul Stock Exchange over the period from 2004 to 2009. The results of panel data regression analysis show that : ( a ) firm size , profitability and liquidity are statistically significant and negatively related to leverage ratios ; and ( b ) growth is statistically significant and positively related to leverage ratios ; thus providing support to the predictions of the Pecking Order Theory. 83)
Fama and French ( 2012 ) [ USA ] , considering a sample of 6113 non-financial firms
over the period from 1963 to 2009 , test the predictions of the trade-off model , the pecking order model , and market conditions model using three pairs of Fama and MacBeth (1973) type cross-sectional regressions formed by segregating : ( i ) new outside financing between issue of equity share and debt ; ( ii ) debt financing between short-term and long-term debt ; and ( iii ) equity financing between issue of share and retained earnings. It is observed that : ( a ) the performance of pecking order is satisfactory till equity issue become common in the early 1980s ; ( b ) the target adjustment prediction of the trade-off model, and the prediction of equity financing response to market valuations by the market conditions Munificence refers to the environment’s capacity to support a sustained growth. Environments with high munificence have abundant resources, low levels of competition and hence high profitability ( Dess and Beard, 1984 ). 14
15
Environmental dynamism reflects the degree of instability or non - predictable change of a given industry ( ibid.).
64
model have statistically significant but rather minor effects on the segregation of new outside financing between equity issues and debt ; ( c ) in case of smaller ( larger ) firms targets for short-term debt appear to be have strong ( weak or non-existent ) effects on the segregation of debt financing between short-term and long-term debt ; ( d ) the predictions of the pecking order and the market conditions models about the segregation of equity financing between share issues and retained earnings are plagued by sticky dividends. 84)
Lim ( 2012 ) [ China ] examines the determinants of capital structure of financial
service firms based on a cross-sectional sample for 36 A-share firms in financial sector listed on the Shanghai and Shenzhen Stock Exchanges over the period 2005 to 2009. The results of the application of multiple regression and stepwise regression analyses indicate that : ( a ) profitability , firm size , non-debt tax shields , earnings volatility and non-circulating shares are significant determinants of leverage in financial sector ; ( b ) capital structure choice is affected by institutional characteristic ; and ( c ) large state ownerships appear to influence capital structure decision even though the determinants of capital structure of financial firms are reasonably similar to those in other industries. The above observations imply that the trade-off theory has limited power for explaining the capital structure decisions of listed Chinese financial companies which appear to follow a different pecking order preferring external financing to internal sources. 85)
Caskey, Hughes and Liu ( 2012 ) [ USA ] , while considering the dynamic nature of
capital structure and potentially delayed market reactions , examine the cross-sectional relation between leverage and future returns based on an unbalanced panel data sample of non-financial firm-year observations ranging from 60,779 to 72,325 over the period of 27 years from 1980 to 2006. The results indicate that significant information about the firm’s future asset growth and probability of financial distress is provided by excess leverage. However, the link between excess leverage and future fundamentals may not be entirely understood by the market. It is observed that future returns are positively predicted by Graham’s ( 2000 ) kink ( a measure of negative excess leverage ) , a phenomenon primarily driven by the market’s delayed reaction to the information in the kink about future asset growth and financial distress. Since excess leverage and leverage are positively correlated, leverage is negatively correlated with future returns in the absence of controls for excess leverage. Simulataneous consideration of both leverage and excess leverage renders future returns to be associated with excess leverage, but not debt levels. The properties of excess leverage are exhibited by Graham’s kink measure in the context of a partial adjustment model. The relationship between the kink and future returns is explained neither by the
65
omission of distress costs nor by a distress risk factor. High ( low ) kink firms are found to have lower ( higher ) probability of financial distress, and lower ( higher) exposure to a distress risk factor challenging the explanation provided by distress cost. Cross-sectional analysis of the relation between the kink and future returns does not seem to support other potential explanations related to stock options as a tax shield, foreign repatriation taxes, and intensity of Research and Development. 86)
Kouki and Said ( 2012 ) [ France ] examine the determinants of corporate capital
structure choice emphasizing the role of capital market imperfections through the tradeoff, pecking order and market timing theories to explain corporate leverage. The analysis is conducted on a sample of 244 French listed companies over the period from 1997 to 2007. The empirical results indicate that : ( a ) trade-off hypothesis and the financing deficit variable of the pecking order hypothesis are complementary ; ( b ) market conditions do not appear to have any significant and meaningful impact on debt ratio without any confirmation of simple or extendend form of market timing effect ; and ( c ) the lagged leverage ratio is relevant in all tests thus confirming the existence of a dynamic process of adjustment to a target level. 87)
Bahng and Jeong ( 2012 ) [ Australia ] investigated ( i ) the possible non-linear
effects in the determinants of capital structures, and ( ii ) the non-linear adjustment behaviour of cross-sectional debt ratios, applying the conditional quantile regression methodology to an unbalanced panel data sample comprising 13,813 firm-years observations over the period 1991 to 2007 16. The results confirmed the existence of non-linear effects between debt ratio and the explanatory variables of firm size and profitability. However, such nonlinear effects did not appear to arise in case of the explanatory variables of asset tangibility and non-debt tax shield. The non-linear speeds of cross-sectional adjustment to target leverage were also confirmed to exist. 88)
Devos , Dhillon , Jagannathan and Krishnamurthy ( 2012 ) [ USA ] examine the
reasons for firms having no debt in their capital structure , based on a sample of public, non-financial firms with zero-debt ( that is, firms remaining unlevered for at least three consecutive years ) during the period 1990 to 2004. The empirical results indicate that zero-debt firms : ( a ) do not appear to have weaker internal or external governance
16
Pooled conditional quantile regression was applied instead of fixed effects- or random effects- conditional quantile regression.
66
mechanisms and the shocks to their entrenchment ( such as takeover threats or the emergence of activist blockholders ) do not precede the debt initiation decisions of these firms, thus leading to the rejection of the hypothesis that entrenched managers undertake zeroleverage policies in order to avoid the disciplinary pressures of debt ; and ( b ) are small, young , conserve cash from cash-flow, are more likely to lease their assets , face stricter covenants and higher all-inclusive costs than comparable control firms when they have access to a line of credit, lose market share in economic downturns and do not voluntarily stockpile debt capacity , thus lending support to the hypothesis that these firms are financially constrained. 89)
Dong , Loncarski , Horst and Veld ( 2012 ) [ Canada ] study market timing and
pecking order in a sample of debt and equity issues and share repurchases of Canadian firms [ 227 corporate debt issues ( made by 64 different companies ), 1,271 corporate equity issues ( made by 664 different companies ), and 1,071 intended share repurchases ( made by 447 different companies ) ] from 1998 to 2007. The findings indicate : ( a ) the existence of an interaction between market timing and pecking order effects ; ( b ) that financially unconstrained firms issue ( repurchase ) equity when their shares are overvalued ( undervalued ) ; ( c ) that overvalued issuers of equity earn lower post announcement long-run returns ; and ( d ) undervalued issuers of equity prefer debt to equity financing. 90)
Atiyet ( 2012 ) [ France ] compares the explanatory power of the Pecking Order
Theory ( POT ) and the Static Trade-off theory ( STT ) based on the studies made by ShyamSunder and Myers ( 1999 ) and Frank and Goyal ( 2003 ) and consideration a panel data sample of 88 firms belonging to SBF 250 index over a period from 1999 to 2005. The empirical results show that pecking order theory favour estimation of the empirical models explaining the financial structure . 91)
Lemma and Nagesh ( 2013 ) [ Africa ] conducted a study to investigate the impact
of firm and industry characteristics and institutional and macroeconomic factors on a firm’s capital structure decision in the context of nine African countries based on sample of 986 non-financial firms over a period of 10 years from 1999 to 2008. The study utilized Generalized Method of Moments ( GMM ) and Seemingly Unrelated Regression ( SUR ) techniques which are robust to data heterogeneity and endogeneity problems. The results indicated that : ( a ) leverage is directly affected by firm size while it is inversely related with profitability ; ( b ) the effect of asset tangibility, non-debt tax shield and dividend payout on leverage is dependent on how the latter is defined ; ( c ) inter-industry variations affect capital structure decision of African firms ; ( d ) income level of host countries
67
moderates the influence of firm-specific factors on capital structure decisions ; and ( e ) legal and financial institutions and macro-economic conditions do matter in the capital structure decisions of African firms. These findings signify the role that probability of bankruptcy, agency costs, transaction costs, tax issues, information asymmetry problems, access to finance and market timing play in capital structure decisions of firms in Africa. 92)
Wellalage and Locke ( 2013 ) [ New Zealand ] explored the impact of firm-specific
factors ( namely, liquidity, tangibility, growth, risk, size and non-debt tax shield ), industryspecific factors ( industry dummy variables ) and corporate governance variables ( namely, foreign share ownership, managerial ownership and non-executive directors on the board ) on the capital structure of Kiwi firms, based on a balanced panel data sample of 40 nonfinancial NZX50 firms listed on the New Zealand Stock Exchange ( NZSE ) over a period of 7 years from 2003 to 2009. Conditional quantile regression methodology was applied to the pooled data 17. The results indicate that firm-specific and industry -specific characteristics rather than corporate governance variables play a significant role in determining leverage levels. The results imply that finance policies need to vary across firm type and firm characteristics, and should match with the different borrowing requirements of listed firms. 93)
Oliveira , Tabak , Resende and Cajueiro ( 2013 ) [ Brazil ] investigate the firm-
specific determinants of capital structure ( namely, size, tangibility, profitability, growth opportunity and risk ) , applying cross-sectional conditional quantile regression methodology to a sample of 394 publicly traded companies over the period 2000 to 2009. The empirical results confirm significant changes in the effects of the explanatory variables depending on the quantile , which may be explained by bankruptcy and agency costs associated to the quantum of debt in the capital structure of firms , relative to each quantile. Conditioning on the determinant , nature of debt and the quantile analysed , the predictions of the tradeoff and pecking order theories are then examined . The resultant observations in respect of size and profitability indicate the suitable application of the pecking order theory with increase in quantile. 94)
Said ( 2013 ) [ France ] , based a sample of French companies listed on SBF 250
index and observed over the period 1997-2007 , examines the determinants of capital structure by analyzing how ownership structure affects debt policy as per the propositions of the agency theory. The empirical observations indicate that : ( a ) a non-linear relationship
17
Pooled conditional quantile regression was applied instead of fixed effects- or random effects- conditional quantile regression which may be said to be panel data - conditional quantile regession in the truest sense.
68
exists between managerial ownership and capital structure ; ( b ) in a static framework, disciplinary role is not executed by outside shareholders for effectively controlling managerial behaviour , and outside shareholdings does not appear to significantly affect debtequity ratio for high levels of managerial ownership ; ( c ) in a dynamic framework, the effect of external investors on debt-equity ratio is significant and negative but the effect of management ownership on debt-equity ratio is not significant. 95)
Bayrakdaroğlu , Ege and Yazıcı ( 2013 ) [ Turkey ] conducted a study to determine
whether capital structure determinants in the emerging market of Turkey support the capital structure theories which have been developed to explain the capital structure choice in developed economies. The balanced panel data sample comprised of 242 companies listed on the İstanbul Stock Exchange ( ISE ) over the period 2000-2009. The dependent variable ( s ) included three different measures of leverage ( short-term, long-term and total debt ratios, all at market values ). The explanatory variables included firm-specific factors ( namely , profitability , tangibility of assets , size, growth opportunities , and non-debt tax shields ) and macro-economic factors ( economic development, inflation and taxes ). Fixed effects panel data regression analysis was applied. From the results it was concluded that trade-off theory was less successful than the pecking order hypothesis in explaining the capital structure of the sample of Turkish companies. 96)
Strebulaeva and Yang ( 2013 ) [ USA ] present the puzzling evidence that, from 1962
to 2009, on an average , 10.2% of large public non-financial US firms have zero debt and almost 22% have less than 5% book leverage ratio , noting that : ( a ) zero-leverage behavior is a persistent phenomenon ; ( b ) dividend-paying zero-leverage firms pay substantially higher dividends, are more profitable, pay higher taxes, issue less equity, and have higher cash balances than control firms chosen by industry and size ; ( c ) firms with higher Chief Executive Officer ( CEO ) ownership and longer CEO tenure are more likely to have zero debt, especially if boards are smaller and less independent ; and ( d ) family firms are also more likely to be zero-levered. 97)
Chung, Na and Smith ( 2013 ) [ USA ] test the trade-off hypothesis against the
pecking order hypothesis for capital structure using an unbalanced panel data sample of 2015 oil exploration firm - year observations over the period 1970 to 2007. Probit and Multinomial Probit regression analyses are applied to three different functional forms : ( a ) quadratic specifications to current and lagged capital structure ; ( b ) a linear splined model that allows capital structures above and below the mean of the sample under consideration to have different slopes ; and ( c ) linear splined models where the relationships
69
of capital structure to outcomes are assumed to be linear over a broad range , but allowed to be different in the upper and lower deciles and quintiles. The results indicate no significant evidence of an economically important optimal capital structure. The issue of low - leveraged firms operating and surviving for a long period of time without being targeted for acquisition , was also addressed by the study. It is observed that firms facing attractive growth opportunities appear to increase leverage or firms raise capital by borrowing when compelled to do so or when poor operating performances reduce equity value , thus lending support to the pecking-order hypothesis. 98)
Jõeveer ( 2013 ) [ European Countries ] explores the significance of firm-specific
factors ( namely , size, profitability, tangibility and age ) and institutional and macroeconomic factors ( namely , inflation, GDP growth, corporate income tax rate, share of foreign owned banks assets to total banking sector assets, share of the three biggest banks assets to total banking sector assets, country credit rating, corruption perception index, shareholder rights protection index and minority shareholder rights protection index ) in explaining variation in leverage using an unbalanced panel data sample of listed and unlisted firms from nine Eastern European countries ( namely , Bulgaria, the Czech Republic, Estonia, Hungary, Latvia, Lithuania, Poland, Romania, and Slovakia ) over a period of eight years from 1995 to 2002. Applications of ANOVA and OLS regression techniques result in the observations that : ( a ) in case of small unlisted companies , the country-specific factors explain most of the variation in leverage , while the firm-specific factors appear to be the major determinants of variation in leverage in case of listed and large unlisted companies ; and ( b ) macroeconomic and institutional factors explain around half of the variation in leverage related to country factors , while unmeasurable institutional differences explain the remainder. 99)
Mokhova and Zinecker ( 2014 ) [ European Countries ] , based on a sample of
manufacturing firms from seven European countries ( namely, Germany, France Czech Republic, Slovakia, Hungary, Poland and Greece ) for the period 2006 to 2010 , examined the influence of macroeconomic factors ( divided into two groups - fiscal policy and monetary policy ) on capital structure choice. The results from the application of correlation and regression techniques indicated that capital structure decision - making process and source of financing were significantly influenced by macroeconomic factors.
70
100) Wellalage and Locke ( 2014 ) [ Sri Lanka ] explore the relationship between capital structure and its firm-specific determinants based on a sample of 158 firms listed on the Colombo Stock Exchange over the period from 2005 to 2010. Application of conditional quantile regression to the pooled data 18 showed that significant differences appear to exist between firms in different quantiles of leverage with the signs of the explanatory variables changing with the quantiles. It is concluded that the capital structure policy of a firm should be responsive to the firm-specific factors and should match with the different borrowing requirements of listed firms. 101) Berzkalne and Zelgalve ( 2014 ) [ Baltic ] , based on a sample of 75 non-financial firms from the Baltic states ( namely, Latvia , Estonia and Lithuania ) over the period from 1998 to 2011 , conducted a study to test and evaluate the performance of pecking order and trade-off theories of capital structure considering Shyam-Sunder and Myers ( 1999 ) model and applying panel data regression methodology. The empirical results show that : ( a ) listed companies in Latvia , compared to the other countries , appeared to have the lowest debt ratio ; and ( b ) companies adjust their debt levels according to target debt , thus lending no support to the pecking order theory. 102) Acedo‐Ramírez and Ruiz‐Cabestre ( 2014 ) [ Europe ] analyze the influence of country-specific differences on capital structure choice indirectly through firm-specific variables ( namely , firm size , non-debt tax shield , effective tax rate , financial distress costs , investment and free cash flow ) , applying system GMM ( Generalized Method of Moments ) technique to an unbalanced panel data sample 19 of non-financial listed companies from five European countries ( namely, France, Germany, Italy, Spain and the United Kingdom ) during the period 1998-2008. Since capital structure decisions may be influenced differently through firm-specific variables owing to the different financial systems ( bankoriented or market-oriented ) of European economies , the determinants of capital structure for each country are initially analysed separately leading to the subsequent examination of the relevancy of the observed differences in capital structure decisions between the United Kingdom and the continental European countries. The results show that : ( a ) the type of financial systems of the countries ( bank-oriented and market-oriented ) and the firm-specific factors account for the major portion of the substantial differences in the
18
Pooled conditional quantile regression was applied instead of fixed effects- or random effects- conditional quantile regression which may be said to be panel data - conditional quantile regession in the truest sense. 19
The unbalanced panel comprises 242 French ( 1,799 observations ), 164 German ( 1,193 observations ), 85 Italian ( 640 observations ), 60 Spanish ( 465 observations ) and 337 British ( 2,431 observations) companies.
71
capital structure choices of firms across five major European countries ; and ( b ) the differences between the capital structure choices of firms in the United Kingdom ( marketoriented economy ) and firms in continental European countries ( bank-oriented economies) are relevant. 103) Arshanapalli and Nelson ( 2014 ) [ USA ] examine the capital structure decision of 3,432 US companies in the years 2006 and 2011 applying cross-sectional conditional quantile regression methodology to explore the predictions of the trade-off and pecking order models. The results indicate evidence of heterogeneity in the determinants of capital structure, the data being more consistent with the trade-off theory than the pecking order theory in 2006 but only economic conditions mattering in 2011. 104) Chang, Chen and Liao ( 2014 ) [ China ] , applying the same methodology of Frank and Goyal ( 2009 ) to a panel data sample of Chinese A-share listed companies ( excluding financial firms and firms listed in the growth enterprise market ) with 13,107 firm-year observations between 1998 and 2009 , identify seven core variables ( namely, profitability, industry leverage, asset growth, tangibility , firm size and state control ) as reliably important factors that are statistically significant and have coefficients of consistent signs across various models. 36% of the variation in book leverage of firms are explained by the seven core variables , whereas 1.4% of the variation are accounted for by the remaining 17 variables. 105) Elsas, Flannery, and Garfinkel ( 2014 ) [ USA ] evaluate the determinants of leverage by analysing how firms paid for 2,073 very large investments between 1989 and 2006, thus complementing existing empirical work on capital structure which typically estimates regression models of leverage for a broad set of firms. It is opined that security issuances should provide information about managers’ attitudes toward leverage as large investments are mostly externally financed. The findings of the study suggest that issued securities move firms toward target debt ratios and that firms also tend to issue more equity following a share price run-up, consistent with both the trade-off hypothesis and managerial efforts to time market sentiment thus extending little support for the standard pecking order hypothesis. 106) Sánchez-Vidal ( 2014 ) [ Spain ] applies quantile regression approach to analyze the leverage determinants for a large sample of companies for the period 2001 to 2011 depending on their level of indebtedness. The findings reveal that : ( a ) coefficients of the independent variables change sign, value and level of significance at different quantiles of the leverage distribution ; ( b ) many factors are no longer significant for highly-leveraged
72
companies ; and ( c ) cash flow variable is crucial if the companies would like to decrease their debt levels. 107) Frank and Goyal ( 2015 ) [ USA ] show that the inverse relation between leverage and profitability, which is regarded as a serious defect of the trade-off theory, is not a defect with the theory but with the use of a leverage ratio in which profitability affects both the numerator and the denominator. Profitability directly increases the value of equity. The predicted offsetting actions of issuing debt and repurchasing equity when profitability rises, and retiring debt and issuing equity when profitability falls , are not taken by the firms. Consistent with variable transactions costs, the adjustment is not generally sufficient to fully undo the profitability shocks. Consequently, the leverage ratio, on average, falls as profitability rises. The study also applies conditional quantile regression to pooled data, apart from OLS regression. 108) Elsas and Florysiak ( 2015 ) [ USA ] propose and apply a new unbiased estimator DPF ( dynamic panel data with a fractional dependent variable ) for estimating the speed of adjustment towards target leverage, based on a sample of 169,787 firm-years observations ( 16,357 firms over the period 1965 to 2009 ). The results suggest that the DPF estimator performs better than the Blundell-Bond ( 1998 ) GMM estimator, the long difference estimator, or the LSDVC estimator. 109) Arce , Cook and Kieschnick ( 2015 ) [ USA ] introduced a function of higher polynomial forms such as terms of second-order or third-order when explaining capital structure decisions. Based on an unbalanced panel data of non-financial firm-year observations ranging from 137,854 to 188,938 over the period from 1965 to 2003 and using the linear model of Lemmon et al. ( 2008 ), it was tested whether the mis-specifications of Lemmon et al. ( 2008 ) model were concerned with non-linear models. The results suggested higherorder terms were significant. 110) Öztekin ( 2015 ) [ International ] examines the determinants of capital structure applying dynamic panel data approach on a large sample of 15,177 firms from 37 countries over the period 1991 to 2006 totalling 101,264 firm-years observations. The findings suggest that : ( a ) the reliable determinants for leverage are firm size, tangibility, industry leverage, profits, and inflation, ( b ) the quality of the countries’ institutions affects leverage and the adjustment speed toward target leverage in significant ways, and ( c ) high quality institutions lead to faster leverage adjustments, whereas laws and traditions that safeguard debt holders relative to stockholders ( for example, more effective bankruptcy procedures and stronger creditor protection ) lead to higher leverage.
73
111) Su ( 2015 ) [ China ] , using firm-level panel data of Chinese listed companies , analyse the effects of the pyramid of inner ownership structure on capital structure, and the differences in those effects between regions with different institutional environments. The results indicate that the layers of pyramid structure have a significantly positive effect on capital structure , with longer layers of the pyramid structure having stronger ‘leverage effect’ as well as the ultimate owner’s motivation to expand debt financing. The chains within a pyramid structure , however , are not found to have any significant effect on capital structure. In the regions having better institutional environment , the effect of the layers of pyramid structure on capital structure becomes smaller compared to the regions having poor institutional environment. 112) Gwatidzo, Ntuli and Milo ( 2016 ) [ South Africa ] apply panel data ( fixed effects ) -conditional quantile regression approach proposed by Canay ( 2011 ) on a sample of 239 listed firms over the period 1996-2010 to investigate the non-linear effects of capital structure determinants on the conditional distribution of leverage. The results suggest that the impact of leverage determinants does not vary with leverage with the exception of asset tangibility and age, whose effect increased with leverage. It is concluded that for the case of South Africa, studies that estimate the impact of leverage at the mean are still valid and appropriate. 113) Ferrarini, Hinojales, and Scaramozzino ( 2017 ) [ China ] assess the financial fragility of the Chinese economy by looking at risk factors in the nonfinancial sector. The results of the application of Quantile regressions [ Quantile Regression for Panel Data ( QRPD ) introduced by Powell ( 2014 ) and Simultaneous Quantile Regression ( SQR ) for crosssectional data over each time period ] to a sample of 983 Chinese listed companies over the period 2009 to 2015 , indicate increasing sensitivity of corporate leverage to some of its core determinants over time. Specifically , profitability is found to increasingly act as a constraint on corporate leverage during recent years , which may contribute to further increases in corporate leverage over time considering the low profitability across the Chinese non-financial corporate sector at the present time.
74
3.3 (1)
Indian Studies Sarma and Rao ( 1969 ) tested the M-M model using cross-sectional analysis for 30
engineering companies for three years-1962, 1964 and 1965. Applying Two Stage Least Squares ( TSLS ) regression on the data, they observed that the magnitude of coefficient of the leverage variable was greater than that of the tax rate variable, thus concluding that debt, apart from its tax advantage, also affects the cost of capital in agreement of the traditional approach and contrary to M-M Hypothesis. (2)
Chakraborty ( 1977 ) , considering a sample of 22 firms belonging to various indus-
tries from the corporate private sector, conducted a comprehensive mainly to : ( a ) investigate the determinants of capital structure with the dependent variable being debtequity ratio and the independent variables being age, total assets, retained earnings, profitability and capital intensity ; and ( b ) calculate the cost of capital and indirectly test the M-M Hypothesis. The major findings of the study included : ( a ) age, retained earnings, and profitability were negatively correlated with the debt equity ratio, while total assets and capital intensity were directly related to it ; ( b ) cost of capital was almost invariant to the debt-equity ratios ; and ( c ) the average cost of capital for firms in the consumer goods industry was the highest while it was the lowest for the belonging to the intermediate goods industry, primarily because of low debt content in the capital structure in case of former category of industry as compared to the later. (3)
Bhat ( 1980 ) conducted an empirical study to examine the impact of seven firm-
specific determinants ( firm size, business risk, growth, profitability, operating leverage, dividend payout and debt service capacity ) of financial leverage on debt-equity ratio. Applying correlation and multiple regression techniques to a sample of 63 companies of the engineering industry over a period of six years from 1973 to 1978, the author concluded that business risk, profitability, debt service capacity and dividend payout appeared to be significant determinants of capital structure. (4)
Pandey ( 1981 ) attempted to analyse the empirical relationship between capital
structure and cost of capital based on a cross-sectional sample of 131 firms from four industries { namely, cotton ( 47 firms ), chemicals ( 32 firms ), engineering ( 32 firms ) and electricity generation ( 20 firms ) } for three years, 1968, 1969 and 1970. Two measures of leverage ( one measure treating preference share capital as part of debt and the other measure treating it as part of equity ) were used and the following explanatory variables , namely , firm size , growth, operating risk, dividend payout and liquidity were incorporated.
75
When the average cost of capital was regressed against leverage, holding the other variables constant, the results ( that the coefficients of both the leverage variables were significantly negative for all the industries for all the years as well as for the pooled data ) supported the traditional approach. A modified model was also used to test the hypothesis whether or not debt financing is advantageous ( the cost of capital will decrease with increase in leverage ) in the absence of its tax effect. When regressions were run for the pooled data of three industries ( as electricity industry was excluded for the lack of sufficient number of observations ), the coefficients of both the leverage variables were found to be significantly negative, thus strengthening the support for the traditional approach. (5)
Venkatesan ( 1983 ) investigated the determinants of capital structure by analyzing
the relationship between seven variables ( namely, size, operating leverage, debt coverage, cash flow coverage, business risk, growth, and industry categorization ) and the financial structure of firms, through the application of correlation and multiple regression analyses on a cross-sectional sample of 66 firms from four industries ( mining, paper, chemical and steel ) over a period of four years from 1977 to 1980. The relationship between industry categorization and financial leverage was examined by the grouping the firms in various leverage classes but no significant and conclusive relationship was found. The impact of the other independent variables on financial structure was examined in two sample classifications , namely , intra-industry and inter-industry. In the intra-industry model, debt coverage ratios were found to be the only significant determinants of financial leverage in respect of all the industries except for steel industry. In the inter-industry model, significant relationship between financial leverage and the explanatory variables ( except growth ratio ) were observed in case of low-levered firms , while no significant common determinants were found in case of medium-levered and high-levered firms. (6)
Pandey ( 1984 ) conducted a questionnaire survey to ascertain the corporate managers’
conceptual understanding of cost of capital and optimum capital structure and their attitude towards the use of leverage. Managers of 30 companies responded. It was found that 87% of the respondents regarded ordinary shares as most expensive source of capital, whereas 77% regarded long-term debt as the cheapest source. The low cost of debt due to tax deductibility of interest and the complicated procedures for raising equity capital were found to be responsible for the strong preference for borrowing.
76
(7)
Pandey ( 1985 ) examined the impact of firm size, industry, profitability and growth
on leverage based on a sample of 743 companies classified into 18 industrial groups for the period 1973-74 to 1980-81. It was observed that : ( a ) about 72 to 80 percentage of the assets of the sample firms were financed by external debt and current liabilities, and trade credit was employed as much as bank borrowings ; ( b ) after 1973-74, the levels of leverage for all the industries exhibited a noticeable increase ; ( c ) the classification of leverage percentage by the type of industry was not indicative any systematic and significant patterns and, moreover, the trends and volatilities associated with the leverage percentages was not supportive of the belief that the degree of leverage was affected by the type of industry ; ( d ) some large-sized firms a large number of small-sized firms was found to employ high level of debt, thus invalidating the commentary of any conclusive statement on the relationship between size and the degree of leverage ; and ( e ) majority of the profitability and growth groups of firms were concentrated within narrow bands of leverage, thus negating the presence of any definite structural relationship between the degree of leverage on the one hand and profitability and growth on the other, even though there was improvement in the degree of leverage, profitability and growth over the period under study. (8)
Rao ( 1989 ) examined the correlation between debt-equity ratio on the one hand and
age, size, retained earnings and profitability on the other, based on a sample of 30 chemical companies over the period 1970-71 to 1981-82. It was observed that : ( a ) significant negative correlation existed between age and debt equity ratio with the indication that possibly younger age of chemical companies tended to be associated with higher debtequity ratio ; ( b ) negative correlation between retained earnings and the debt-equity ratio indicated that a company with higher volume of retained earnings had low debt-equity ratio ; ( c ) in case of high debt-equity ratio, the profitability declined due to large payment of interest ; ( d ) positive correlation existed between debt-equity ratio and the size measured in terms of total assets, and net assets. (9)
Babu and Jain ( 1998 ) , in an attempt to gather evidence of the pecking order
hypothesis in India , conducted a survey through questionnaires sent by mail to about 1,300 selected non-government, public limited companies listed on the Bombay Stock Exchange. The basis of selection was on a random sampling method using the official directory of the Bombay Stock Exchange. The response received was only 91 companies. The survey indicated four goals pursued by corporate firms in India which includes maximizing equity per share, maximizing aggregate earnings, maximizing net worth per
77
share and maximizing rate of return on investment in assets. On examining the pecking order hypothesis, the survey asked the financial managers to rank preferences in financing for all the 9 available long-term funds. The results indicated that retained earnings had been the most preferred source of finance in India ( with a score of 8.25 ) while debt was ranked the second most important source of financing ( with a score of 6.66 ). The preference shares had been the least preferred one with a score of only 2.54. From this survey study, there is some practical evidence of the pecking order hypothesis utilized by finance managers in India in terms of order of preference in financing. (10) Kakani and Reddy ( 1999 ) investigated the determinants of the capital structure based on a sample of 100 manufacturing firms listed on the Bombay Stock Exchange during the periods 1984 -89 ( pre-liberalization period ) and 1991-95 ( post liberalization period ). Three measures of leverage ( dependent variable ) were used : total debt ratio, shortterm debt ratio and long-term debt ratio. The eleven explanatory variables included : asset structure, capital intensity, non-debt tax shields, growth, uniqueness, size, profitability, earnings volatility, net exports, regulation and corporate strategy. The econometric methods of Kruskal-Wallis one way analysis of variance by ranks, correlation analysis, linear and stepwise multiple regression analyses were applied. Profitability ( negatively related to all the measures of leverage ), capital intensity, earnings volatility and non-debt tax shields ( negatively related to short-term and total debt ratios ) and uniqueness ( positively related to short-term and total debt ) were found to be statistically significant ; whereas firm size and corporate strategy were found to be insignificant. Regulated firms and growth-oriented firms were found to have more total and long-term debt during the pre-liberalization period. In the liberalized period, net exports seemed to have grown in importance in determining total and long-term debt. (11) Majumdar and Chhibber ( 1999 ) examined the relationship between the levels of debt in the capital structure and performance ( or profitability ) based on a cross-sectional sample of 1043 firms over the period 1988 to 1994. The application of weighted least squares regression method revealed significantly negative relationship between leverage and profitability , which may be accounted for by the structure of capital markets in India with short-term and long-term lending institutions being government owned. It is also asserted that corporate governance mechanisms , applicable to western economies , will not work in the Indian context unless the supply of loan capital is privatized.
78
(12) Bhaduri ( 2002 ) attempted to study the capital structure choice of developing countries through a case study of the Indian corporate sector. The study was based on a sample of 363 manufacturing firms across nine broad industries ( Heavy , Drugs and Pharmaceutical, Chemical, Cement, Textile, Automobile Ancillaries, Paper, Electrical Machinery, Food, Sugar and Beverage ) for the period 1989-90 to 1994 -95. Three measures of the dependent variable ( leverage ) { namely , total debt ratio , short-term debt ratio and longterm debt ratio } were used. The independent variables included cash flow deficit , asset structure, size , non-debt tax shields , profitability , growth , age , financial distress ( volatility of earnings and probability of financial distress ), signaling ( dividend payout and business group dummy) uniqueness and industry dummies. The study developed a two-stage model of optimal capital structure. In the first stage, it used a partial adjustment model to analyse the lagged adjustment towards the optimal capital structure in the presence of restructuring costs. The second stage of the model utilized a factor analytic technique to explain observed variations in optimal capital structure across firms. The evidence suggested that the optimal capital structure choice could be influenced by factors such as growth, cash flow, size, uniqueness and industry characteristics. The results also suggested the possibility of costly restructuring by the Indian firms and a differential cost of adjustment for longterm and short-term borrowing. The speed of adjustment towards optimal capital structure is higher for short-term borrowing than for long-term debt. (13) Guha-Khasnobis and Bhaduri ( 2002 ) attempted to provide some insights into the dynamics of the optimal capital structure choice of the Indian corporate sector based on a balanced panel sample of 697 manufacturing firms over the period 1990-1998. The econometric applications included Generalized Methods of Moments ( GMM ) estimation of a dynamic panel data model and Cohorts Analysis. The results suggested a possibility of : ( a ) costly restructuring for the Indian firms and differential costs of adjustment for long term and short-term borrowing , the speed of adjustment towards the optimal capital structure being much higher for the short term borrowing than the long term one ; and ( b ) optimal capital structure being strongly influenced by factor the factors like size, asset structure, profitability and short-term financial distress cost. (14) Bhole and Mahakud ( 2004 ) analyzed the trends in the corporate capital structure in respect of Indain public limited and private limited companies based on panel data model. Period analysis was also conducted to examine the impact of liberalization on the determinants of capital structure. It was found that leverage ratios have increased significantly during 1966-2000. Also apparently, dependence on debt was more in the case of
79
public limited companies than private limited companies unlike in countries like USA, UK and Australia. The pecking order of funds in India was found to be borrowings followed by trade dues , external equity and reserves and surplus. It was also observed that variables like cost of borrowing, cost of equity, size of firm, collateral value of assets, liquidity and non-debt tax shields appear to be the major determinants of corporate capital structure in India. (15) Sen and Pattanayak ( 2005 ) examined the issue of capital structure choice of Indian banking sector by applying exploratory factor analysis technique to a sample of 82 Indian Banks comprising public sector banks, private and foreign banks for the period from 1996 to 2002. The results of the study suggested that liquidity, size, efficiency and growth, quality of assets, profitability and service diversification seemed to the most critical factors influencing the capital structure of the Indian banking companies. (16) Sahoo and Omkarnath ( 2005 ) conducted a study to : ( a ) analyze the capital structure of the Indian corporate sector ; ( b ) examine whether any shift has taken place in the financing pattern of the Indian corporate sector after the implementation of financial liberalization in the early 1990s ; and ( c ) discuss the factors determining the debt-equity choice of Indian private sector firms. The above-mentioned issues were empirically analyzed with Multiple Regression methodology based on a sample of Large Public Limited Companies ( whose number varied from time to time ) for the period 1980-81 to 2003-04. The study found out that profitability and asset structure were the most significant factors determining capital structure choice. (17) Narender and Sharma ( 2006 ) examine the capital structure policies adopted by the profit-making central Public Enterprises ( PEs ) based on a sample of 47 firms for the period from 1994-1995 to 2004-2005. The results , based on fixed effects panel data methodology , suggest : ( a ) tangibility of assets plays a significant role in determining the leverage ; ( b ) the PEs are utilizing their internal resources and not debt for expansion and financing ; and ( c ) the PEs are mobilizing long-term resources for meeting shortterm requirements. It is concluded that the PEs may be said to adopt pecking order theory while framing their capital structure policies. (18) Rastogi , Jain and Yadav ( 2006 ) conducted a study to : ( a ) examine significant variations in the profile of debt-financing arise due to differences in industry , firm size and age group of corporate enterprises in India ; ( b ) analyse the trend in debt financing practices among the various groups during the period 1992 to 2003 ; and ( c ) ascertain the impact of the liberalized environment on the debt financing decisions of different
80
groups of the sample firms, in phase-2 ( 1998-2003 ) vis-a-vis phase-1 ( 1992-1997 ) of the liberalized business scenario. The study was based on stratified sample of 601 corporate enterprises belonging to fourteen industry groups , and segregated into four firm size classes and two age categories. The evidences suggested that industry and firm size were observed to be the significant factors influencing the composition and maturity structure of debt financing decisions which did not vary significantly across age categories. (19) Saravanan ( 2006 ) examines the impact of ownership pattern , profitability , asset composition , size, risk and growth opportunities on the debt-equity choice of Indian manufacturing firms , based on a sample of 423 firms comprising seven industry classes ( namely, Chemicals , Machinery , Metal and metal products , Textiles , Transport , Non-metal and mineral products , Food and beverage , and Other miscellaneous manufacturing ) over the period 1992-93 to 2001-02. The results of Multiple Regression Analysis indicate that growth oppor-tunities , tax shield , business risk and ownership structure ( promoters’ shareholding ) are significantly positively related to debt-equity ratio, whereas firm size is significantly negatively related to debt-equity ratio. (20) Maji and Ghosh ( 2007 ) test the Static Trade-Off and Pecking Order theories considering four firm-specific determinants of leverage ( namely, size , tangibility , profitability and dividend ) and based on a sample of 160 Indian companies selected from nine manufacturing sectors ( namely, Automobiles, Cement, Computer software, Computer hardware, Electronics, Pharmaceuticals, Steel, Tea and Textiles ) over a period of 14 years from 1990-91 to 2003-04. Multiple regression technique with first-difference method is applied to the data with firms being classified according to leverage, profitability, size and industry. It is concluded that the findings of this study neither support the Trade-off theory , nor the Pecking Order theory ; though much evidence in favor of the former is provided by the results. (21) Reddy and Babu ( 2008 ) conducted a study to : ( a ) analyse the nature and pattern of Indian corporate finance in general, and ( b ) examine the structure of selected pharmaceutical companies listed on the National Stock Exchange ( NSE ). The target adjustment and pecking order models were tested applying pooled Ordinary Least Squares ( OLS ), Generalised Least Squares ( GLS ) , and Fixed Effect regressions. It was observed that : ( i ) the companies were financially stable and mostly financed their investments from retained earnings, and for the rest, they depended more on equity rather than debt finance ; ( b ) the pharmaceutical companies financed their investments mainly through internal funds and
81
the rest through equity share capital. The results obtained contradicted the pecking order model and supported the target adjustment model. (22) Nandy ( 2008 ) examined the influence of four macroeconomic factors ( inflation, stock market performance, debt interest rate and GDP growth rate ) on corporate capital structure based on a sample of 30 companies comprising the BSE Sensex over the period 199798 to 2003-04. Application of correlation analysis resulted in the findings that debt-equity ratio was : ( a ) negatively related to inflation in case of 17 companies ; ( b ) negatively related to stock market performance in case of 24 companies ; ( c ) positively related to debt interest rate in case of 23 companies ; and ( d ) negatively related to GDP growth rate in case of 22 companies. It was concluded that study will encourage the management of the companies to take into account the macroeconomic factors while designing their capital structure. (23) Dogra and Gupta ( 2009 ) conducted a survey study to examine the sources of funds of Small and Medium Enterprises ( SMEs ) operating in the state of Punjab and to analyse the relationship between capital structure of a firm and its characteristics, based on a sample of 50 SME manufacturing units. The results of Pearson chi-square test , Analysis of Variance ( ANOVA ) and factor analysis revealed that capital structure was significantly and highly associated with type of firm , age of firm, growth of firm, degree of competition and level of capital investment. (24) Datta and Agarwal ( 2009 ) attempted to study the determinants of capital structure in Indian scenario based on a panel data sample of 76 non-financial Indian firms listed in the BSE200 Index for the period 2003-2007. The study considered six explanatory variables,namely, profitability, firm size, non-debt tax shield, growth opportunities, tangibility, and tax. Fixed effects panel data regression analysis was applied. It was observed that financing with internal funds, according to the Pecking-Order theory, seemed to have emerged as a major feature of corporate capital structure. Some other determinants, however, had patterns of influences in accordance with the postulates of Trade-Off and Agency Costs theories. It was concluded that the capital structure pattern on an average appeared to portend well for long term development of Indian corporate sector. (25) Sinha and Ghosh ( 2009 ) hypothesize that the firms which follow the Pecking Order Theory ( POT ) may consistently move towards the Market Timing Theory ( MTT ) with dynamic revision , arguing that the cost of asymmetric information related to the equity ( or debt ) financing reduces in the overvalued ( or undervalued ) equity market and in the absence of significant overvaluation or undervaluation, firms finance through internal
82
equity. Hence, by applying a time varying “dynamic market timing measure” , the study examines firms’ market timing strategy to explain the behavior of the cost of asymmetric information. In the case of debt financing, the study confirms that the cost of asymmetric information involves dynamic revision in the short run, but the same disappears over the long run periods when firms tend to follow the MTT consistently. On external equity, the results suggest that firms’ successful market timing lacks persistency and does not happen consistently over the long run study period. (26) Sinha and Ghosh ( 2010 ) , based on a sample of 268 firms from six industries ( namely , automobile , auto ancillary , chemicals , cement , consumer durables, and construction ) examine the adjustment speed in the dynamic capital structure choice of the firms , applying a dynamic Partial Adjustment Model ( PAM ) and extending the work of Drobetz and Wanzenried ( 2006 ). It is observed that firms’ dynamic recapitalization is subject to changes in the firm-specific as well as macroeconomic variables, where both the target leverage and the adjustment speed are determined by firms’ reactive and / or proactive adjustment behaviors. (27) Pathak ( 2010 ) examined the relative importance of six factors ( namely, tangibility of assets, growth, firm size, business risk, liquidity, and profitability ) in the capital structure decisions of publicly traded Indian firms, based on a sample of over 135 firms listed on the Bombay Stock Exchange during the period of 1990-2009. Ordinary least squares regression technique was applied. It is found that factors such as tangibility of assets, growth, firm size, business risk, liquidity, and profitability have significant influences
on capital structure of Indian firms. (28) Bhattacharjee ( 2010 ) conducted an empirical study of the determinants of capital structure based on a balanced panel data sample of 22 Information Technology ( IT ) firms listed on the Bombay Stock Exchange ( BSE ) over the period 2003-04 to 2007-08. Correlation analysis and panel data regression analysis were applied. The results suggested ‘growth’ to be the only significant factor affecting capital structure of IT firms. (29) Rajagopal ( 2010 ) attempted to test the portability of capital structure theories across developed and developing economies like India. The study was based on a cross-sectional sample ranging from 1110 to 1163 manufacturing firms comprising 22 industries over a period of five years from 1998-2002. Leverage was measured by six ratios { ratios of book values of total debt, long-term debt and short-term debt to book values ( and quasimarket values ) of total assets }. The explanatory variables included firm size, tangibility , non-debt tax shields, profitability, market-to-book ratio and business risk. Multiple OLS
83
regression was applied. The results showed that a common set of independent variables appeared to be significant in explaining the cross-sectional variation in leverage ratios in a developed economy such as the US and an emerging market such as India. This study provided strong evidence that capital structure theory is potentially portable across developed and developing countries, and that traditional theory is quite certainly applicable to India. (30) Mahakud and Misra ( 2010 ) investigated the role of adjustment costs and other firm-specific variables like tangibility, growth opportunity, size of the company, profitability, volatility, non-debt tax shields and uniqueness of the company in the determination of capital structure based on a panel data sample of 793 Indian manufacturing companies during the period 1995-1996 to 2006-2007. A dynamic partial adjustment model along with the Generalised Method of Moments ( GMM ) technique was used to test the dynamics of capital structure. The results indicated the firms have a target capital structure. The adjustment speed towards the target capital structure appeared to be reasonably high and it varied with the definition of leverage ratio. All these evidences were consistent with the trade-off theory of corporate capital structure. (31) Mahakud and Mukherjee ( 2011 ) conducted a study to identify the determinants of target capital structure and the adjustment speed to target capital structure based on a balanced panel data sample of 891 Indian manufacturing companies over the period from 1993-94 to 2007-08. Application of Generalized Method of Moments ( GMM ) estimation technique to a dynamic partial adjustment model showed that : ( a ) size , profitability , growth opportunity , tangibility and uniqueness appeared to be significant determinants of target capital structure ; and ( b ) the speed of adjustment to target capital structure is significantly influenced by financial constraints , external financing cost , distress cost , ownership and macro-economic conditions. (32) Deepa and Azhagaiah ( 2011 ) attempted to determine the predictors of capital structure in the beverages and alcohol industry in India and also to find out the approach followed by these firms to decide their capital structure. A cross-sectional sample of 21 companies listed on the Bombay Stock Exchange during the period 2000-01 to 2009-10 was considered. Correlation analysis and multiple regression analysis were conducted. Tangibility and profitability were found to be the major determinants of capital structure. The findings suggested that Indian beverage and alcohol industry tend to follow pecking order theory.
84
(33) Panigrahi ( 2011 ) analyzed whether the location of a firm affects its capital structure decisions. The analysis was conducted on a sample of 300 private sector companies, comprising 20 different sectors for the period 1999-2000 to 2007-2008 , duly grouping them on the basis of their regions as : eastern ( 34 ) , western ( 135 ) , southern ( 85 ) and northern ( 46 ). The findings showed that the region or location of a company strongly influences the quantum of inflow of funds. (34) Mishra ( 2011 ) conducted a study to identify the determinants of capital structure of Public Sector Undertakings ( PSUs ) , based on a sample of 41 profit making manufacturing PSUs over the period 2006 to 2010. The independent variables were identified in accordance with Agency Cost Theory , Pecking Order Hypothesis and other established capital structure models. The results of multiple regression analysis indicate that leverage is significantly and negatively affected by asset structure , profitability and tax
and
positively by growth. (35) Naha and Roy ( 2011 ) analyze the impact of product market behaviour captured from three different angles - structure ( concentration ratio ) , conduct ( advertising expenditure ) and performance ( return on assets ) } - on debt ratios ( short-term and longterm ) of 50 manufacturing firm belonging to eight different industries over a period of nineteen years from 1985-86 to 2004-05 , after controlling for other determinants ( namely , firm size, tangibility , non-debt tax shield , financial distress , growth , potential debt tax shield and industry dummy variables ) of capital structure and under-scoring the role of economic liberalization ( through structural break dummy variable ) , using panel data regression20 analysis. It is observed that the structure and conduct is significant in influencing only the short-term debt ratio whereas performance is consistently having a negative impact on both short-term and long-term debt ratios. The structural break dummy is also significant irrespective of the maturity structure of debt. (36) Agarwal , Iyer and Yadav ( 2012 ) investigated capital structure practices in the Indian industry, through a sample of top 500 companies classified in 19 industries for a 10 year period from 1998 to 2007. The study considered 67 variables including the leverage variables. The relationship of leverage ratios with market capitalization and Earnings Per Share ( EPS ) was also explored. Multi-objective criteria for processing capital structure decisions was identified and justified based on the findings of the past researches
20
Using industry dummy variables in ( cross-sectional ) fixed effects panel data regression will inevitably result result in perfect collinearity as is evident from the paper. So the results of Pooled OLS are reported thus mitigating the power of the analysis to a large extent.
85
and the empirical survey conducted as a part of this study. In the empirical survey, CFOs as respondents were investigated for their goals, priorities, motivations, constraints and practices for capital structure decision making. Goal Programming model was selected from different multi-objective optimization techniques. The model was found to be capable of providing satisfying solutions to multiple goals simultaneously by minimizing the deviation from the objective function after assuming that the decision maker is an optimist and does not attempt to satisfy all objectives fully. (37) Mohapatra ( 2012 ) investigated the impact of profitability, operating leverage and industry class on the capital structure, considering a cross-sectional sample of 626 nongovernment and non-financial companies ( with paid up capital of rupees one crore and above ) for the period 1987-88 to 2009-10. Applying correlation analysis and analysis of variance ( ANOVA ), he found that capital structure of Indian firms get significantly influenced by the industry class and operating leverage whereas profitability does not have significant bearing on the capital structure. (38) Rasoolpur ( 2012 ) analyzed the determinants of corporate capital structure based on a balanced panel data sample of 298 manufacturing firms ( out of the top 500 firms listed on the BSE and selected on the basis of the turnover for the year 2004-2005 ) over the period 1995-96 to 2005-06. Applying pairwise correlation analysis and fixed-effects panel data regression analysis they found uniqueness and liquidity to be the important determinants of capital structure. (39) Srivastava ( 2012 ) conducts a study in oder to identify important determinants of capital structure and also to verify the applicability of trade-off theory and pecking order theory. The period of study , from 1977-78 to 2006-2007, is divided into two equal subperiods designated as pre- and post- liberalization periods. The data for the study comprises of consolidated sources and uses of funds of Indian Public Limited companies obtained from various issues of RBI bulletin. Two formulations of capital structure, the dependent variable, namely, total debt to total assets ratio and long-term debt to total assets ratio , are considered. The explanatory variables include firm size , asset structure , non-debt tax shield ( NDTS ) , cash , growth opportunities and profitability. The explanatory power of the model measured in terms of R2 for the total time period is 62% and 66% for total debt and long-term debt respectively. However , the values of R2 for the same variables are 93% and 98% in the pre-liberalization period , and 86 % and 80% for post-liberalization. More-over , as confirmed by Chow Test , the statistical significance of variables also changed in different time periods showing that variables have varied importance in various
86
time periods. Thus the regression analysis suggests applicability of both trade-off theory and pecking order theories. (40) Krishnankutty and Chakraborty ( 2013 ) investigated the determinants of capital structure considering a balanced panel data sample of 213 non-financial companies { comprising Bombay Stock Exchange ( BSE ) 500 Index } during the period 2002-2011. Two measures of capital structure were proposed : natural logarithm of total debt , and total debt to total assets ratio. The independent variables included asset structure , profitability , non-debt tax shield , debt capacity , financial risk , size , liquidity and long-term leverage. Since most of the variables showed skewed and peaked distributions the study relied upon conditional quantile regression analysis as an appropriate tool. Five quantiles ( 5th , 25th , 50th, 75th and 95th ) were considered. It was found that long-term leverage, assets structure and size were positively determining the capital structure at the 5 th quantile. For the quarter quantile ( 25th ) and median quantile ( 50th ) financial risk , long-term leverage and size were positively , and debt capacity was negatively determining the capital structure. Moreover, financial risk and long-term leverage were positively determining the capital structure at the 75th quantile. However, in case of 95th quantile long-term leverage and nondebt tax shield were positively determining the capital structure while debt capacity was negatively determining capital structure. Thus there were marked non-linearities in the determinants of capital structure. (41) Ray ( 2013 ) examined the determinants of capital structure, based on a cross-sectional sample of 39 cement companies listed on the Bombay Stock Exchange ( BSE ) during the period 1991-92 to 2011-12. Considering leverage ( Debt-Equity ratio ) as the dependent variable, the explanatory variables included asset collateral, asset composition, age of firm, size of firm, business risk, growth rate, flexibility, profitability, non-debt tax shield. Pearson correlation analysis and Two Stage Least Square regression analysis were conducted. The results showed that asset composition, size and non-debt tax shields were having statistically positive relationship with debt-equity ratio , whereas age, profitability and asset collateral were having significant negative relations with leverage. The other factors such as business risk , flexibility and growth opportunities did not seem to have any significant impact on capital structure.
87
(42) Basu and Rajeev ( 2013 ) made an attempt to answer two crucial issues : ( a ) whether capital market regulations exert any influence on capital structure decisions of Indian corporate firms, and ( b ) how to measure the capabilities of firm-specific factors to explain two theories of capital structure namely, static trade-off theory and pecking order hypothesis. Static panel data model, as proposed by Driscoll and Kraay ( 1998 ) , was employed considering an unbalanced panel data sample of 1154 firms for a period of 21 years ( 1989-2009 ). Evidence was found in favour of the argument that institutional factors matter in financing decisions of corporations. The results suggested that capital market regulations in India have adverse impact on the use of public debt and favorable impact on the use of equity capital. Moreover , firm-specific factors are more capable of explaining trade-off theory rather than explaining the information asymmetry in the public domain explaining the fact that private lenders have superior firm-specific information which helps firms in mobilizing resources from private sources rather than from the market. (43) Rasoolpur and Warne ( 2013 ) examined the leverage decisions of textile and ready-made garments industry of the Indian corporate sector considering a sample of 19 firms of that industry ( out of top 500 manufacturing firms selected on the basis of the turnover for the year 2004-2005 ) over the period 1995-96 to 2005-06. that : ( i ) 47.29 percent of the firms are lying in 0-100 percent capital structure range which is much below the standard ratio of 2:1 ; ( ii ) 33.50 percent of the firms are lying in 100-200 percent capital structure range which is less than the standard ratio of 2:1 ; ( iii ) 5.42 percent of the firms are lying in 190 to 210 percent capital structure range which is close to the standard ratio of 2:1 ; ( iv ) 19.21 percent of the firms are lying in more than 200 percent capital structure range which is well beyond the standard ratio of 2:1. It may thus be inferred that in this industry, around half of the firms may be said to employ low level of leverage, one-third of firms may be said to employ moderate level of leverage and rest of the firms may be said to employ high level of leverage in their capital structure. (44) Panda, Mohapatra and Moharana ( 2013 ) analysed the determinants of capital structure of Indian Steel Industry based on a sample of 66 companies over the period 2000 to 2010. Considering eight independent variables ( namely , asset structure , profitability, growth opportunities , business risk , size , non-debt tax shields , uniqueness and liquidity ) from early studies and employing correlation analysis, multiple regression and stepwise regression techniques, it was found that only three variables ( namely , profitability , growth
88
and risk ) were having significant impact on the debt ratio of these steel companies and were following the postulates of the trade-off theory. (45) Poddar and Mittal ( 2014 ) conducted a study to examine the determinants of capital
structure based on a balanced panel of 5 steel companies ( Tata Steel , Steel
Authority of India , Hindustan Zinc , Bhushan Steel and Ahmedabad Steel ) over a period of 14 years from 1998 to 2011. The explanatory variables included profitability, liquidity, size and interest coverage. Variants of panel data regression analyses [ namely , ( a ) intercept and slope are constant across time and but error term captures differences over time and cross-sections ; ( b ) slope coefficients are constant but the intercept varies across crosssections ; ( c ) slope coefficients are constant but the intercept varies across time ; ( d ) slope coefficients are constant but the intercept varies over time and cross-sections ; and ( e ) intercept and slope vary across cross-sections ] were applied. The results showed profitability and liquidity to be the significant determinants of capital structure and also the existence of pronounced individual company effect but not the time effect. (46) Sinha and Samanta ( 2014 ) examine the impact of eight firm-specific determinants of corporate capital structure, namely, firm size, tangibility, growth opportunities, profitability, non-debt tax shields, operating risk, liquidity and firm age, over the entire conditional distribution of leverage through the application of conditional quantile regression methodology on a balanced panel data related to a selected sample of 76 Indian pharmaceutical companies listed on the Bombay Stock Exchange over a period of 10 years from 200203 to 2011-12. The coefficients of the explanatory variables are estimated at five quantiles, namely, 5th , 25th , 50th , 75th and 95th . The empirical results indicate marked non-linearities in the relationships between leverage and its firm-specific determinants with the estimated regression coefficients of the explanatory variables changing magnitudes and statistical signi-ficance accompanied by change of signs ( in some cases ) at different quantiles. The study, though limited to a particular industry, affirms the existence of non - linearities in capital structure determinants of corporate firms in the Indian scenario. (47) Handoo and Sharma ( 2014 ) seek to identify the most important determinants of capital structure based on a sample of 870 listed Indian firms comprising both private sector companies and government companies over the period 2001-2010 , considering ten independent variables ( namely , profitability , growth , asset tangibility , size , cost of debt , liquidity, financial distress , tax rate , debt serving capacity , and age ) and three dependent variables ( namely , total debt ratio, long-term debt ratio and short-term debt ratio ) through
89
the application of multiple regression analysis. It is concluded that factors such as profitability, growth , asset tangibility, size , cost of debt, tax rate, and debt serving capacity have significant impact on the leverage structure chosen by firms in the Indian context. (48) Ghosh ( 2015 ) examines the interrelationships among leverage, debt maturity and source of debt based on a sample of 1556 manufacturing firms over the period 1996 to 2012. Application of 3SLS ( Three Stage Least Squares ) regression technique shows that : ( a ) the three variables are interrelated with each other tending to complement or substitute the other ; ( b ) disaggregating firms on the basis of equity and board presence, the effect of leverage on debt maturity is the highest for firms that do not exhibit close relationships with banks ; and ( c ) having no seat on the firm board makes it difficult for banks to exercise control over the firm’s indebtedness. (49) Tiwari and Krishnankutty ( 2015 ) attempted to analyze the firm-specific determinants of capital structure ( namely , size , tangibility , profitability , non-debt tax shields and growth ) for Indian firms and to investigate whether the capital structure models derived from Western settings provide convincing explanations for capital structure decisions of the Indian firms. Conditional Quantile Regression ( CQR ) technique was applied to a balanced panel data sample of 298 firms ( from the BSE 500 firms ) over the period from 2001 to 2010. The results showed marked non-linear behaviour of the determinants of leverage over the entire conditional distribution of leverage. (50) Chadha and Sharma ( 2015 ) study the key determinants of capital structure of Indian manufacturing companies in order deduce the applicability of trade - off theory and pecking order theory , considering a sample of 422 companies listed on Bombay Stock Exchange over the period from 2003-2004 to 2012-2013. The
results of panel data
regression analysis indicate that size, age, asset tangibility, growth, profitability, non-debt tax shield, business risk, uniqueness and ownership structure are significantly related to capital structure. It is concluded that a mix of both the theories may be said to explain the capital structure of Indian manufacturing sector. (51) Chaklader and Chawla ( 2016 ) investigated the following six determinants of capital structure ( namely, growth , profitability , tangibility , liquidity , size and non-debt tax shield ) of firms comprising the NSE CNX 500 over the period 2008-2015 , with a view to find out whether the results are in line with Pecking Order Theory ( POT ) or TradeOff Theory ( TOT ) of capital structure. The results of the study were found to be more inclined towards TOT with growth, profitability, size, tangibility and non-debt tax shield supporting the predictions of TOT and liquidity supporting the predictions of POT.
90
(52) Sathyanarayana, Harish and Kumar ( 2017 ) examined the relationship between various identified determinants ( profitability, tangibility, growth rate, business risk, size and non-debt tax shield ) and its impact on financial leverage ( CS ) decisions of Capital goods, FMCG, Infrastructure and IT sectors in Indian Stock market. Data was collected from the published financial statements of quoted firms in the Indian stock market from the above mentioned sectors for a period of ten years from 2006 to 2015 . After testing the data for multicollinearity, linear multiple regression model has been used to investigate the impact of chosen independent variables on CS ( leverage ) decisions in Indian capital market. Moreover , residual diagnostic tests , such as serial correlation test , heteroskedasticity test , normality and CUSUM test have been run to assess the strength of the constructed regression model. The results show that profitability , tangibility and growth were the major determinants in case of capital goods sector and profitability , tangibility , growth , size and non-debt tax shield were the major factors for the FMCG sector. For the Infrastructure sector the major factors were growth , business risk and size and in case of the IT sector the important determinants were profitability , business risk and size. The study revealed inconsistency in independent variables influencing the financial leverage component, though there is statistical support for the proposed determinants with respect to profitability and growth rate influencing the financial leverage. (53) Sinha and Samanta ( 2017 ) analyse the impact of eight firm-specific determinants of capital structure ( namely, firm size, tangibility, non-debt tax shields, growth opportunities, profitability, distance from bankruptcy, liquidity and dividend payment ) over the entire unconditional distribution of leverage ( measured by quasi market value of debt to equity ratio, MLEV ) through the application of panel data-unconditional quantile regression methodology ( fixed effects - recentered influence function regression technique ) on a balanced panel data related to a selected sample of 593 Indian manufacturing companies listed on the Bombay Stock Exchange ( BSE ) of India over a period of 18 years from 1997-98 to 2014-15, with a view to test the existence of non-linear behaviours of the determinants of capital structure . The coefficients of the explanatory variables are estimated at seven quantiles ( namely, 10th , 25th , 40th , 50th , 60th , 75th and 90th ). The empirical results indicate marked non-linearities in the relationships between leverage and its firm-specific determinants with the estimated regression coefficients of the explanatory variables changing magnitudes and statistical significances accompanied by change of signs ( in case of tangibility and growth ) at different unconditional quantiles of leverage.
91
3.4
Conclusion
The major empirical studies based on the modern theories of capital structure and conducted in the contexts of International and Indian scenarios have been discussed in this chapter, with a view to identify the research problem ( s ) and to formulate the research objectives of this study in the following chapter.
CHAPTER 4
92
Chapter 4 Research Methodology 4.1
Introduction
The generic meaning of the term ‘research’ may interpreted as a combination of the two equations : ( 1 ) research = ‘re-’ + ‘search’
(1)
where ‘re-’ means “Again ; anew ” 1 , and ‘search’ means “ To make a careful examination or investigation ” 2 ; thus implying the interpretation of the term ‘ research ’ as ‘ To make a careful examination or investigation again ’. ( 2 ) research = ‘re’ + ‘search’
(2)
where ‘re’ means “About ; concerning” 3 , originating from Latin rē , ablative case of rēs, meaning ‘thing’ 4 , and ‘search’ has already been defined in ( 1 ) above ; thus implying the interpretation of the term ‘ research’ as ‘ To make a careful examination or investigation about a thing ’. Combining the above two interpretations, the term ‘ research ’ may be generically defined as ‘ To make a careful examination or investigation about a thing, again ’. The literal meaning of the term ‘ methodology ’ is “ a body of practices, procedures, and rules used by those who work in a discipline or engage in an inquiry ; a set of working methods. ” 5 or a “system of methods and principles used in a particular discipline.” 6 Hence , we may define ‘research methodology’ literally as ‘ A system of methods , principles, procedures or rules used in a particular discipline or field of inquiry in order to make a careful examination or investigation about a thing relevant to that discipline or field of inquiry, again.’
1
The American Heritage® Dictionary of the English Language.
2
Ibid.
3
Oxford English Dictionary.
4
The American Heritage® Dictionary of the English Language.
5
Ibid.
6
Collins English Dictionary.
93
Some formal definitions of ‘research’ and ‘research methodology’ are cited below : “ Research is a systematic process of collecting, analysing, and interpreting information (data) in order to increase our understanding of the phenomenon about which we are interested or concerned ” ( Leedy and Ormrod , 2010 , p. 2 ). “ The term ‘research’ refers to the systematic method consisting of enunciating the problem, formulating a hypothesis, collecting the facts or data, analysing the facts and reaching certain conclusions either in the form of solution (s) towards the concerned problem or in certain generalisations for some theoretical formulation ” ( Kothari , 2004 , p. 2 ). “ Underlying and unifying any research project is its methodology. The research methodology directs the whole endeavor : It controls the study, dictates how the data are acquired, arranges them in logical relationships, sets up an approach for refining and synthesizing them, suggests a manner in which the meanings that lie below the surface of the data become manifest, and finally yields one or more conclusions that lead to an expansion of knowledge ” ( Ibid., p. 6 ). According to Leedy and Ormrod ( 2010 , pp. 2-5 ) , research : ( a ) originates with a problem or question ; ( b ) requires an unambiguous articulation of a goal ; ( c ) requires a specific plan or design for proceeding ; ( d ) generally divides the principal problem into smaller sub-problems which can be managed with ease ; ( e ) is guided by specific research hypotheses ; ( f ) accepts certain valid and critical assumptions ; ( g ) requires the collection of data and interpretation of findings for the resolution of the problem initiating the research ; ( h ) is cyclical or more exactly helical in nature. This chapter comprises of eight more sections excluding this first section. The second section deals with the identification of the research problem(s) based on the literature review conducted in the previous chapter and description of the objective(s) for undertaking this research study . The third section prescribes the determinants of capital structure to be selected for this study along with their theoretical interpretations based on the capital structure theories. The fourth section deals with the formulation of the dependent and the independent variables. The fifth section presents the nature and source of data and the procedure of sample selection . The sixth section discusses briefly the statistical techniques for preliminary analyses of the sample data. The seventh section provides a detailed discu-
94
ssion on the econometric framework of the study. The eighth section presents the specific hypotheses to be tested in order to provide valid answers to the research problem(s) and objective(s) framed in the second section of this chapter. The last section concludes this chapter.
4.2 4.2.1
Research Problem(s) and Research Objective(s) Identification of Research Problem(s)
The theories of corporate capital structure and the empirical evidences based on these theories , have respectively been discussed and reviewed in the previous two chapters. The process of identification of research problem(s) to be addressed by this study will comprise of the following sequential stages based on the relevant observations and comments made after a detailed perusal of the review of prior empirical studies:-
( I ) Stage 1 ( A ) Observations and comments :( 1 ) An important area of research in corporate capital structure is the issue of nonlinearity in capital structure choices of firms . This has been studied from the following perspectives: ( a ) application of polynomial functional forms of second-order or third-order . Arce et al. ( 2015 ) tested whether the mis-specifications of the linear model of Lemmon et al. ( 2008 ) are concerned with non-linear models and confirmed that higher-order terms were significant ; ( b ) application of threshold level model used to find a threshold debt ratio where capital structure decisions can be explained or split by two different linear functions. If the current debt ratio is below this threshold debt ratio, the firm will increase its debt ratio and vice versa ( Nieh et al. , 2005 ) ; ( c ) application of ‘Quantile Regression’ ( QR ) methodology which , unlike standard or mean regression , properly captures the heterogeneous relationships between the variable measuring capital structure ( say , V ) and its determinant variables over the entire distribution of V. Fattouh et al. ( 2005 , 2008 ) , Galvao and Montes-Rojas ( 2010 ) , Bahng and Jeong ( 2012 ) , Oliveira et al. ( 2013 ) , Krishnankutty and Chakraborty ( 2013 ) , Sánchez ( 2014 ) , Wellalage and Locke ( 2013 , 2014 ) , Arshanapalli and Nelson ( 2014 ) , Choi et al. ( 2014 ) , Sinha and Samanta ( 2014a ) , Sinha and Samanta ( 2014b ), Frank and Goyal ( 2015 ) , Aviral and Raveesh ( 2015 ) , Tiwari and Krishnankutty ( 2015 ) , Gwatidzo et al. ( 2016 ) , Ferrarini et al.
95
( 2017 ) and Sinha and Samanta ( 2017 ) explore the non-linear effects in the firm-specific determinants of capital structure applying QR methodology and conclude the existence of such effects ; but all the above studies , except Sinha7 and Samanta ( 2017 )8 , have applied ‘Conditional Quantile Regression’ ( CQR ) methodology , which may not be appropriate , as the interpretations of the regression coefficients become problematic when covariates are added to the CQR model ; the technique of ‘Unconditional Quantile Regression’ ( UQR ) is said to overcome this problem.9 ( 2 ) The approach mentioned in 1 ( c ) is more popular among the researchers than approaches mentioned in 1 ( a ) and 1 ( b ) , and needs a further re-look through the application of Unconditional Quantile Regression. ( B ) Identified Research Problem ( partial statement ) : To re-examine the issue of nonlinearity in capital structure choices of corporate firms through the application of the hitherto unapplied technique of ‘unconditional quantile regression’.
( II ) Stage 2 ( A ) Observations and comments :( 1 ) A substantially major portion of such studies are devoted to the trade-off and pecking order theories lending support to the following observations by Miglo ( 2013 , p. 2 ) : ( a ) “ …researchers have extensively tested trade-off and pecking order theories” , ( b) “ Compared to trade-off and pecking order theories , theoretical aspects of market timing theory are underdeveloped ” ; ( c ) “…signaling theory of capital structure lacks empirical support regarding some of its core predictions” ; and ( d ) “…agency theories of capital structure helped to explain many financing strategies and qualitative phenomena related to capital structure. They do not seem however to compete successfully against major theories in terms of finding optimal capital structure quantitatively.” ( 2 ) The applicability of the theories of capital structure in the practical scenario of capital structure decision making process may be ascertained empirically : ( a ) directly , through the analysis of the core models formulated on the basis of the fundamental postulates of these theories ; or 7
The present researcher.
8
To the best of the knowledge of the present researcher, this is the first study on determinants of capital structure which has considered UQR technique and this study forms an integral part of this thesis. 9
Detailed discussion in section 7 of this chapter.
96
( b ) indirectly , through the analysis of the implications or predictions of these theories on the determinants of capital structure , that is , factors affecting capital structure ; ( c ) indirectly , through survey studies conducted on managers of corporate firms. ( 3 ) The number of prior studies following approach mentioned in 2(b) far outnumber those traversing the path of the approaches mentioned in 2(a) and 2(c) . ( 4 ) Application of quantile regression methodology to approach mentioned in 2(a) , particularly about the dynamic aspects of the capital structure theories , has only been followed by Galvao and Montes-Rojas ( 2010 ) in the conditional quantile regression framework . An econometric model for the dynamic aspects of the capital structure theories based on the unconditional quantile regression framework is yet to be formulated . ( B ) Identified Research Problem ( partial statement ) : To re-examine the issue of nonlinearity in capital structure choices of corporate firms through the application of the hitherto unapplied technique of ‘unconditional quantile regression’ with a view to indirectly ascertain the applicability of Trade-off Theory , Agency Cost theory and Pecking Order Theory through the analysis of the predictions of these theories on the determinants of capital structure. ( III ) Stage 3 ( A ) Observations and Comments :( 1 ) Most of the studies on the determinants of capital structure have been conducted in the cases of developed economies such as USA, UK, Germany, France, Japan, etc. Studies in the context of emerging economies like India have gained momentum in the last few years. Prior research studies , such as Booth et al. ( 2001 ) , De Jong et al. ( 2008 ) and Rajagopal ( 2010 ) have concluded that the firm-specific determinants of capital structures of developed economies are very much relevant for emerging economies as well. Also , previous empirical studies in the Indian context , a vast majority of which have considered the manufacturing sector , have produced mixed results regarding the impact of these determinants on leverage, there is scope for further research on this issue. ( 2 ) Given the advantages of panel data regression models over cross-sectional regression models ( as discussed in the fifth section of this chapter ) some studies
10
in the Indian
context , to the best of the knowledge of the present researcher , have applied panel data
Guha-Khasnobis and Bhaduri ( 2002 ), Bhole and Mahakud ( 2004 ), Maji and Ghosh ( 2007 ), Reddy and Babu ( 2008 ), Datta and Agarwal ( 2009 ), Mahakud and Misra ( 2010 ), Mahakud and Mukherjee ( 2011 ), Rasoolpur ( 2012 ), Basu and Rajeev ( 2013 ), Poddar and Mittal ( 2014 ), Chadha and Sharma ( 2015 ) and Sinha and Samanta ( 2017 ) . 10
97
regression methodology in the truest sense , that is , fixed effects or random effects regression models and not ‘ pooled regression models ’. Considering the studies on the application of quantile regression , only Galvao and Montes-Rojas ( 2010 ) , Gwatidzo et al.( 2016 ) , Ferrarini et al.( 2017 ) and Sinha11 and Samanta ( 2017 )12 consider panel data quantile regression models ( with firm fixed effects 13 ) in the truest sense. ( B ) Research Problem ( final statement ) : To re-examine the issue of non-linearity in capital structure choices of corporate firms through the application of the hitherto un-applied technique of ‘panel data unconditional quantile regression’ with a view to indirectly ascertain the applicability of Trade-off Theory , Agency Cost theory and Pecking Order Theory through the analysis of the predictions of these theories on the firm-specific determinants of capital structure of Indian manufacturing firms.
4.2.2
Objectives of Research Study
The objectives of this research study may be stated as follows : ( 1 ) To choose the appropriate panel data model from three mutually exclusive panel data models ( Pooled Ordinary Least Squares Model, Fixed Effects Model and Random Effects Model ) for analysing the impact of firm-specific determinants of capital structure on the mean of the variable measuring capital structure ( say , “ V ” ) in respect of Indian manufacturing companies listed on the Bombay Stock Exchange ( BSE ), with a view to indirectly assess the applicability of Trade -Off Theory ( TOT )14 , Agency Cost Theory ( ACT ) and Pecking Order Theory ( POT ) for an average company with mean ( average ) level of leverage. ( 2 ) To apply ‘unconditional quantile regression technique’ on the chosen panel data model for analysing the differential impact ( that is , non-linear behaviour ) of the firm - specific determinants of capital structure over the entire unconditional distribution of the variable “ V ” in respect of Indian manufacturing companies listed on the Bombay Stock Exchange ( BSE ) , with a view to assess indirectly the applicability of Trade - Off Theory ( TOT ) , Agency Cost Theory ( ACT ) and Pecking Order Theory ( POT ) at different quantiles of
11 12
The present researcher. This unpublished conference paper is an integral part of this thesis.
In the context of the study of capital structure determinants, ‘ year fixed effects’ control for the macroeconomic determinants of capital structure and models with only ‘ year fixed effects’ through the inclusion of time dummy variables, cannot be construed to be panel data fixed effects model in the truest sense. 13
14
Traditional Static Trade-Off Theory or Tax Shield-Bankruptcy Cost (TS-BC) Theory.
98
the unconditional distribution of V, that is , for representative 15 companies having varying ( for example , ‘very low’ , ‘low’ , ‘moderate’ , ‘high’ or ‘very high’ ) levels of leverage. The justifications for choosing Indian manufacturing companies listed on the BSE may be explained as follows :( 1 ) India has been chosen because : ( a ) it is one of the fastest growing emerging market economies in the world, and ( b ) the researcher was able to get access to database of Indian companies only. ( 2 ) Manufacturing companies have been chosen so as to facilitate comparison of the nonlinear behaviour of the determinants of capital structure to be examined in this study with ( linear or non-linear ) behaviour examined by the previous Indian studies, a vast majority of which have considered the manufacturing sector. ( 3 ) The Bombay Stock Exchange ( BSE ) 16 , having far greater number of listed companies than its nearest rival, the National Stock Exchange ( NSE )17, has been considered as the study seeks to form a panel data sample comprising a substantial size of cross-sectional units over a substantially long period of time.
15
A representative company at a particular quantile of leverage represents a company with the corresponding level of leverage , for instance , a representative company at the 50th quantile represents a company with median ( or moderate ) level of leverage , or a representative company at the 10 th quantile may be said to represent a company with very low level of leverage. 16
BSE became the country’s first listed stock exchange on 3rd February, 2017.
More than 5500 companies are listed on BSE making it world’s largest exchange in terms of listed companies, http://www.bseindia.com/static/about/introduction.aspx? accessed on 28.03.2017. The number of listed companies on NSE is 1571,https://www.nseindia.com/corporates/corporateHome.html accessed on 28.03.2017 . 17
99
4.3
Determinants of Capital Structure
A number of determinants likely to affect corporate capital structure decision have been enumerated by empirical literature. These determinants act as proxies for the various factors of market imperfections ( such as taxation, bankruptcy costs, agency problems and asymmetric information ) on which the widely discussed modern theories of capital structure postModigliani and Miller ( M-M ) Hypothesis [ namely, Trade-Off Theory ( TOT ), Agency Cost Theory ( ACT ) and Pecking Order Theory ( POT ) ] are based. A vast majority of prior empirical studies as reviewed in the previous chapter consider only the firm-specific determinants of capital structure after controlling for : ( i ) unobserved industry effects
18
through the inclusion of Industry Median Leverage or
Industry Dummy Variables ; and / or ( ii ) unobserved macro-economic and institutional factors 19 through the utilization of Time Dummy Variables. Moreover, all the prior studies dealing with the non-linearity issue, as mentioned above, consider only the firm-specific determinants. This study will consider the following firm-specific determinants of capital structure : ( i ) firm size, ( ii ) tangibility, ( iii ) non-debt tax shield, ( iv ) profitability, ( v ) growth opportunities, ( vi ) bankruptcy risk ( distance from bankruptcy ) , and ( vii ) liquidity , after controlling for the unobserved industry effects and the unobserved macro-economic and institutional factors through the inclusion of Industry Dummy Variables and Time Dummy Variables respectively. The implications of Trade-Off Theory ( TOT ) 20 , Agency Cost Theory ( ACT ) and Pecking Order Theory ( POT ) on these determinants are explained below.
18
Industry - specific determinants ( or observed industry effects ) have been considered by MacKay and Phillips ( 2005 ) who use a firm’s industry position variable, that is , a proxy for natural hedge measuring a firm’s proximity to the median industry capital-labor ratio ) after controlling for industry concentration ratio ( Herfindahl- Hirschman Index, HHI ) and by Kayo and Kimura (2011) who use munificence and dynamism with industry concentration ( HHI ) as a control variable. Some studies such as Shuetrim et al.(1993), Rajan and Zingales (1995), Hirota (1999), Booth et al.(2001), Deesomsak et al.(2004) , Eldomiaty et al. (2005), Nandy (2008), Fan et al. (2010), Mahakud and Mukherjee (2011), Lemma and Nagesh (2013), Bayrakdaroğlu et al.(2013), Karin (2013), Mokhova and Zinecker (2014) use specific macro-economic and institutional determinants. 19
20
This is the traditional TOT or more specifically the ‘Tax Shield - Bankruptcy Cost Theory’ ( TS-BCT ).
100
( 1 ) Firm size ( A ) TOT : Firm size may be said to serve as a proxy for expected bankruptcy costs or an inverse proxy for the probability of bankruptcy ( Rajan and Zingales , 1995 ). Larger firms tend to be more diversified ( Remmers et al., 1974 ; Ferri and Jones , 1979 ) , enjoy easier access to the capital markets, receive higher credit ratings for their debt issues, pay lower interest rates on borrowed funds ( Pinches and Mingo , 1973 ; Ferri and Jones , op. cit. ) , and are likely to exhibit lower volatility in profits, cash flows, and firm values which would lower the probability of costly bankruptcy or financial distress, and would allow larger firms to take on larger debt burdens ( Titman and Wessels , 1988 ; Parsons and Titman , 2009 ). Thus , firm size will be positively related to debt. ( B ) ACT : Larger firms tend to provide more information to lenders , have lower agency costs of debt and comparatively lower costs of monitoring than smaller firms ( Fama and Jensen , 1983 ). So , larger firms are expected to employ relatively higher levels of debt. Grinbalt and Titman ( 1998 ) argue that the conflicts between debtholders and equity shareholders are experienced by smaller firms more severely than that by larger firms because : ( i ) smaller firms being more flexible are better able to enhance the risk of their investment projects, and ( ii ) top managers of smaller firms are more likely to be the major shareholders who may be said to have incentives for making choices benefitting equity shareholders at the expense of debtholders. So , lower debt ratios are likely to be exhibited by smaller firms. Hence , firm size will be positively related to leverage. ( C ) POT : Firm size may also serve as a proxy for the information available with outside investors , which should increase their preference for equity relative to debt ( Rajan and Zingales , op. cit.). Larger firms having a better reputation ( Diamond , 1989 ) will have less asymmetric information problems, should tend to have more equity than debt, and thus have lower leverage. Hence , firm size and leverage are expected to be negatively related.
101
( 2 ) Tangibility The tangibility of the assets of a firm serves as a proxy for costs of financial distress which include bankruptcy costs and agency costs. A firm with more tangible assets may be said to have greater ability of issuing secured debt and revealing less information about future profits ( Booth et al., 2001 ). ( A ) TOT : The effect of the collateral value of assets of a firm on its gearing level is represented by the tangibility of its assets ( Rajan and Zingales , 1995 ). Tangible assets are said to provide high collateral value relative to intangible assets implying that these assets can support more debt. Also, tangible assets often reduce the costs of financial distress because they tend to have higher liquidation value than intangible assets ( Harris and Raviv, 1990 ). Hence , tangibility and leverage may be said to be positively related. ( B ) ACT : ( i ) Galai and Masulis ( 1976 ) , Jensen and Meckling ( 1976 ) , and Myers ( 1977 ) argue that shareholders of leveraged firms are driven by the objective of expropriating wealth from the bondholders and have an incentive to invest sub-optimally. The collateralizaton of debt, that is, securing debt against tangible assets, tends to restrict the borrower in using funds for a specific project ; and depending on the value of the assets utilized for collateralization, the creditors secure an improved guarantee of repayment, thus reducing the asymmetric information costs of issuing debt ( Titiman and Wessels , 1988 ). Harris and Raviv ( 1990 ) argue that tangible assets have higher value on liquidation implying that liquidation is often the optimal strategy when the firm is financially distressed. However , when liquidation happens to be the best course of action, managers, due to considerations of self-interest, tend to suppress relevant information necessary for liquidation. In such a situation, debt can ensure availability of information because default on debt obligations triggers investigation into the operations of the firm. Hence, tangibility may be expected to positively related to leverage. ( ii ) Following Grossman and Hart ( 1982 ) who suggest that higher debt levels diminish the tendency of managers to consume more than the optimal level of perquisites because of the increased threat of bankruptcy, Titman and Wessels ( 1988 ) argue that the agency costs of managers consuming more than the optimal level of perquisites increases for firms with less collateralizable assets because shareholder monitoring costs of capital outlays of such firms will probably be higher than those that have more collateralizable assets. So,
102
firms with less collateralizable assets may choose higher levels of debt in order to restrict their managers’ consumption of perquisites ( Ibid. ). Hence , tangibility and leverage may be said to be negatively related. ( C ) POT : Frank and Goyal ( 2009a ) argue that tangible assets having low information asymmetry renders less costly issuance of equity shares , thus implying that firms with higher tangibility should have lower leverage ratios. Hence , tangibility and leverage may be expected to be negatively related. ( 3 ) Non-Debt Tax Shields ( NDTS ) ( A ) TOT : The existence of non-debt tax shields ( such as depreciation and investment tax credits ) provides an alternative and relatively cheaper means of reducing the burden of income tax thus reducing the interest tax shield on debt ( Cloyd et al. , 1997 ). DeAngelo and Masulis ( 1980 ) argue that at higher levels of leverage, ceteris paribus, the marginal savings from an additional unit of debt declines as non-debt tax shields increase because of the increased probability that the potential debt tax shields ( that is , interest tax shields ) will be partially or fully lost through bankruptcy. Hence NDTS may be expected to be negatively related to leverage. ( 4 ) Profitability ( A ) TOT : A firm with higher profitability will have greater tax advantages of using debt and lesser probability of its failure to pay interest on debt. So, profitable firms, whose expected bankruptcy costs are lower and expected tax shields are higher, would employ more debt. Moreover, profitable firms enjoy better borrowing terms as lenders prefer such firms. If future profitability can be suitably proxied by past profitability , then profitable firms would tend to borrow more due to greater probability of repayment of loans ( Gaud et al. , 2005 ). Hence , profitability is expected to be positively related to leverage. ( B ) ACT : Profitable firms tend to employ debt to prevent wasteful activities by managers. For firms with higher free cash flows ( and hence higher profitability ) , high level of debt may be said to act as a constraint for managerial discretion ( Jensen , 1986 ). The higher the free
103
cash flows of the firm and the agency costs of equity, a higher level of debt should be used to discipline managerial behaviour ( Huang and Song , 2006 ). Hence , profitability and leverage are expected to be positively related. ( C ) POT : Donaldson ( 1961 ) opines that firms will prefer to raise capital from retained earnings , then from debt and finally from issuing new equity shares on the basis of the increasing level of transaction costs of these sources of financing. Similar conclusions are reached by Myers ( 1984 ) and Myers and Majluf ( 1984 ) while explaining corporate financing decisions in the presence of asymmetric information. Firms with large retained earnings will tend to have less debt because the more profitable the firm, the greater the availability of internal capital and the less the requirement for funds from external sources ( Myers , 1984 ). Hence, leverage and profitability may be expected to be negatively related. ( 5 ) Growth Opportunities ( A ) TOT : The bankruptcy costs will be greater for firms with larger growth opportunities. These kind of firms may thus be reluctant to take on large amounts of debt so that bankruptcy probability is not increased ( Ibid.). Hence, leverage and growth opportunities may be expected to be negatively related. ( B ) ACT : Firms with growth opportunities may find it difficult and costly to rely on debt for financing , as the degree of risk may be high for growth oriented investments. Growth opportunities may be regarded as real options with associated agency costs, thus making it difficult for a firm to borrow against them than against tangible fixed assets ( Myers , 1977 ). Hence, leverage and growth opportunities will be negatively related. ( C ) POT : A firm with valuable investment opportunities is expected to rely on debt to finance such investments in order to maintain its debt-equity ratio as its equity increases due to large retention of earnings ( Myers , 1984 ; Myers and Majluf , 1984 ). Thus leverage and growth opportunities are expected to have a positive relationship.
104
( 6 ) Bankruptcy Risk ( Distance from Bankruptcy ) ( A ) TOT : Financially healthy companies , with low probability of bankruptcy and high distance from bankruptcy , tend to have low levels of debt. Evidence is found by Byoun ( 2008 ) that greater the Altman Z score ( which is used as a proxy for the distance from bankruptcy ), the lower the degree of leverage ( Kayo and Kimura, 2011 ). Mackie Mason ( 1990 ) opines that firms with lower Altman Z score are expected to be less financially secure and therefore should be closer to facing zero tax rate. Hence , distance from bankruptcy may be expected to be negatively related to leverage. ( B ) POT : Higher probability of bankruptcy will lead potential investors to require a higher rate of return, making it more expensive to issue equity due to more information asymmetries ; so risky firms, with higher probability of bankruptcy and thus with lower distance from bankruptcy, will tend to have higher levels of debt. Hence, distance from bankruptcy and leverage are expected to be negatively related. ( 7 ) Liquidity ( A ) TOT : Firms with higher liquidity ratios, having greater ability to meet outstanding short-term obligations tend to support a comparatively higher debt ratio ( Ozkan , 2001 ). Hence , leverage may be expected to be positively related to liquidity. ( B ) ACT : Firms with substantial free cash flows tend to employ more debt because of the disciplining benefit of debt which leaves the managers with little free cash flows to squander , after meeting the debt servicing obligation ( Jensen , 1986 ). Hence , leverage is expected to be positively related to liquidity. ( C ) POT : Firms with high level of liquid assets, being in the possession of more internal funds, tend to borrow less. Firms preferring internal sources of finance tend to create liquid reserves from retained earnings in order to finance future investments thus reducing their need for external funds ( Myers and Majluf , 1984 ). Hence , leverage and liquidity are expected to be negatively related. The following table summarizes the theoretical implications of the selected determinants on TOT, ACT and POT along with some instances of past empirical studies in which the relationship between such determinants and the dependent variable measuring leverage was found to be significant at conventional levels of statistical significance :
105
Table 4.1 : Theoretical Implications of Capital Structure Determinants Determinants
Firm Size
Tangibility
Non-Debt Tax Shield
Study
Relationship
Theory
( A ) International Studies Taub (1975), Marsh (1982), Harris & Raviv (1990, 1991), Chiarella et al.(1992), Rajan & Zingales (1995), Wiwatanakantang (1999), Hirota (1999) , Ozkan (2001), Booth et al. (2001), Pandey (2001), Bevan & Danbolt (2002,2004), Voulgaris et al.(2004), Deesomsak et al. (2004), Fattouh et al.(2005), Gaud et al.(2005), Eldomiaty et al.(2005), Huang & Song (2006), Delcoure (2007), de Jong et al.(2008), Antoniou et al.(2008), Fattouh et al.(2008), Psillaki & Daskalakis (2009) , Frank & Goyal (2009a) , Sheik & Wang (2011), Kayo & Kimura (2011), Bahng & Jeong (2012), Lemma & Nagesh (2013), Bayrakdaroğlu et al.(2013), Berzkalne & Zelgalve (2014) ( B ) Indian Studies Rao (1989), Booth et al. (2001), Bhaduri (2002), Sahoo & Omkarnath (2005), Narender & Sharma (2006), Maji & Ghosh (2007) , Pathak (2010) , Mahakud & Mukherjee (2011), Ray (2013) , Basu & Rajeev (2013)
Positive
TOT & ACT
( A ) International Studies Chudson (1945), Gupta (1969), Kester (1986), Titman & Wessels (1988), Rajan & Zingales (1995) [ only for Germany ], Fattouh et al. (2005), Fattouh et al. (2008), Icke & Ivgen (2011), Bahng & Jeong (2012) ( B ) Indian Studies Bhaduri(2002), Guha-Khasnobis & Bhaduri (2002), Saravanan(2006), Mahakud & Mukherje (2011), Handoo & Sharma (2014)
Negative
( A ) International Studies Rajan & Zingales (1995), Hirota(1999), Bevan & Danbolt (2002, 2004), Voulgaris et al. (2004), Deesomsak et al. (2004), Huang & Song (2006), Delcoure (2007), Frank & Goyal (2009a), Kayo & Kimura (2011) ( B ) Indian Studies Booth et al. (2001), Guha-Khasnobis & Bhaduri (2002), Sahoo & Omkarnath (2005), Narender & Sharma (2006), Pathak (2010), Rajagopal (2010), Mahakud & Mukherjee (2011), Majumdar (2012), Ray (2013), Basu & Rajeev (2013) , Handoo & Sharma (2014)
Positive
TOT & ACT
( A ) International Studies Pandey (2001), Eldomiaty et al. (2007), Psillaki & Daskalakis (2009), Sheik & Wang (2011) ( B ) Indian Studies -
Negative
POT & ACT
( A ) International Studies Delcoure ( 2007 ) ( B ) Indian Studies Kaur & Rao (2009) , Naha & Roy (2011), Mahakud & Mukherjee (2011), Ray (2013)
Positive
-
( A ) International Studies Wiwattanakantang (1999), Deesomsak et al. (2004), Huang & Song (2006), Frank & Goyal (2009a)
Negative
TOT
POT
106
Table 4.1 : Theoretical Implications of Capital Structure Determinants Determinants
Study
Relationship
Theory
( A ) International Studies ( B ) Indian Studies Kaur & Rao (2009)
Positive
TOT & ACT
( A ) International Studies Chudson ( 1945 ), Titman & Wessels ( 1988 ), Rajan & Zingales ( 1995 ), Wiwattanakantang ( 1999 ), Hirota ( 1999 ), Pandey ( 2001 ), Booth et al. ( 2001 ), Bevan & Danbolt ( 2002, 2004 ), Voulgaris et al. 2004 ), Deesomsak et al. ( 2004 ), Huang & Song ( 2006 ), Eldomiaty et al. ( 2007 ), Delcoure ( 2007 ), Frank & Goyal ( 2009a ), Psillaki & Daskalakis ( 2009 ), Sheik & Wang ( 2011 ), Kayo & Kimura ( 2011 ), Icke & Ivgen ( 2011 ) ( B ) Indian Studies Chakraborty (1977) , Rao (1989) , Kakani & Reddy (1999 ), Majumdar & Chibber (1999) , Booth et al. ( 2001), Guha-Khasnobis & Bhaduri (2002), Narender & Sharma (2006), Maji & Ghosh (2007), Pathak (2010), Rajagopal (2010) , Mahakud & Mukherjee (2011), Majumdar (2012),Ray (2013), Basu & Rajeev (2013), Handoo & Sharma (2014)
Negative
POT
( A ) International Studies Gupta (1969), Titman & Wessels (1988), Booth et al. (2001), Eldomiaty et al. (2005) ( B ) Indian Studies Booth et.al (2001), Bhaduri (2002), Saravanan (2006), Pathak (2010), Naha & Roy (2011), Mahakud & Mukherjee (2011),Majumdar (2012), Basu & Rajeev (2013), Handoo & Sharma (2014)
Positive
POT
( B ) Indian Studies Sahoo & Omkarnath (2005), Kakani & Reddy ( 1999 ), Rajagopal (2010)
Profitability
Growth Opportunities
( A ) International Studies Rajan & Zingales (1995), Wiwattanakantang (1999), Hirota (1999), Huang & Song (2006) , Frank & Goyal (2009a), Kayo & Kimura (2011) ( B ) Indian Studies Kaur & Rao (2009), Rajagopal (2010) , Ray (2013)
Negative
TOT & ACT
( A ) International Studies Deesomsak et.al al. ( 2004 ) ( B ) Indian Studies Booth et.al (2001) , Saravanan ( 2006) , Kaur & Rao (2009)
Positive
-
( A ) International Studies Eldomiaty et.al (2007), Delcoure (2007), Psillaki & Daskalakis (2009) ( B ) Indian Studies Kakani & Reddy (1999), Pathak (2010)
Negative
TOT & POT
Bankruptcy Risk
107
Table 4.1 : Theoretical Implications of Capital Structure Determinants Determinants
Liquidity
4.4
Study
Relationship
Theory
( A ) International Studies Martin & Scott (1974), Feidakis & Rovolis (2007) ( B ) Indian Studies Narender & Sharma (2006), Kaur & Rao (2009), Basu & Rajeev (2013)
Positive
TOT & ACT
( A ) International Studies Voulgaris et.al (2004), Deesomsak et.al (2004), de Jong et.al ( 2008 ), Sbeiti (2010), Sheik & Wang (2011), Icke & Ivgen (2011) ( B ) Indian Studies Pathak (2010), Mahakud & Mukherjee (2011), Rasoolpur (2012)
Negative
POT
Formulation of Variables
The formulation of the various measures of the dependent and the independent variables are explained below.
4.4.1
Dependent Variable
The capital structural measure of financial leverage is the dependent variable in this study. The following four different empirical measures of financial leverage have been suggested by Rajan and Zingales ( 1995 ) : ➢ The ratio of total ( non-equity ) liabilities to total assets ➢ The ratio of short-and long-term debt to total assets ➢ The ratio of debt to net assets, where net assets are total assets less accounts payable and other current liabilities ➢ The ratio of total debt to capital concluding that the effects of past financing decisions is probably best represented by the ratio of total debt to capital. Deesomsak ( 2004 ) also adopts the debt to capital ratio as the measure of leverage. Welch ( 2007 ) suggests that the focus of researchers should be either on debt scaled by capital or total liabilities scaled by total assets. Moreover , in many studies 21 , debt ratio has been expressed both in terms of book value and market value arguing that :
21
Rajan and Zingales ( 1995), Wiwattanakantang ( 1999), Booth ( 2001), Almazan ( 2002 ), Gaud et al. ( 2005), Mckay and Philips ( 2005 ), Kayo and Kimura ( 2011 ) , Elsas et al. ( 2013 ) to name a few.
108
➢ Book values of leverage are more closely related to a firm’s assets-in-place than to a firm’s growth opportunities ( Myers , 1977 ). Market values incorporate the present values of future growth opportunities and debt issued against these values can distort future real investment decisions ( Shyam-Sunder and Myers , 1999 ). ➢ Book values are insensitive to changes in share price. So managers, when setting their financial structure, tend to focus on book values rather than market values as and this does not reflect on the rebalancing process of capital structure ( Graham and Harvey , 2001 )
➢ The theory of capital structure suggests that debt ratios should be measured in market value terms ( Marsh , 1982 ) ➢ Market values are closer to the intrinsic firm value and tend to reflect the potential of future leverage precisely. Thus, the measurement of market leverage is more realistic compared to book leverage ( Kayo and Kimura , 2011 ) Two measures of the dependent variable [ namely , Market Leverage ( MLEV ) and Book Leverage ( BLEV ) ] , based on the debt to capital ratio in terms of ( quasi ) market value and book value , have been chosen in this study : Table 4.2 : Formulation of Dependent Variable ( s ) Dependent Variable
Formulation Book Value of Debt Book Value of Debt + Net Worth + Book value of Preference Shares
BLEV
MLEV
where, Net Worth ( NW ) [ as defined in CMIE Prowess database ] = Paid up equity capital + Capital contribution or funds by govt., others + Equity Share application money or advances + Equity Capital suspense & other account + Convertible warrants + Reserves ̶ Revaluation reserves ̶ Miscellaneous expenses not written off. Book Value of Debt Book Value of Debt + Market value of equity shares + Book value of Preference Shares where , Market value of equity shares = Market value per share * Number of shares outstanding
Note : Debt includes long-term and short-term debt excluding current liabilities and provisions.
109
4.4.2
Independent Variables
The following determinants of capital structure , as mentioned in section 3.4 , have been chosen for the study :( i ) firm size , ( ii ) tangibility , ( iii ) non-debt tax shield , ( iv ) profitability , ( v ) growth , ( vi ) bankruptcy risk ( distance from bankruptcy ), and ( vii ) liquidity. Industry and time dummy variables are also included as control variables to account for the industrial classification and macroeconomic factors respectively. The formulation of these variables are enumerated in the following table.
Table 4.3 : Formulation of Independent Variables Variable
Formulation
References
Natural logarithm of Net Sales
Titman &Wessels (1988), Rajan & Zingales (1995) , Wiwattanakantang (1999), Ozkan (2001) , Booth et al.(2001) , Pandey (2001) , de Jong et al. (2008) , Antonoiu et al. (2008) , Chakraborty (2010), Kayo & Kimura (2011)
( 2 ) Tangibility ( TANG )
Net Fixed Assets ---------------------------------Total Assets ( TA )
Harris & Raviv (1991), Rajan & Zingales (1995) , Booth et al. (2001) , Bevan & Danbolt (2004) , Huang & Song (2006) , Shah & Khan (2007) , Kayo & Kimura (2011) , Handoo & Sharma (2014) , Öz-tekin (2015)
( 3 ) Non-Debt Tax Shields ( NDTS )
Depreciation & Amortizations ------------------------------------------TA
Titman & Wessels (1988) , Wiwattanakantang(1999), Chen (2004), Deesomsak et al.(2004), Huang & Song (2006), Delcoure( 2007), Chakraborty (2010)
( 4 ) Profitability ( PROF )
( EBDAIT / TA ) EBDAIT = Earnings Before Depreciation, Amortisation, Interest and Tax
Bevan & Danbolt (2004) , Deesomsak et al. (2004) , Chakraborty (2010) , de Jong et al.(2008) , Fattouh (2008) , Frank & Goyal (2015)
( 5 ) Growth Opportunities ( GROW )
MV of equity shares + ( BV of Total Assets BV of Net Worth ) BV of Total Assets ( TA )
Deesomsak et al.(2004),Huang & Song (2006), Eldomaity (2008), Antoniou et al.(2008 ), de Jong et al. (2008), Kayo & Kimura (2011), Frank & Goyal ( 2015 )
( 6 ) Bankruptcy Risk [ Distance from Bankruptcy ( DFB ) ]
Altman’s Z score ( 1968 ) as modified by MacKie-Mason ( 1990 ) : Z = 3.3 ( EBIT / TA ) + 1.0 ( Sales / TA ) + 1.4 ( Retained earnings / TA ) + 1.2 ( Working capital / TA )
MacKie-Mason (1990),Graham (2000 ), Byoun (2008), Kayo & Kimura (2011)
( 1 ) Firm Size ( SIZE )
110
Table 4.3 : Formulation of Independent Variables Variable
Formulation
References
( 7 ) Liquidity ( LIQ )
Current Ratio = ( Current Assets / Current Liabilities )
Ozkan (2001), Deesomsak et al. (2004), Eldomaity (2008), de Jong et al. (2008), Sbeiti (2010) ,Mat Nor et al.(2011), Handoo & Sharma ( 2014 )
Control Variables
Industry Dummy Variable ( IND ) Time Dummy Variable ( Year )
4.5 4.5.1
INDj = 1 when the company belongs to industry j , = 0 otherwise.
Scott (1972), Scott & Martin (1975), Ferri & Jones (1979) , Titman & Wessels (1988 ), Graham & Harvey (2001), Eldomaity (2008), Su (2015)
YEARt = 1 when the year is t , = 0 otherwise.
Eldomaity(2008), Su(2015),Eun &Wang ( 2016 )
Data and Sample Selection Type of Data
The data to be analysed has been selected as that of ‘micro panel data’ type with large N ( number of cross-sections ) and small T ( length of the time series ) . Panel data ( or longitudinal data ) consist of repeated observations on the same cross section ( for e.g. individuals , households , firms , or cities ) over time ( Wooldridge , 2010 ). The advantages of panel data over cross-sectional and time-series data , as listed by Klevmarken ( 1989, pp. 523-29 ) , Hsiao ( 2003 , pp. 3-7 ) , Baltagi ( 2005 , pp. 4-7 ) and Gujarati ( 2003 , pp. 637-38 ) may be summarised as follows : ( 1 ) Panel data is able to control for individual heterogeneity by suggesting that individuals, firms, states or countries are heterogeneous. Time-series and cross-section studies which are not able to control this heterogeneity run the risk of obtaining biased results. ( 2 ) Panel data give more informative data, more variability, less collinearity among the variables, more degrees of freedom and more efficiency by combining time series of crosssectional observations. ( 3 ) Panel data are better able to study the dynamics of adjustment. Cross-sectional distributions that look relatively stable hide a multitude of changes. ( 4 ) Panel data are better able to identify and measure effects that are simply not detectable in pure cross-section or pure time-series data. ( 5 ) Panel data models allow the researcher to construct and test more complicated behavioural models than purely cross-section or time-series data.
111
( 6 ) Micro panel data gathered on individuals , firms and households may be more accurately measured than similar variables measured at the macro level and biases resulting from aggregation over firms or individuals may be reduced or eliminated.
4.5.2
Source of Data
The data is obtained from the secondary source of CMIE Prowess IQ database which is a powerful internet-based application for querying CMIE’s database on performances of listed and unlisted Indian companies.
4.5.3
Selection of Sample
The final sample include 601 Indian manufacturing companies listed on the Bombay Stock Exchange ( BSE ) comprising the eleven broad industrial divisions ( as classified in the Prowess IQ database ) over a period of 18 years from 1997-98 to 2014-15. Since, panel data model will be applied to the analysis of capital structure which is supposed to be a long-term phenomenon, a balance between the ‘length of the period of time’ and the ‘number of cross-sectional units’ has been maintained, so that the sample includes a longer time period with a sizeable number of cross-sectional units. The sample selection has been conducted on the basis of multistage stratified cluster sampling based on certain inclusion and exclusion criteria as shown below : Table 4.4 : Procedure of Sample Selection Sl. No. 1. 2.
3.
Particulars Indian manufacturing companies ( super population ) Less : Indian manufacturing companies which are either unlisted or not listed on the BSE Indian manufacturing companies listed on BSE and comprising all of the eleven broad industrial divisions ( as classified in the CMIE Prowess IQ database ), namely, Consumer Goods, Chemicals & Chemical Products, Construction Materials, Diversified, Drugs & Pharmaceuticals, Food & Agro-based Products, Machinery, Metals & Metal Products, Miscellaneous Manufacturing, Textiles and Transport Equipment [ population considered for this study ]
Number of companies 10981 8375
2606
112
Sl. No.
Particulars
Number of companies
4.
Less: Companies not fulfilling all of the following criteria : ( a ) having positive net worth and positive sales ; ( b ) incorporated before the first period (1997-98) of the study ; and ( c ) maintaining their identities and reporting annual financial statements continuously for the entire period of the study without any missing figures for the variables under study
2005
5.
Final sample of Indian manufacturing companies listed on the BSE, falling under all of the eleven broad industrial divisions and fulfilling all the criteria mentioned in ( 4 ) above
601
The following table shows the classification of companies in the final sample of 601 companies on the basis of the manufacturing divisions to which they belong : Table 4.5 : Classification of Companies in Final Sample as per Industrial Division Sl. No.
Industrial Division ( code used )
Number of companies in the sample
Percentage of companies in the sample
1
Consumer Goods ( CG )
25
4.16%
2
Chemicals and Chemical Products ( CCP )
126
20.97%
3
Construction Materials ( CM )
32
5.32%
4
Diversified ( D )
8
1.33%
5
Drugs and Pharmaceuticals ( DP )
39
6.49%
6
Food and Agro-based Products ( FAP )
66
10.98%
7
Machinery ( M )
66
10.98%
8
Miscellaneous Manufacturing ( MM )
28
4.66%
9
Metals and Metal Products ( MMP )
63
10.48%
10
Textiles ( T )
89
14.81%
11
Transport Equipment ( TE )
59
9.82%
Total
601
100.00%
It is observed from the above table that corporate firms belonging to the Chemicals and Chemical Products ( CCP ), Textiles ( T ), Food and Agro-based Products ( FAP ), Machinery ( M ), and Metals and Metal Products ( MMP ) divisions constitute 68.22% of the final
113
sample of companies, with CCP division registering the highest percentage followed by T FAP, M and MMP. The other six manufacturing divisions constitute the remaining 31.78% of the final sample of companies with Diversified ( D ) division registering the least percentage among all the manufacturing divisions.
4.6
Techniques for Preliminary Analyses of Data
The preliminary analyses of data will be conducted based on the following statistical and econometric techniques applying Stata®/SE 14.1 , Eviews® 9 Enterprise Edition and Microsoft® Excel 2016 softwares :-
( 1 ) Descriptive summary of the basic statistical measures of minimum, maximum, mean, median, standard deviation and coefficient of variation of the variables for the pooled data as well as for the individual time periods , along with relevant charts. These measures are enumerated below : ( a ) Minimum : The lowest value of a variable from a set of values. ( b ) Maximum : The highest value of a variable from a set of values. ( c ) Mean : Mean or Arithmetic Mean ( to be precise ) is a measure of central tendency regarded as a representative or typical value of the entire frequency distribution. In other words, mean is a single figure which represents the whole series of observations with their varying sizes, lies between the smallest and the largest observations, and is defined as the sum of the observations divided by the number of observations. For a frequency distribution, it is given by : Mean = ( f x / N )
(3)
where x = random variable ; f = frequency, and N = f = total frequency. ( d ) Median : The median, a positional measure of central tendency, is the value of the middle-most term in a set of observations arranged in order of magnitude so that half of the observations in the data set lie on either side of it. The median for grouped frequency distribution is given by : Median = l1 + { ( N / 2 – FC ) / fm } * d where l1 = lower boundary of median class, N = total frequency, FC = Cumulative frequency below l1, fm = frequency of median class, d = width of the median class.
(4)
114
( e ) Standard deviation : It is defined as the positive square root of the arithmetic mean of the squares of deviations of the values of a variable from their arithmetic mean. It is usually denoted by σ. For a simple series : σ=√
∑ 𝑥2 𝑛
∑𝑥 2
− (
𝑛
)
(5)
where x = variable, n = number of observations. For a frequency distribution : σ=√
∑ 𝑓𝑥 2 𝑁
∑ 𝑓𝑥 2
− (
𝑁
)
(6)
where f = frequency , N = sum of the frequencies.
( 2 ) Correlation analysis : Pairwise correlation analysis of the variables will be conducted by the application of linear correlation coefficient which is a statistical parameter used to define the strength and nature of the linear relationship between two variables or attributes. Karl Pearson’s product moment correlation coefficient between two variables x and y is given by : r x, y = { cov ( x, y ) / x * y }
(7)
where cov ( x, y ) = covariance of x and y ; x and y are the standard deviations of x and y respectively. Evans ( 1996 ) suggests the following degree or strength for the absolute value of r : • 0.0 -0.19 “very weak” • 0.20 -0.39 “weak” • 0.40 -0.59 “moderate” • 0.60 -0.79 “strong” • 0.80 -1.0 “very strong”.
( 3 ) Linear Trend Analysis of Variables : Considering that our data is of ‘panel data’ type with cross-sectional unit ( that is , firm ) i ( i = 1,2, 3,…,N ) observed over a period of t years ( t = 1,2,3,…,T ) , the cross-sectional mean of a variable ( averaged over the crosssectional units ) for each ‘time period’ , is regressed against ‘time period’ t , in an Ordinary Least Squares ( OLS ) linear regression framework , to yield the linear trend line which helps in depicting any significant patterns or trends in the data over the sample period. A positive (negative) coefficient on the ‘time period’ variable indicates increasing ( decreasing ) trend experienced by the variable on an average. ̅ t = [ ( 1 / N ) ∑N Let Y i=1 Yi t ] , the cross-sectional mean of a variable Y for each time period t , be assumed to be linearly related to t . The OLS regression model is given by :
115
̅t = + * t + t Y
(8)
where is the intercept ; is the regression coefficient of t ; and t is the stochastic error term
assumed to be identically and independently distributed following Normal
distribution with mean zero and standard deviation ( σε ). The null hypothesis of ‘no linear trend’ ( or , = 0 ) is tested against the alternative hypothesis of ‘linear trend’ ( or , 0 ). This linear trend analysis of the variables ( independent and dependent ) will augment the panel data unit root tests for these variables explained below.
( 4 ) Test for Stationarity of Variables : A times series is said to be ( weakly ) stationary if the mean and autocovariances of the series are independent of time. In case of panel data, the cross-sectional units generate multiple ‘time series’ . Let us consider the following first-order autoregressive process , AR(1) : yit = it + i yit-1 + it
(9)
where i = 1,2,3,…,N cross-sectional units ( or panels) observed over t =1,2,3,..,T time perods, it represents the deterministic part of the model which may include panel-specific intercepts ( fixed effects ) , a panel-specific time trend , or nothing ( in which case yit is predicated to have a zero mean for all panels ) , i are the first-order autoregressive coefficients , and it is the zero -mean stochastic error term assumed to be mutually independent across panels. If i < 1 , yi is said to be ( weakly ) stationary at level. However, if i 1 , then yi contains an unit root and is said to be non-stationary at level. The following panel data unit root tests will be applied in this study : ( A ) Levin-Lin-Chu ( LLC ) ( 2002 ) Test
22
: This test assumes that all cross-sectional
units have the same autoregressive coefficient , that is , i = for all i . The null hypothesis is 1 ( that is , each individual time series contain a unit root ) , and the alternative hypothesis is < 1 ( that is , each time series is stationary ). This test consists of three steps as follows : ( i ) performing separate Augmented Dickey Fuller ( ADF ) regressions for each cross-section , ( ii ) estimating the ratio of long-run to short-run standard deviations and (iii) computing the adjusted t-statistic which is asymptotically normally distributed with zero mean and variance of one. The test requires : ( a ) the panel data to be strongly balanced, ( b ) the specification of the number of lags used in each cross-sectional ADF regression, and ( c ) the choices of kernel used in the computation of average standard deviation. Moreover , the exogenous variables used in the test equations may include no
22
Details in Balatagi ( 2009 , p. 240 ) .
116
exogenous regressors , individual constant terms ( fixed effects ) or linear trends or both. LLC recommend using the test for panels of moderate size with N between 10 and 250 and T between 25 and 250 , arguing that the standard panel procedures may not be computationally feasible or sufficiently powerful for panels of this size. If T is very large , standard unit-root tests can be applied to each panel , owing to small expected gains from aggregation. However , for very large N and very small T, usual panel data procedures are recommended . ( B ) Im-Pesaran-Shin ( IPS ) ( 2003 ) Test 23 : A major limitation of the LLC test is the assumption that all panels have the same value of . This assumption is relaxed by the IPS test which allows separate i for each panel. The null hypothesis is H0 : i = 0 for all i ( that is , all panels have a unit root ) and the alternative hypothesis H1 : i < 0 for i = 1,…, N1 and i = 0 for i = N1 + 1 ,…, N with 0 < N1 < N ( that is , the proportion of stationary panels is non-zero ) where N1 denotes the number of stationary panels. As N tends to infinity , the proportion ( N1 / N ) tends to a non-zero fraction , thus allowing some of the panels to possess unit roots under the alternative hypothesis. W t-bar statistic , the standardised version of the IPS test statistic , t-bar ( 𝑡̅ ) , is reported.
23
Deails in Baltagi , op. cit., p. 242.
117
4.7 4.7.1 4.7.1.1
Empirical Framework Panel Data-Least Squares ( LS ) Models 24 Formulation
The study seeks to apply Quantile Regression ( QR ) methodology to the sample data. However, as the basic structure of the data for this study is that of a panel data -type, the baseline econometric model will be a Panel Data-Least Squares ( LS ) regression model which will be chosen from the Pooled Ordinary LS model , Fixed Effects model and Random Effects model after performing the appropriate tests , as discussed in the next subsection ( 4.7.2 ) , will then be applied to the chosen panel data model. Based on the final sample ( Section 4.5.3 ), let us consider a balanced panel data structure consisting of N (= 601 ) firms belonging to M (= 11 manufacturing industrial divisions ) [ all the firms belonging to the same respective industry for the entire time period ], each firm being observed over a period of T (=18 ) years. This data structure with NT (= 10818 ) number of observations is of the large N small T-type ( T < N ). Considering that data is collected at the firm level, the panel data LS models for firm i ( i = 1,2, 3,…,N ) belonging to an industrial division j ( j = 1,2,3,…,M ) and observed over a period of t years ( t = 1,2,3,…,T ) , are given below : Model 1 : Panel Data – Least Squares ( LS ) Model 𝑀 𝑇 Yi j t = + ∑𝐾 𝑘=1 βk * Xi j t ( k ) + ∑𝑡=2 γt * Dt + ∑𝑗=2 δj * IND j + ηi + ui j t
( 10 )
where Yi j t = dependent variable ( MLEV or BLEV ) = intercept ; Xi j t ( k ) = K [ = 7 ( seven ) ] time-varying independent variables SIZE, TANG, NDTS, PROF, GROW, DFB, and LIQ ; β = regression coefficients of the Xi j t variables ; Dt = 17 ( seventeen ) 25 time-varying time dummy variables for t ( = 2,3,…,T) to account for the macro-economic and institutional country-specific factors ; γ = regression coefficients of the Dt variables ;
24
This section is based on the discussions in Schmidheiny ( 2016 ) , Cameron and Trivedi ( 2005 ) , Wooldridge ( 2009 ) , Baltagi ( 2005 ) and Greene ( 2010 ). ( T – 1 ) or ( 18 – 1 ) time dummy variables to avoid falling into the dummy variable trap of perfect collinearity as intercept ( ) has been included. 25
118
IND j =10 ( ten ) 26 time-invariant 27 industry dummy variables [ equal to one , if the ith firm belongs to industrial division j ( = 2, 3,…,M ) and equal to zero , otherwise ] representing the industry classifications of the firms ; δ = regression coefficients of the IND j variables ; ηi = time-invariant unobserved and heterogenous firm-specific effects ; ui j t = stochastic error terms. The ‘j’ subscript will be ignored from X and u in the following discussions in order to
avoid unwarranted complexities.
4.7.1.2
General Assumptions
( G1 ) Linearity E ( ui t ) = 0 , E ( ηi ) = 0. The model is linear in parameters , , γ , δ , effects ηi and error term ui t .
( G2 ) Strict Exogeneity E ( ui t | Xi 1 (1)…Xi T (K) , D2…DT , IND2…INDM , ηi ) = 0. The model is strictly exogenous , that is , the respective error terms are uncorrelated with the explanatory variables ( of past , present and future time periods ) , control variables and the firm-specific effects.
( G3 ) Independence Model 1 : { Xi1(1)…Xi T (K) , D2…DT , IND2…INDM , Yi 1…Yi T } N i=1 ( i.i.d , that is , independently and identically distributed ). The observations are assumed to be independent across firms but not necessarily across time.
( G4 ) Variance of Error ( a ) Homoscedastic and no serial correlation ( i ) Homoscedastic : V ( ui t | Xi1(1)…Xi T(K) , D2…DT , IND2…INDM , ηi ) = σ2u > 0 and < . ( ii ) No serial correlation : Cov ( ui t , ui s | Xi1(1)…Xi T (K) , D…DT , IND2…INDM , ηi ) = 0 for all s t.
( M – 1 ) or ( 11 – 1 ) industry dummy variables to avoid falling into the dummy variable trap of perfect collinearity as intercept ( ) has been included. 26
27
Firms do not change industrial divisions over time.
119
( b ) Heteroscedastic V ( ui t | Xi1(1)…Xi T (K) , D2…DT , IND2…INDM , η i ) = σ2u,it > 0 and < . ( c ) Serial correlation of unknown form ̶ 1 < Corr ( ui t , ui s | Xi1(1)…Xi T (K) , D2…DT , IND2…INDM , ηi ) < 1 for all s t.
4.7.1.3 Specific Assumptions and Estimation of Random Effects ( RE ) , Fixed Effects ( FE ) and Pooled Ordinary Least Squares ( POLS ) Models ( I ) Random Effects ( RE ) Model ( 1 ) Assumptions : ( RE1 ) ηi 0. ( RE2 ) Uncorrelated Effects
E ( ηi | Xi1(1)…Xi T (K) , D2…DT , IND2…INDM ) = 0 The firm -specific effect is a random variable that is uncorrelated with the time -varying and time -invariant explanatory variables. ( RE3 ) Variance of Effect
( i ) Homoscedastic : V ( ηi | Xi1 (1)…Xi T (K) , D2…DT , IND2…INDM ) = σ2η > 0 and < . ( ii ) Heteroscedastic : V ( ηi | Xi1(1)…Xi T (K) , D2…DT , IND2…INDM ) = σ2η,i > 0 and < . ( RE4 ) Identifiability
( , Xi1(1)…Xi T (K) , D2…DT , IND2…INDM ) are not linearly dependent and 0 < V ( Xi t (k) ) < and 0 < V̂ ( Xi t (k) ) < . The regressors including the constant () are not perfectly collinear and all time-varying regressors have non-zero and finite variances and not too many extreme values. ( RE5 ) Error Variance in Equicorrelated Random Effects Model
Putting vi t = ηi + ui t in equation ( 10 ) , we get from assumptions G3 , G4 (a) , RE2 and RE3 (a) : ( i ) Homoscedastic : V ( vi t | Xi1(1)…X i T (K) , D2…DT , IND2…INDM ) = σ2v = σ2 + σ2u for all i, t. ( ii ) No cross-sectional serial correlation : Cov ( vi t ,vp s | Xi1(1)…Xi T (K) , Xp1(1)…Xp T (K) , D2…DT , IND2…INDM ) = 0 for all s, t and i p. ( iii ) Presence of contemporaneous serial correlation : Corr ( vi t , vi s | Xi 1(1)…Xi T (K) , D2…DT , IND2…INDM ) = (σ2 / σ2v ) for all i and s t.
120
( 2 ) Estimation of Random Effects Model The Random Effects ( RE ) estimator of ( ̂ RE , say ) , which is the Feasible Generalized Least Squares ( FGLS ) estimator , is obtained by the transformation of the dependent and the independent variables whereby the error terms in the transformed model are uncorrelated across all cross-sectional units ( N ) and all time periods ( T ) . ̂ RE is consistent and asymptotically normally distributed 28 under assumptions G1 to G4 , RE2 , RE3 and RE4 when N tends to infinity even if T is fixed , thereby allowing its approximation in samples with many cross-sectional units. However, the small sample properties for ̂ RE cannot be established . Consistent estimation of asymptotic variance of ̂ RE is possible assuming homoscedasticity and no serial correlation of the error term in the equicorrelated model ( RE5 ). Moreover, allowing for arbitrary conditional variances and for autocorrelation of the combined error, vit [assumptions G4(c) and RE3(b) ] , the asymptotic variance can be consistently estimated with the cluster-robust covariance estimator treating each cross-sectional unit as a cluster. In case of large samples , the usual tests ( z-, Wald-) can be performed in both cases ( Schmidheiny , 2016 , pp. 6-7).
( II ) Fixed Effects ( FE ) Model ( 1 ) Assumptions : ( FE1 ) ηi 0.
The presence of firm-specific unobserved effects representing individual heterogeneity is assumed to be captured by the intercept term ( ) , which means that every firm gets its own intercept ( i ) in the form of ηi , while the slope coefficients ( s ) are the same for all the firms.29 ( FE2 ) Correlated Effects
E ( ηi | Xi1(1)…Xi T( K) , D2…DT IND2…INDM ) 0. The firm-specific effect is a random variable that is assumed to be correlated with explanatory variables ( time-varying and time-invariant ). So, the time-invariant industry dummy variables in Model 1 will be perfectly collinear with ηi. Hence equation ( 10 ) [ Model 1 ] will reduce to the following FE model :
As Wooldridge ( 2009, p. 174) opines “ Even though the yi are not from a normal distribution, we can use the central limit theorem…to conclude that the OLS estimators satisfy asymptotic normality, which means they are approximately normally distributed in large enough sample sizes.” 28
29
Note that is not represented as [ i (k) ] in equation ( 10 ) .
121
Model 2 : Panel data -Fixed Effects ( FE ) Model 𝑇 Yi t = + ∑𝐾 𝑘=1 βk * Xi t ( k ) + ∑𝑡=2 γt * Dt + ηi + ui t
( 11 )
( FE3 ) Identifiability
Model 2 : ( Ẍi t (1)…Ẍ i t (K) ) are not linearly dependent and 0 < V ( Ẍi t (k ) ) < for all k , where Ẍi t (k) = ( Xi t (k) ̶ 𝑋̅i (k) ) and 𝑋̅i (k) = ( 1 / T ) ∑𝑇𝑡=1 Xi t ( k ) . The time-varying explanatory variables are not perfectly collinear, that they have non-zero within-variance ( that is , variation over time for a given individual ) and not too many extreme values. ( 2 ) Estimation of FE Model ( A ) FE Estimator or Within Estimator
The within estimator of the slope coefficients k and γt estimates the within model by OLS treating the firm-specific effects ( ηi ) as nuisance or incidental parameters that are not of intrinsic interest. Subtracting time averages ̅ Yi = ( 1 / T ) ∑Tt=1 Yi t , ̅ Xi k = ( 1 / T ) ∑Tt=1 Xi t k , and ̅ t = ( 1 / T ) ∑Tt=2 Dt from Yi t , Xi t k and Dt respectively , Model 2 [ equation ( 11 ) ] reduces D to : Ÿi t = ∑Kk=1 βk * Ẍi t (k) + ∑Tt=2 γt * D̈t + ü i t
( 12 )
̅i ) , Ẍi t k = ( Xi t k ̶ X ̅i k ) , D̈t = ( Dt ̶ D ̅ t ) and ü i t = ( ui t ̶ u̅i ) . where Ÿi t = ( Yi t ̶ Y The unobserved firm-specific effects ( ηi ) and the intercept ( ) cancel out. The fixed effects within estimator of ( ̂ FE , say ) is unbiased under assumptions of G1, G2, G3 and FE3. Under the assumption of normally distributed and homoscedastic errors with no serial correlation [ G4(a) ] , ̂ FE is said to be normally distributed in small samples with its variance being unbiasedly estimated applying the usual OLS estimator in the transformed model. Under assumptions G1 to G4 and FE3 , ̂ FE is consistent and asymptotically normally distributed when N tends to infinity even if T is fixed , thereby allowing its approximation in samples with large N. If heteroscedasticity and serial correlation of unknown form [ G4(c) ] are taken into consideration , the asymptotic variance of ̂ FE can be consistently estimated with cluster-robust covariance estimator treating each N as a cluster. The usual tests ( z-, Wald- ) for large samples can be performed ( Ibid. , p. 7 ).
122
( B ) Least Squares Dummy Variables ( LSDV ) Estimator
The LSDV estimator is estimated using pooled OLS model which includes a set of ( N-1) dummy variables ( as the intercept is included in the model to avoid the dummy variable trap of perfect collinearity ) and identify the cross-sectional units ( N ) , and thus an additional (N-1) parameters. The time-invariant industry dummy variables are dropped owing to their perfect collinearity with the ( N-1 ) dummy variables. The LSDV estimators of k , which are numerically similar to the within estimators discussed above , are consistent under the same set of assumptions. The LSDV estimators of the additional parameters for the ( N-1) dummy variables are , however , inconsistent in non-linear fixed effects models , when the number of parameters tends to infinity , due to the incidental parameters problem ( Ibid., pp. 9-10 ) , whereby the estimation of the common parameters are contaminated. ( C ) First Difference ( FD ) Estimator
The FD estimator of k and γt estimates the first difference model by OLS. Subtracting one period lagged values Yi, t-1 , X i, t-1 (k) and Dt-1 from Yi t , Xi t (k) and Dt respectively, Model 2 [ ( equation ( 11 ) ] reduces to : Ẏi t = ∑Kk=1 βk * Ẋi t (k) + ∑Tt=2 γt * Ḋt + u̇
it
( 13 )
where Ẏi t = ( Yi t ̶ Yi, t-1 ) , Ẋi t (k) = ( Xi t (k) ̶ Xi, t-1 (k) ) , Ḋt = ( Dt ̶ Dt-1 ) and u̇ i t = ( ui t ̶ ui, t-1 ). The firm-specific effects ( ηi ) and the intercept ( ) cancel out. The FD and FE estimators are numerically when T = 2. Under similar set of assumptions as the FE estimator , the FD estimator consistently estimates βk . The FE estimator is more efficient than the FD estimator when the error terms are not serially correlated , and the FE estimator is used more than the FD estimator as the FE model is generally stated , explicitly or implicitly , with serially uncorrelated errors ( Wooldridge , 2009 ).
( III ) Pooled Ordinary Least Squares ( POLS ) Model ( 1 ) Assumption : ( POLS1 ) ηi = 0.
( 2 ) Estimation of POLS Model The POLS estimator , ignoring the panel structure of the data , estimates , , γ and δ by regressing Yi t on , Xi1(1)…Xi T(K) , D2…DT and IND2…INDM . In small samples , the POLS estimator : ( a ) is unbiased under the assumptions G1, G2, G3, RE2 and RE3 ; and ( b ) is normally distributed , assuming G4 and normally distributed stochastic error and cross - sectional specific errors. In large samples with N tending towards infinity , it is consistent and asymptotically normally distributed under G1 to G4, RE2 and
123
RE3. But , the POLS estimator is inefficient and the usual standard errors are incorrect resulting in invalid tests ( for instance , t-, F-, z-, Wald- ). Cluster-robust covariance estimator treating each cross-sectional unit as a cluster may be said to yield correct standard errors ( Schmidheiny , 2016 , pp. 5-6 ).
4.7.1.4
Tests for Choosing Appropriate Panel Data Model
Based on the above specifications, the choice between Pooled OLS, Fixed Effects and Random Effects Models may be made by implementing the following tests :-
( 1 ) Pooled OLS ( POLS ) versus Fixed Effects ( FE ) Models Chow test or Restricted F-test : FE model and POLS model are the Unrestricted ( UR ) and Restricted ( R ) models respectively. Null Hypothesis ( H0 )
: η i ( i = 1,2,3,…,N ) = 0.
Alternative Hypothesis ( H1 ) : η i ( i = 1,2,3,…,N ) 0. The F statistic under H0 given by : ( ESSR ̶ ESS UR ) / ( N ̶ 1 ) F = -----------------------------------------∼ F ( N-1 ), ( NT ESS UR / ( NT – N – K )
–N –K)
( 14 )
where
ESSR and ESSUR = Error Sum of Square for the Restricted and Unrestricted models respecttively ; NT = total number of observations ; K = number of regressors in the Unrestricted model. If H0 is rejected, then FE model may be preferred to POLS model. This test is automatically implemented in Stata after the commands xtreg (with fe option ) or areg.
( 2 ) Pooled OLS ( POLS ) versus Random Effects ( RE ) Models In order to test the presence of random effects in the underlying POLS model , a Lagrange Multiplier ( LM ) test developed by Breusch and Pagan ( 1980 ) and based on OLS residuals may be implemented after a RE regression. Under the null hypothesis ( H0 ) of variance of the Pooled OLS residuals being equal to zero, the LM statistic30 follows a chi-square distribution with one degree of freedom. If H0 is rejected, then RE model may be preferred to POLS model. The test is implemented by the Stata command xttest0 after xtreg with re option. 30
Details in Baltagi ( 2005, pp. 59-60 ).
124
( 3 ) Fixed Effects ( FE ) versus Random Effects ( FE ) Models The choice between FE and RE models is primarily based on the consideration of the correlation between the unobserved cross-sectional - specific effects and the regressors. Hausman ( 1978 ) proposed a test
31
based on the difference between the RE and FE
estimates , with the null hypothesis that there should not systematic or substantial difference between the two estimates. Under the null hypothesis of no correlation between the unobserved cross-sectional - specific effects and the regressors , FE estimate is consistent but inefficient , whereas RE estimate is the Best Linear Unbiased Estimator ( BLUE ), consistent and efficient. Under the alternative hypothesis , FE estimate is consistent , but RE estimate is inconsistent ( Baltagi op. cit., p. 66 ). A statistically significant difference may be interpreted as evidence against the RE assumption of uncorrelated unobserved cross-sectional specific effects ( Wooldridge , 2010 , p. 288 ) , and the FE model may be said to be preferable to the RE model. The Stata command hausman performs this test. However , the assumption of this test that one estimator should be fully efficient under the null hypothesis will often be violated if heteroskedasticity or serial correlation is present within panels , as the RE estimator will not be fully efficient in that case. A cluster - robust Hausman32 test , based on bootstrap resampling over cross-sectional units treated as cluster , may be implemented by the Stata command rhausman ( Boris , 2015 ) in such a scenario. A test for comparing FE and RE estimates may also be viewed as a test of overidentifying restrictions. The FE estimator incorporates the orthogonality conditions that the regressors and the stochastic error term are uncorrelated . Additional orthogonal conditions or overidentifying restritions, that the regressors and the group-specific error are uncorrelated , are used by the RE estimator . These overidentifying restritions which are responsible for the increased efficiency of the RE estimator compared to the FE estimator might not be valid. The null hypothesis is that they are valid , and rejection of the null suggests that they are not and hence the random effects estimator that uses them is inconsistent . Arellano ( 1993 ) and Wooldridge ( 2002 ) describe the artificial regression approach in which a RE model is reestimated with the additional regressors in the form of deviations of the original regressors from their respective means. A Wald test of the joint significance of these additional regressors is preformed. A large-sample chi-squared test statistic ( Sargan-Hansen statistic ) , which : ( a ) is asymptotically similar to the traditional Hausman test of FE versus RE
31
Details in Baltagi ( 2005 , pp. 66-74 ) , Greene ( 2010 , pp. 379-380 ) and Cameron ( 2005 , pp. 715-719 ).
32
Details in Cameron ( 2005, p. 718 ).
125
estimators under conditional homoscedasticity and in case of balanced panel data ; ( b ) considers heteroscedasticity and cluster-robust variances ; and ( c ) guarantees the generation of a positive test-statistic , is reported without correcting for any degrees of freedom ( Schaffer and Stillman , 2016 ). This test is implemented by the Stata command xtoverid.
4.7.1.5
Tests for Heteroscedasticity , Serial Correlation , Cross-sectional Dependence and Multicollinearity in Panel Data Models
( 1 ) Heteroscedasticity When heteroscedasticity is present the standard errors of the estimates will be biased and robust standard errors should be computed for correcting for the possible presence of heteroscedasticity. In the context of panel data, the most likely deviation from homoscedastic errors will be the error variances specific to the cross-sectional unit. When the error process is homoscedastic within cross-sectional units but its variance differs across units, it gives rise to group-wise heteroscedasticity. ( a ) Pooled OLS Model : White ( 1980 ) proposed a general test for heteroskedasticity in the error distribution which may be non-normal , by regressing the squared residuals on all distinct regressors , cross-products , and squares of regressors. The test statistic , a Lagrange multiplier measure , is distributed Chi-squared under the null hypothesis of homoscedasticity and alternative hypothesis of unrestricted heteroscedasticity. Wooldridge ( 2000 , pp. 259-260 ) suggested a special version of the White Test by specifying the fitted option in which the predicted values from the original regression and their squares are used in place of the individual regressors, their squares, and their cross-products, conserving on the many degrees of freedom used by White test for models with just a moderate number of independent variables. The Stata commands ‘ whitetst ’ and ‘whitetst , fitted ’ after ‘regress’ yields respectively White and White-Wooldridge tests. ( b ) Fixed Effects Model : Following Greene ( 2000 , p. 598 ) , a modified Wald statistic for group-wise heteroskedasticity in the residuals of a fixed-effect regression model will be calculated applying the Stata command xttest3 after xtreg with fe option. The resulting test statistic is distributed Chi-squared under the null hypothesis of homoscedasticity , that is , 𝜎𝑖2 = 𝜎 2 for i = 1,2,...,N, where N is the number of cross-sectional units. The modified Wald statistic is workable even when the assumption of normality is violated, at least in asymptotic terms.
126
( 2 ) Serial Correlation ( a ) Pooled OLS : In this case , Durbin -Watson ( 1951 ) Test 33 for first-order serial correlation [ AR(1) ] can be implemented. The null hypothesis of the test is that there is no first-order autocorrelation. The range of the values of the Durbin-Watson d statistic is from 0 to 4. Under the null hypothesis , d is equal to 2. Positive autocorrelation ( ρ34 > 0 ) and negative autocorrelation ( ρ < 0 ) may be said to be present for values of d less than 2 and for values of d greater than 2 respectively. ( b ) Fixed Effects Model : The Durbin -Watson statistic has been generalized to the fixedeffects panel model by Bhargava et al.( 1982 ). Lagrange Multiplier ( LM ) statistics for firstorder serial correlation has been derived by Baltagi and Li ( 1991,1995 ) and Baltagi and Wu ( 1999 ). Drukker (2003) has developed a simple test [ Wooldridge-Drukker ( WD ) test ] for first-order serial correlation extending an original proposition by Wooldridge ( 2002 ) based on the OLS residuals of the first-differenced model. Wooldridge ( 2002 ) suggests a test for the absence of an unobserved effect (i ). Under the null hypothesis of σ2η = 0, the errors, uit are serially correlated. Simulation results provided by Drukker ( 2003 ) show that the test has good size and power properties in reasonably sized samples ( Born and Breitung, 2016 ). The Stata command xtserial after xtreg implements this test. However, these tests have their limitations. In the case of Bhargava et al. (1982) statistic , the distribution depends on N ( the number of cross-sectional units) and T ( the number of time periods ) and , therefore , the critical values have to be provided in large tables depending on both dimensions. Baltagi and Li ( 1995 ) noted that, for fixed T, their test statistic does not possess the usual χ2 limiting distribution due to the ( Nickell ) bias35 in the estimation of the autocorrelation coefficient. Furthermore, these tests are not robust against temporal heteroscedasticity. Inoue and Solon (2006) proposed a portmanteau test deriving an LM statistic for the more general null hypothesis : E(uit uis) = 0 for all t < s, where uit denotes the error of the panel regression. Since this test is not designed to test against a particular alternative it is suitable for panels with small T . In case of moderate or large T , the size and power properties suffer because the dimension of the null hypothesis increases with T2 ( Born and Breitung , 2016 ). A heteroscedasticity-robust ( HR ) test for first order serial correlation,
33
Details in Gujarati ( 2003 , pp. 467-472 ).
34
ρ is the first-order autocorrelation coefficient.
35
The LSDV estimators for dynamic fixed effects model remains biased with the introduction of exogenous variables if T is small ; details of derivation in Nickell (1981).
127
whereby demeaned residuals are regressed backwards on lagged forward demeaned residuals using a heteroscedasticity and autocorrelation robust estimator, has been proposed by Born and Breitung ( 2016 ). An F-test is then performed on the estimated coefficients by calculating the ( time dependent ) heteroscedasticity-robust HR statistic that is asymptotically equivalent to this F-test. The Stata commands xthrtest after xtreg with fe option implements this test. ( c ) Random Effects Model : The Wooldridge-Drukker ( WD ) test mentioned above is applicable for random effects model as well. ( 3 ) Cross-sectional Dependence Since the panel structure of the data considered in this study is that of large N-small T type ( that is , large number of cross-sectional units and small number of time periods ) , the following three tests will be applied to this data for testing cross-sectional dependence of the error term ( Hoyos and Sarafidis , 2006 ) : ( a ) Pesaran’s CD Test : Pesaran ( 2004 ) developed the CD statistic based on an average of pairwise correlation coefficients of OLS residuals from the individual regressions in the panel to test the null hypothesis of cross-sectional independence in panel data models with small T ( time dimension ) and large N ( cross-sectional dimension ) . The CD statistic is given by : 2T
N CD = √N(N−1 ) ( ∑N-1 i=1 ∑j=i+1 𝑝̂ 𝑖𝑗 )
( 15 )
where 𝑝̂ 𝑖𝑗 is the sample estimate of the pairwise correlation of the residuals . Under the null , the CD statistic has an asymptotically normal distribution as N → ∞ and T sufficiently large . The stata command xtcsd , pesaran after xtreg ( with fe or re option ) implements this test . ( b ) Friendman’s Test : Friedman ( 1937 ) proposed a non-parametric test based on the average of the Spearman’s rank correlation coefficients across pairs of cross-sectional disturbances . Friedman’s statistic ( FR ) is given by : FR = ( T − 1) { ( N − 1) Ra + 1} where Ra =
2 N(N−1 )
( 16 )
N (∑N-1 i=1 ∑j=i+1 𝑟̂𝑖𝑗 ) and 𝑟̂𝑖𝑗 is the sample estimate of the rank correlation
coefficient of the residuals. FR is asymptotically 2 distributed with ( T 1 ) degrees of freedom , for fixed T as N gets large. The stata command xtcsd , friedman after xtreg ( with fe or re option ) implements this test.
128
( c ) Frees Test : The CD and FR both involve the sum of the pairwise correlation coefficients of the residual matrix which implies that cases of cross-sectional dependence with alternating signs of the correlations are likely to missed by these tests , that is , large positive and negative correlations in the residuals will cancel each other out during averaging. Frees test does not suffer from this drawback ( Ibid.). Frees ( 1995 ) derives a non-parametric test based on the average of the squared Spearman rank correlation coefficients across pairs of cross-sectional disturbances. Frees’ statistic ( FRE ) is given by : FRE = N { ( R2a ( T 1 ) 1 } where R2a =
2 N(N−1 )
( 17 )
N 2 (∑N-1 i=1 ∑j=i+1 𝑟̂ 𝑖𝑗 ) and 𝑟̂𝑖𝑗 is the sample estimate of the squared rank
correlation coefficient of the residuals. FRE , asymptotically follows the Q-distribution which is a ( weighted ) sum of two χ2-distributed random variables and depends on the size of T. The stata command xtcsd , frees after xtreg ( with fe or re option ) implements this test. ( 4 ) Multicollinearity The term ‘multicollinearity’ , coined by Ragner Frisch ( 1934 ) refers to the existence of a perfect or exact linear relationship among all or some explanatory variables of a regression model ( Gujarati , 2003 ). However, even extremely high multicollinearity ( so long as it is not perfect ) does not violate OLS assumptions and OLS estimates are still BLUE or Best Linear Unbiased Estimators ( Ibid.). The only problem with very high or near multicollinearity is that it will be hard to obtain coefficient estimates with small standard errors. But this problem will also faced when the number of observations are small ( Achen , 1982 ). Goldberger ( 1991, p. 249 ) coined the term “micronumerosity” as diametrically opposite to multicollinearity , where exact micronumerosity denotes a sample size of zero. Near micromunerosity is said to arise when the number of observations barely exceeds the number of parameters to be estimated ( Gujarati , 2003 ). As mentioned in Section 4.5.1 , panel data is plagued with the problem of multicollinearity among the explanatory variables to a lesser extent than that of pure time series or cross-sectional data. Nevertheless, multicollinearity will be tested on the respective panel data regression model through the application of Variance Inflation Factor ( VIF ) which quantifies the degree of inflation of the variance of a coefficient estimate in the presence of multicollinearity. Each of the explanatory variables of a particular regression model are regressed on the remaining explanatory variables and the VIF for the kth explanatory variable is given by : VIFk = 1 / ( 1 – R2k )
( 18 )
129
where , R2k is the R2 ( coefficient of determination ) value obtained from the regression. Several researchers 36 have suggested that the rules of thumb for large VIFs of 5 or 10 are based on the associated R2k values of 0.80 or 0.90 respectively. O’Brien ( 2007 ) advices to treat these rules of thumb with caution when making decisions to reduce multicollinearity ( for instance , eliminating one or more predictors ) and suggests that researchers should also consider other factors ( such as, sample size ) which influence the variability of regression coefficients ( Murray et al., 2012 ). VIFs cannot be directly calculated as a postestimation option after the Stata ( panel data ) commands xtreg and areg. However, they can be computed as a postestimation option after the Stata command regress considering crosssectional dummy variables for fixed effects model only , though this method may be tedious if the number of cross-sectional units is quite large. The postestimation option varinf in Eviews appears to be very useful in this matter.
4.7.1.6
Robust Standard Errors
When some of the assumptions of the underlying regression model are violated , robust standard errors should be relied upon in order to ensure valid statistical inference. Some of the suggestions about the use of robust standard errors and the Stata commands to be applied for their implementation are summaried below ( Hoechle , 2007 ) :( 1 ) If the residuals are independently distributed but heteroscedastic , then heteroscedasticity-consistent Eicker-Huber-White 37 standard errors should be used by applying the option vce ( robust ) after the Stata command xtreg or reg. ( 2 ) If the residuals are heteroscedastic and correlated within clusters but uncorrelated between clusters , then the option cluster( ) after the Stata command xtreg or reg allows the computation of Rogers (1993) or clustered standard errors which are robust to heteroscedasticity and autocorrelation provided the panel identifier ( variable representing the crosssectional units ) is taken as the cluster( ) variable. ( 3 ) If the residuals are autocorrelated with AR(1) , the Stata command xtregar will suffice. ( 4 ) If the residuals are heteroscedastic and autocorrelated of moving average type with lag length q [ that is , M ( q ) ] then standard errors developed by Newey and West ( 1987 ) derived from generalized method of moments-based covariance matrix , which is an extension
36 37
for instance , Hocking and Pendelton ( 1983 ) , Craney and Surles ( 2002 ) . Eicker ( 1967 ) , Huber ( 1967 ) and White ( 1980 ) .
130
of White’s estimator and which may be applied through the Stata command newey with option force. ( 5 ) If the residuals are heteroscedastic and temporally as well as cross-sectionally correlated, then the following methods may be applied : ( i ) Parks -Kmenta 38 method based on feasible generalized least-squares ( FGLS ) implemented by the Stata command xtgls with option panels ( correlated ) ; however , this method is infeasible for microeconometric panels like that of the present study , that is , when T is less than N ; and as shown by Beck and Katz ( 1995 ) this method tends to produce standard error estimates which are unacceptably small ; ( ii ) Beck and Katz ( 1995 )’s panel corrected standard errors ( PCSE ) method which rely on OLS coefficient estimates and which may be implemented by the Stata command xtpcse ; however , it is convincingly demonstrated by Beck and Katz (1995) that their large-T asymptotics-based standard errors perform well in small panels , but if N is large compared to T and if the ratio ( T / N ) is small , the finite sample properties of this estimate are rather poor and imprecise ; ( iii ) Driscoll and Kraay ( 1998 ) standard errors ; Driscoll and Kraay ( 1998 ) , applying Monte Carlo simulations , show that the presence of even modest cross-sectional or spatial dependence can render OLS standard errors to be largely biased when N is large , and present conditions under which a simple modification of the standard nonparametric time series covariance matrix estimator yields estimates of the standard errors that are robust to general forms of spatial and temporal dependence as T → ∞ ; however , if T is small , the problem of consistent nonparametric covariance matrix estimation appears to be much less tractable ; parametric corrections for spatial correlation are possible only if strong restrictions are palced on their form. For typical micropanels with N much larger than T, estimating this correlation is impossible without imposing restrictions, since the number of spatial correlations increases at the rate N2 , while the number of observations grows at rate N ; for instance , in case of macro panels where N = 100 countries observed over T = 20 to 30 years, N is still larger than T requiring the need for prior restrictions on the form of spatial correlation ( Baltagi , 2005 ; Driscoll and Kraay , 1998 ). Hoechle ( 2007 ) suggests the use of Driscoll Kraay standard errors when residuals are heteroskedastic, autocorrelated with moving average type with lag length q [ that is , MA(q) ] and cross-sectionally depen-
38
Parks ( 1967 ) and Kmenta ( 1986 ) .
131
dent , through the application of the Stata command xtscc , but adivices that caution should be exercised when applying this estimator to panels with large N and small T .
4.7.2
Quantile Regression ( QR ) Methodology
This study seeks to anlayse the panel data sample through the application of Quantile Regression ( QR ) technique . Ordinary least squares regression techniques calculate the average effect of the explanatory variables on the ‘average firm’ , the focus on which may hide important features of the underlying relationship ( Coad and Rao , 2006 ). Mosteller and Tukey ( 1977, p. 266 ) explain this lucidy as follows : “What the regression curve does is give a grand summary for the averages of the distributions corresponding to the set of x’s. We could go further and compute several regression curves corresponding to the various percentage points of the distributions and thus get a more complete picture of the set. Ordinarily this is not done, and so regression often gives a rather incomplete picture. Just as the mean gives an incomplete picture of a single distribution , so the regression curve gives a correspondingly incomplete picture for a set of distributions”. Application of QR methodology helps
to obtain
more complete picture(s) of the relationship(s) between the
response variable and the explanatory variable(s). Quantiles are points taken at regular intervals from the cumulative distribution function ( CDF ) of a random variable. A quantile is basically a percentile which is a general term for median ( 50 th quantile ) , quartile ( 25th quantile ) , pentile ( 20th quantile ) , decile ( 10th quantile ) and so on . A researcher can account for unobserved heterogeneity and heterogeneous covariates effects through the application of QR models , while he/she is allowed to to control for some unobserved covariates by the inclusion of fixed effects through the availability of panel data ( Canay , 2011 ). The differential impacts of covariates along the distribution of an outcome can be understood through the pragmatic approach provided by QR ( Borah and Basu , 2013 ) . Davino et al. ( 2014 , pp. 2-3 ) compare mean , median and quantiles as follows: Let y be a random variable. Its unconditional mean is defined as the centre c of the distribution which minimizes the squared sum of deviations, that is , as the solution to the following minimization problem : μ = argmin E ( y – c ) 2
( 19 )
c
The median minimizes the absolute sum of deviations : m
e
= argmin E | y – c | c
( 20 )
132
Now, the -th quantile is the value z such that P ( y z ) = 0. Considering the Cumulative Distribution Function ( CDF ) : Fy ( z ) = F ( z ) = P ( y z )
( 21 )
the quantile function is defined as its inverse : q
y (
) = q ( ) = Fy-1 ( ) = inf { z : F ( z ) > }
( 22 )
for ϵ [ 0, 1 ] If F (. ) is strictly increasing and continuous , then F y-1 ( ) is the unique real number z such that F ( z ) = .39 QR estimates the effect of an independent variable on a dependent variable, with covariates being allowed to act as controls , similar to that of OLS . In constrast to OLS, QR provides estimates of these effects at different points of the distribution of the dependent variable rather than just the mean ( as in OLS ) and is insensitive to outliers on the dependent variable ( Porter , 2015 ).
4.7.2.1
Conditional Quantile Regression ( CQR )
The quantile regression [ or more appropriately, Conditional Quantile Regression ( CQR ) ] model as applicable to cross-sectional data was introduced by Koenker and Bassett (1978). Let ( yi, xi ), i = 1, 2, …, n, be a sample from some population where y is the dependent variable and x is a ( K × 1 ) vector of regressors. Assuming that the qth quantile of the conditional distribution of yi is linear in xi, the conditional quantile regression model may be written as : yi = qθ ( yi / xi ) + εθ i qθ ( yi / xi
)
= inf { y : Fi ( y | x ) θ } = xi βθ
qθ ( εθ i / xi ) = 0 where q
θ
( 23 ) ( 24 ) ( 25 )
( yi / xi ) denotes the θ-th conditional quantile of yi conditional on the regressor
vector xi ; β is the unknown vector of parameters to be estimated for different values of θ in ( 0,1) ; ε is the vector of error terms ; and Fi ( y | x ) denotes the conditional distribution function. The entire distribution of y, conditional on x, can be traced by varying the value of from 0 to 1.
39
Gilchrist ( 2000 ) as referred to in Davino et al. ( 2014 , p. 2 ) .
133
Koenker and Hallock ( 2001 ) shows that the estimator for βθ is obtained by solving the following linear programming problem :
arg min ∑𝑛𝑖=1 θ ( yi xi βθ )
( 26 )
q
where θ ( ) is a loss function defined as : θ*
if 0
θ ( ) =
( 27 ) (1θ)*
if 0
The loss function assigns a weight of θ to positive residuals and a weight of ( 1 θ ) to negative residuals. For conditional quantile regression, interpretation of the coefficients is in relation to the quantiles of the distributions defined by the covariates ( the conditional distribution ), rather than the unconditional distribution of the dependent variable. Thus including covariates or fixed effects not only adjusts for selection bias, but also redefines the quantiles, thus changing the interpretations of the variables ( Porter , 2015 ). As Porter ( 2015 , pp. 342-343 ) elucidates “ Just as OLS yields the effect of a variable at the mean of y, we also wish to know the effect at other quantiles of y, not quantiles of y defined within subgroups. The main issue here is that inclusion of control variables in a conditional regression model is necessary to deal with selection bias, just as in the case of OLS, yet inclusion of these covariates changes the interpretation of the quantiles. Moreover, as additional covariates are included, the interpretation of the quantiles changes, making comparisons across different model specifications problematic.” Thus , in our case , for example , the 90th conditional quantile of the dependent variable ( MLEV or BLEV ) , conditioned on the covariates SIZE ( firm size ), PROF ( profitability ) and GROW ( growth opportunities ) may represent “highly levered small-sized firms having high profitability and low growth opportunities” rather than just “highly levered firms” ( in which we are interested ) as correctly represented by the 90th unconditional quantile of MLEV or BLEV. Porter ( 2015 , p. 343 ) further reiterates “ the growing consensus in the literature is that many researchers have inadvertently misused conditional quantile regression for many years, by interpreting the results as if they came from an unconditional quantile regression model. In other words, they have interpreted their coefficients as if they were the effect on the quantile of y, rather than quantiles of y defined within groups based on their set of covariates”. As a result , many researchers are applying Unconditional Quantile Regression ( UQR ).
134
4.7.2.2 4.7.2.2.1
Unconditional Quantile Regression ( UQR ) Concept of UQR
Firpo et al. ( 2009 ) developed the Unconditional Quantile Regression ( UQR ) model which is not influenced by any right-hand-side variables as the quantiles are defined pre-regression ( Killewald and Bearak , 2014 ; Borgen , 2016 ). In UQR , selection bias can be adjusted by incorporating fixed effects without the quantiles being redefined ( Borgen , 2016 ). Unlike CQR , UQR being a non-parametric regression method is robust to a mis-specified set of conditioning variables ( Maclean et al. , 2014 ). The method consists of running a ( conditional ) regression of a transformation - the ‘Recentered Influence Function’ ( RIF ) - of the outcome variable on the explanatory variables. The RIF is an extension of the concept of ‘Influence Function’ ( IF ) used in robust statistics to address the effect , or influence, of adding or removing a particular observation on the value of a distributional statistic 40 . Let ( yi, xi ) , i = 1, 2, …, n, be a sample from some population where y is the dependent variable and x is a ( K × 1 ) vector of regressors.The Influence Function, IF ( yi ; ν, Fy ) of a distributional statistic ν ( Fy ) of the cumulative ( unconditional ) distribution function ( Fy ) of y represents the influence of an individual observation yi on that distributional statistic. If the statistic ν ( Fy ) is added back to the influence function , IF the ‘Recentered Influence Function’ ( RIF ) is obtained ( Firpo et al. , 2009 ). For a quantile of Fy, RIF ( yi ; q, Fy ) = q + [ 1{ yi ≤ 𝑞 } / fy ( q ) ]
( 28 )
where [ 1{ yi ≤ 𝑞 } ] / fy ( q ) is the Influence Function of yi ; 1{ yi ≤ q } is an indicator function for whether yi is above or below quantile , that is, 1{ yi ≤ q } equals one when the value of yi is less than or equal to the value of yi at quantile q , and zero otherwise ; fy ( q ) is the probability density of the unconditional distribution function of y at q . The RIF could serve as the outcome variable in an OLS model yielding a so-called RIFOLS model ( Ibid.) 41 . When the conditional expectation of RIF ( yi ; q, Fy ) is modeled as a function of the explanatory variables xi we get the RIF-OLS regression model given by : E [ RIF ( yi ; q, Fy ) | xi ] = m ( xi ) [ say ]
40 41
( 29 )
such as mean, variance, quantiles, etc.
RIF-logit and RIF-NP are two other methods described by Firpo, Fortin, and Lemieux ( 2009 ) for estimating UQR. The differences between the three estimation methods are minor in their applications. ( Borgen, 2016, p. 405 ).
135
This RIF regression can be viewed as an UQR. This is because by the definition of RIF and the law of iterated expectations : Ex E [ RIF ( yi ; q, Fy ) | xi ] = q ( Fy ) or Ex E [ m ( xi ) ] = q ( Fy )
( 30 )
The average derivative of the UQR given by Ex ( d m ( xi ) / d xi ) 42 can be interpreted as any OLS statistic , that is , the predicted impact of the effect of a small change in the location of the distribution of the explanatory variables x on the unconditional quantile of the dependent variable y, ceteris paribus ( Ibid.). The probability density of the unconditional distribution of y at q is estimated using a kernel density estimator 43 as :
𝑓̂𝑦 (𝑞̂𝜃 ) =
1 𝑁𝑏
∑𝑁 𝑖=1 𝐾𝑦 (
𝑦𝑖 − 𝑞̂𝜃 𝑏
)
( 31 )
where 𝐾𝑌 (. ) is a kernel function ; 𝑏 is a positive scalar parameter called the bandwidth, the size of which determines how smooth or spiky the estimated density curve is ; and two times b is the window width ; 𝑞̂𝜃 is estimator of the θ th quantile of the unconditional distribution of Y. The kernel function K (· ) is a continuous function, symmetric around zero, that integrates to unity and satisfies additional boundedness conditions ( Cameron and Trivedi , 2005 ). Following Lee ( 1996 ) , Cameron and Trivedi ( 2005 , p. 299 ) assume that the kernel function K ( z ) satisfies the following conditions : ( i ) K ( z ) is symmetric around zero and is continuous. ( ii ) K ( z ) dz = 1, z K ( z ) dz = 0, and | K ( z ) | dz < ∞. ( iii ) Either ( a ) K ( z ) = 0 if | z | ≥ z0 for some z0, or ( b ) | z | K ( z ) → 0 as | z | → ∞. ( iv ) z2 K ( z ) dz = κ, where κ is a constant. Kernel functions work better if they satisfy condition ( iiia ) rather than just the weaker condition ( iiib ). Then restricting attention to the interval [−1, 1 ] rather than [−z0, z0 ] is simply a normalization for convenience, and usually K ( z ) is restricted to z ∈ [−1, 1 ]. However, the existing literature does not provide much advice for the determination of the appropriate kernel function and bandwidth choice in respect of unconditional quantile regression. As a result, researchers tend to rely on the defaults of their respective software 42
where ( d m ( xi ) / d xi ) is the ( K x 1 ) vector of partial derivatives { δ m ( xi ) j / ( δ xi ) j } ( j = 1,2,..,K ) which is equal to [ d q ( Fy ) / d xi ] 43
Kernel density estimator, introduced by Rosenblatt ( 1956 ), is a non-parametric estimator ( which does not rely on distributional assumptions ) for estimating the probability density function of a random variable.
136
packages. However, it is better to report the results of alternative kernel functions for sensitivity analysis ( Porter , 2015 ). The optimal bandwidth parameter varies with the kernel and is much more important than the kernel ( Cameron and Trivedi , 2005 ). Nine alternative combinations of the three commonly used kernel density functions [ Epanechnikov ( or quadratic or parabolic ) , Gaussian ( or Standard Normal ) and Uniform ( or Rectangular ) ] and three optimal bandwidth choices [ Silverman ( Silverman , 1986 ) , Hä rdle ( Hä rdle et al. , 1992 ) and Scott ( Scott , 1992 ) ] will be considered in this study. These measures are tabulated below following Cameron and Trivedi ( 2005 , 300 ), Zucchini et al. ( 2003 ) , Porter ( 2015 ), Guidoum ( 2015 ) and Jann ( 2007 ) : Table 4.6 : Kernel Density Functions and Optimal Bandwidth Parameters
Kernel
Epanechnikov Gaussian Rectangular ( Uniform )
Kernel Function K(z)
Measure of “roughness” R(K) = {K ( z )}2 dz
Variance of the Kernel [ 𝝈𝟐𝑲 ] = 2 z K ( z ) dz
Canonical Bandwidth ( ) =
(3/5)
(1/5)
( 15 ) 0.2
100 %
( 1 / 2 )
1
( 1 / 4 ) 0.2
95.12 %
(1/2)
(1/3)
( 9 / 2 ) 0.2
92.95 %
( 3 / 4 ) * ( 1 – z2 ) * 1(|z |