Does it really matter how a firm diversifies? Assets-in

0 downloads 0 Views 399KB Size Report
Jan 31, 2017 - growth in multiple businesses (growth options diversification). ... diversification and a firm's value, with said strategy being more ... tion of the firm's activity across segments and a relevant commitment in each. ... (1993) find that incremental strategies lead to better performance in uncertain and dynamic envi-.
Journal of Corporate Finance 43 (2017) 316–339

Contents lists available at ScienceDirect

Journal of Corporate Finance journal homepage: www.elsevier.com/locate/jcorpfin

Does it really matter how a firm diversifies? Assets-in-place diversification versus growth options diversification Pablo de Andrés a, Gabriel de la Fuente b, Pilar Velasco c,⁎ a b c

Universidad Autónoma de Madrid, Department of Finance, Faculty of Economics, Cantoblanco Campus, 28049 Madrid, Spain Universidad de Valladolid, Department of Financial Economics, Faculty of Economics, Avda. Valle del Esgueva 6, 47011 Valladolid, Spain Universidad de Alcalá, Department of Economics and Business, Faculty of Economics, Plaza Victoria s/n, 28802 Alcalá de Henares, Madrid, Spain

a r t i c l e

i n f o

Article history: Received 11 March 2016 Received in revised form 25 January 2017 Accepted 28 January 2017 Available online 31 January 2017 JEL classification: L25 G32 D22 C34

a b s t r a c t This study analyses whether the effect of corporate diversification on a firm's market value depends on how this strategy is implemented. According to the real options approach, two opposite diversification strategies may be implemented: one based on fully exercising available options (assets-in-place diversification) and the other aimed at seeding new opportunities for future growth in multiple businesses (growth options diversification). We propose an index to measure these two diversification patterns and we explore their impact on firm market value for a sample of U.S. firms during 1998–2014. We find that as a firm's diversification strategy shifts towards a growth options pattern, it becomes a more value-enhancing strategy. © 2017 Elsevier B.V. All rights reserved.

Keywords: Corporate diversification Growth opportunities Firm value Investment strategy Self-selection

1. Introduction The linkage between diversification and firm value has constituted a prolific area of research. Existing literature offers abundant yet inconclusive evidence on the topic, with the research question becoming widely known as the diversification puzzle.1 Although the bulk of the research provides evidence that diversified firms trade at a discount relative to non-diversified companies in their industries (Berger and Ofek, 1995; Servaes, 1996; Stowe and Xing, 2006; Borghesi et al., 2007; Hoechle et al., 2012; Kuppuswamy and Villalonga, 2016), other works call these findings into question, and report a non-statistically significant relationship (Villalonga, 2004b; Elsas et al., 2010), a quadratic relationship (Palich et al., 2000; Andrés et al., 2014), or even premiums for diversifying (Campa and Kedia, 2002; Villalonga, 2004a). Some studies have suggested that these inconsistent findings may be due to one or more of the following three reasons: biases in data (Villalonga, 2004a), methodological issues (Campa and Kedia, 2002; Villalonga, 2004b), or the presence of moderating factors in the diversification-value relationship (Rajan et al., 2000). The identification of moderating factors suggests that the impact of this strategy on performance may not be homogeneous across firms but rather dependent on aspects that might enable certain enterprises ⁎ Corresponding author. E-mail addresses: [email protected] (P. de Andrés), [email protected] (G. de la Fuente), [email protected] (P. Velasco). 1 Martin and Sayrak (2003), Erdorf et al. (2013), and Ahuja and Novelli (2016) provide three complementary surveys of the corporate diversification literature.

http://dx.doi.org/10.1016/j.jcorpfin.2017.01.011 0929-1199/© 2017 Elsevier B.V. All rights reserved.

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

317

to create more value than others. Such moderating factors could be classified into three categories: market and institutional level factors (Fauver et al., 2003; Rudolph and Schwetzler, 2013), industry level factors (Santaló and Becerra, 2008), and firm level factors (Rajan et al., 2000). Among the firm-specific characteristics that may account for differences in diversification value outcomes, one strand of literature explores the role of a firm's growth opportunities. Bernardo and Chowdhry (2002) attribute a significant role to growth opportunities when accounting for the diversification discount: unisegment firms have more options to expand whereas diversified firms may have exhausted some of these. Concurring with this line of thinking, Borghesi et al. (2007) claim that pure-play firms and their diversified industry-peers exhibit different growth potential. Once the age of the firm, used as a proxy for growth opportunities, is controlled for, the initially displayed discount decreases. Andrés et al. (2016) report that growth options partly mediate the relationship between diversification and a firm's value, with said strategy being more value-enhancing insofar as it increases growth options to a greater extent. In a similar vein, Ferris et al. (2002) analyse diversification for a sample of international joint ventures and show that diversification is only value-destroying in enterprises that have a poor set of growth opportunities. Alternatively, Stowe and Xing (2006) offer evidence that differences in growth opportunities are not the main driver of the diversification discount, since the discount persists even after such differences are controlled for. Finally, Holder and Zhao (2015) argue that prior evidence on the diversification discount fails to consider the impact of diversification on changes in the value of a firm's growth options. They find that the diversification discount may be the joint result of the increase in value in below-average performers exploring new growth opportunities through unrelated diversification, and the decrease in value in above-average performers exploiting their current growth opportunities through related diversification. In light of these findings, there appears to be a need to examine the role played by growth options in the diversification-value puzzle in greater depth. Specifically, two further questions arise: first, whether diversification strategies can differ depending on their divergent effect when shaping a firm's growth options portfolio; and second, whether such differences might have an impact on corporate value. We adopt a real options approach to identify two contrasting diversification patterns: one based on fully committed entry into new markets (assets-in-place diversification), and the other aimed at seeding simultaneous growth options in multiple businesses (growth options diversification). Logically, no firm adopts either of these ‘extreme’ diversification patterns but rather an intermediate one, which might be closer to one or the other. Measuring the proximity to one of these extreme patterns is no easy task since neither a firm's strategy nor its real options portfolio are directly observable.2 To address this problem, we propose a proxy that can reflect some of the key effects that emerge from each strategy. To the best of our knowledge, no prior research has attempted to measure this diversification pattern.3 Assets-in-place (AiP, hereafter) diversification would be reflected in a firm's portfolio characterised by a highly uniform distribution of the firm's activity across segments and a relevant commitment in each. In contrast, growth options (GO, hereafter) diversification would show an unequal distribution across businesses with a low commitment in a number of them. Taking the concept of fit as profile deviation (Venkatraman, 1989), we propose a two-dimensional index, which combines an inequality-based measure of a firm's sales distribution across divisions (inter-segment component) and a measure of the scale of commitment in each (intra-segment component). The question we can then ask is whether the proximity to one of these poles makes any difference to the value of diversification. Amihud and Lev (1981) argue that the critical issue for the valuation effect of diversification is what kind of risk is reduced by diversification and whether stockholders could diversify this in their individual portfolios. If investors can diversify at a lower cost than enterprises, corporate diversification would destroy value. However, diversification aimed at providing the firm with growth options might not be so easily replicated. A GO diversification strategy boosts the firm's flexibility to adjust its decisions as uncertainty is resolved, and is geared towards exploring further opportunities in new industries before fully committing. As a result of this flexibility, corporate diversification may reduce risk and serve as ‘strategic insurance’ (Raynor, 2002), which cannot be replicated by investors. The most an individual investor could hope to achieve is to replicate the options portfolio by acquiring the stocks of such firms. However, the value of this replicating portfolio would be less than the value of the diversified firm's options portfolio since the optimal joint exercise of an options portfolio always proves more efficient than the sum of the individual optimal exercise of each option. Such arguments lead us to hypothesise that GO diversification might have a positive effect on value. Some prior research results seem to support the superiority of the GO diversification strategy for creating value over the AiP strategy. For example, Teplensky et al. (1993) find that incremental strategies lead to better performance in uncertain and dynamic environments, such as emerging markets, since they avoid full commitment of resources, while past performance acts as a feedback mechanism for future strategic decisions. In a similar vein, Andreou et al. (2016) report a discount among enterprises moving one time from a single segment to multiple ones and a premium among enterprises that undertake this strategy several times. Mitton and Vorkink (2010) also concur with these findings and show that the valuation effect of diversification is positively related to return skewness, which is consistent with superiority of the GO pattern insofar as positive skewness of a firm's stock returns approximates real options relevance in its total asset mix (Trigeorgis and Lambertides, 2014).4 Overall, this empirical evidence suggests that GO diversification may be more value enhancing than AiP diversification. 2 As is well known, unobservability is by no means an uncommon issue in social sciences, since most variables in their models are unobservable and proxies are used instead. 3 The only related precedent is Klingebiel and Adner (2015), who develop a classification of product innovation strategies to distinguish between real options logic and alternative resource allocation regimes. 4 Mitton and Vorkink's explanation is not based on RO. They argue that corporate diversification reduces stock return skewness as a consequence of return compensation, similar to what happens when combining stocks in a portfolio. Should investors prefer positive skewness, a firm's relative value would be discounted as skewness is reduced by segment diversification.

318

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

Based on this logic, we evaluate whether these diversification patterns explain the effect this strategy has on a firm's value. Our analysis is conducted on an unbalanced panel sample of 3558 U.S. firms (27,955 firm-year observations) for the period 1998–2014. Empirical models are estimated alternatively using the Heckman two-step procedure and the two-stage least squares (2SLS) to control for the potential endogeneity of diversification. We find strong evidence of the relevance of the diversification pattern in explaining diversification value outcomes, with GO diversification clearly emerging as a more value-creating strategy. The rest of the paper is structured as follows. Section 2 explains both diversification patterns and develops a proxy to measure them. The following section describes our empirical models, estimation methodology, data set, and variables. Section 4 provides some preliminary analyses of the performance of our proposed diversification pattern index. Section 5 describes our main empirical results and robustness tests. A discussion of the implications of the findings, limitations, and directions for future research is offered in the concluding section.

2. Measuring the diversification pattern: a quantitative proposal Traditional diversification indices are geared towards capturing the scope of diversification in terms of distributing firm business activity across segments. Yet, by themselves, they fail to provide useful information about how diversification investment is made. Degree and pattern of diversification constitute two different features of this strategy, each requiring specific measures. From the real options lens, two contrasting investment patterns can be distinguished (Bowman et al., 1992): a one-step strategy, which mainly involves full-scale commitments by making large sunk investments, versus a growth option strategy, entailing minor commitments in strategic areas, which serve as platforms for future investments. These two investment patterns translate to opposing ways of diversifying: assets-in-place (AiP) diversification versus growth options (GO) diversification. In AiP diversification, the firm holds a large position in each business it is involved in. Diversification into a new business is conceived as a one-shot investment strategy, meaning, immediate exercise or abandonment of previously acquired growth options. This strategy allows firms a fully committed entry into new markets to exploit potential economies of scope and synergies at the expense of taking a greater risk in these commitments, losing the flexibility to readjust the strategy along the way, and achieving limited exploratory capacity development. This diversification path prioritises exploiting available opportunities rather than remaining open to waiting for the best moment to take action. AiP diversification may correspond to a greater extent to the traditional notion of diversification under which the diversification strategy ‘consumes’ a firm's growth options in return for achieving strategic advantages such as synergies and market power. In GO diversification, the main objective of diversification is to develop further strategic options in new businesses (Williamson, 2001). Each investment is regarded as ‘a foothold in preparation for the next decision’ (Bowman et al., 1992: 98). The firm undertakes small-scale entries into several businesses, which act as ‘platforms’ for future expansion (Kogut and Kulatilaka, 1994; Laamanen, 1999). Through a GO diversification, the firm continuously builds and keeps its strategic options open for the future (Williamson, 2001), thereby encouraging flexibility and learning. As investment conditions evolve, firms will maintain, expand, exercise, or abandon these options while acquiring new options to diversify. GO diversification enables firms to react to uncertainty and fresh information as well as gain experience in a new field and explore further opportunities before fully committing. However, this ‘wait and see’ approach is by no means free of costs such as those emerging from the risk of pre-emption, which can threaten traditional sources of diversification efficiency. Overall, GO diversification is likely to enhance the value of the firm's growth options portfolio to a greater extent than its AiP value. Although a firm's selection of the particular type of diversification strategy is not directly observable, we can proxy it by measuring its effects on observable variables such as the unequal distribution of sales across segments and the scale of commitment within each segment. In particular, we propose an index (DIVPAT) based on these two dimensions: inter-segment distribution (INTER) and intrasegment commitment (INTRA).

2.1. Inter-segment distribution (INTER) INTER measures the degree of inequality in the distribution of the firm's level of diversification across its different businesses. AiP diversification would result in a uniform distribution of the firm's activity across its different segments. In contrast, GO diversification would reflect an unequal distribution as a result of combining major participation in its core businesses with minor exploratory investments in potential industries.5 Overall, INTER is intended to offer such an overview of the investment strategy followed by the firm as reflected by its business portfolio distribution.

5 Due to database limitations, we are only able to consider ‘foothold investments’ to a certain extent. The SFAS 131 reporting standard requires firms to separately report information about an operating segment if it represents 10% or more of total revenues, total profit, or total assets. Otherwise, segments ‘may be considered reportable and separately disclosed if management believes that information about the segment would be useful to readers of the financial statements’ (FASB 2008: 6). In our sample, for segments other than the core business, the proportion of segment sales over a firm's total sales ranges from 0 to 0.0799 on average (summary statistics available upon request). This is below the required threshold of 10% and supports the notion that our sample considers foothold investments of sufficiently low-scale participation.

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

319

We approximate this dimension by a Gini index applied to the distribution of business sales, and computed for each firm i and year t, as follows.6 First, firm i's segments are sorted into ascending order of sales to compute the following quotient: n−1

∑ P i;s −Q i;s

INTERi ¼



s¼1

ð1Þ

n−1

∑ P i;s s¼1

where s stands for each firm i's segment (s ranges from 1 to n-1), n represents the total number of firm i's segments, Pi,s denotes the cumulative proportion of i's sales from segment 1 to segment s, Q s,i denotes the cumulative proportion of i's total diversification from segment 1 to segment s, which is calculated as: s

2

∑ pi;k

Q i;s ¼

k¼1

ð2Þ

HERF i

where pi,k denotes the relative proportion of segment k's sales in firm i, and HERFi is the Herfindahl index defined by Hirschman (1964): n

2

HERF i ¼ ∑ pi;k

ð3Þ

k¼1

INTER takes values between 0 and 1. An INTER value equal to zero reflects perfect equality, and the higher the INTER, the greater the inequality. Thus, the nearer the INTER is to zero, the closer the firm's diversification pattern is to AiP diversification, whereas the nearer the INTER is to one, the closer the firm's diversification pattern is to GO diversification.7 2.2. Intra-segment commitment (INTRA) In order to appropriately evaluate whether a firm's diversification profile is closer to either AiP diversification or GO diversification, we also need to capture the extent of its commitment in each business. With this aim in mind, we incorporate for each firm and segment a measure of the standard AiP commitment position. Specifically, we use a multiplier approach to estimate the sales each firm would obtain from each business segment if it had an average industry commitment (imputed sales).8 We calculate imputed salesi,s as the product of firm i's total assets in year t (TAi) and the average multiple of sales to total assets as computed for all public listed firms (both single-segment and diversified) operating in industry s9: imputed salesi;s ¼ TAi  Inds ðsales=TAÞ

ð4Þ

To evaluate each firm i's scale of participation in each segment s, we compute a commitment ratio (commitmenti,s), which compares its sales in segment s (salesi,s) against the measure of the imputed sales in segment s: commitment i;s ¼

salesi;s imputed salesi;s

ð5Þ

A commitment above or equal to one will indicate that firm i holds an above-average involvement in that industry, in line with an AiP pattern of investment. Otherwise, it will display an option pattern based on under-developed participation in that business, which may serve as a platform for follow-on investments. Next, we compute the intra-segment component (INTRA) of our index, which seeks to capture the company's overall degree of involvement in all its business segments. INTRA is measured by the ratio of the number of firm i's segments displaying commitment ratios above or equal to one over firm i's total number of segments. INTRA ranges between zero and one, and is positively related to the AiP pattern of diversification: the closer the firm is to AiP diversification, the higher the INTRA.10 2.3. The diversification pattern index (DIVPAT) Finally, we combine INTRA and INTER in a two-dimensional index capturing the diversification pattern (DIVPAT), which is devised on the basis of the Euclidean distance. Following Venkatraman (1989), we take a concept of fit as profile deviation to analyse the 6

INTER and INTRA are computed yearly. However, we do not specify the subscript t for the sake of simplicity. For single-segment firms, we compute INTER equal to zero. We measure commitment in terms of sales due to the wider availability for this variable at a segment level. Prior works such as Rudolph and Schwetzler (2014) also note this point. 9 Sales are scaled by total assets to compute commitment in relative terms, thereby making this measure comparable across different-sized enterprises. We compute the mean ratio for each industry s at the 4-digit code level. 10 If a company follows an AiP strategy in all its businesses, its INTRA would by definition be one. 7 8

320

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

degree of adherence of a firm's diversification pattern to an externally specified profile. In our analysis, that externally specified profile serving as a reference will be the extreme case of AiP diversification, and the index to capture the Euclidean distance can generally be expressed for firm i in year t as: DIVPAT i;t ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi r 2  2 INTERi;t ‐INTERAiP;t þ INTRAi;t ‐INTRAAiP;t

ð6Þ

where INTERAiP and INTRAAiP denote, respectively, the value of the INTER and INTRA variables in period t for the extreme case of AiP diversification. The reference profile can be specified either theoretically or empirically (Venkatraman, 1989). From a theoretical standpoint, in extreme AiP diversification, INTER would equal zero, representing perfect equality, whereas INTRA would equal one, indicating that every segment of the firm holds a commitment above the industry mean. Thus, our diversification pattern index is defined as the deviation (Euclidean distance) of the firm's diversification pattern (in INTER and INTRA) from the extreme AiP diversification (INTER = 0, INTRA = 1): 

DIVPAT i;t ¼

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi r 2  2 INTERi;t ‐0 þ INTRAi;t ‐1

ð7Þ

As DIVPAT⁎i,t represents the degree of deviation of the firm's diversification path from the case of extreme AiP diversification (theoretically specified), the higher this index, the closer a firm's diversification profile is to the GO pattern, while the lower this index, the closer it is to the AiP pattern. 3. Research design: data, variables and econometric models 3.1. Database, sample selection, and description Our initial sample comprises all publicly traded U.S. companies (13,494 firms) included in the Worldscope database during the period 1998–2014.11 To mitigate survivorship bias problems, we consider both active and currently inactive12 firms. We use Worldscope as our premier source of data, and employ annual data at both the 4-digit SIC code industry segment and company level. Three additional information sources are used to supplement this data: Datastream to collect market data; the Bureau of Economic Analysis (U.S. Department of Commerce) to obtain macroeconomic data; and, Orbis to extract data on each firm's number of patents.13 As shown in Table 1, we select the sample by applying Berger and Ofek's (1995) criteria to ensure that our results are comparable to prior literature. First, we drop firm-years with any segment operating in the financial services industry (SIC codes 6000–6999) or with non-positive sales. In addition, we remove observations with missing data on total capital, total sales, and segment-level sales. We also require firms to have total sales greater than or equal to $20 million, and the sum of segment sales to be within the range of 99% to 101% of a firm's reported total sales. All these Berger and Ofek's requirements restrict our dataset to 30,952 firm-year observations, with 68% corresponding to unisegment firms and 32% to diversifiers.14 Finally, we remove firm-year observations with negative common equity (1606 observations) and outlying observations of the main variables included in our analysis15 (1391 observations). Our final dataset for estimation purposes is an unbalanced panel sample of 27,955 firm-year observations, comprising a total of 3558 companies for the 1998–2014 period. 3.2. Econometric approach, empirical models, and variables We empirically evaluate whether the pattern of diversification, as measured by DIVPAT⁎, explains the effect of diversification on market values (discounts/premiums). To deal with the potential endogeneity of the diversification decision, we alternatively apply two different estimation techniques: the Heckman two-step and the two-stage least squares (2SLS) with instrumental variables. These two approaches allow us to verify the robustness of our estimates by treating endogeneity at different levels: Whereas the former enables us to control for the potential self-selection of the decision to diversify or not to diversify, as done in prior literature, the latter allows us to explicitly account for the endogeneity associated with how a firm diversifies (diversification pattern). 3.2.1. Heckman two-step The Heckman two-step procedure allows self-selection linked to the diversification decision to be controlled (Heckman, 1979). Selectivity appears when diversification is not assigned randomly across companies, with firms either self-selecting to diversify or to remain focused (Campa and Kedia, 2002; Villalonga, 2004b; Miller, 2006). Factors affecting firms' propensity to diversify may also 11 As of 15 December 1997, the new SFAS 131 reporting standard became effective for fiscal years in the United States, replacing the previous SFAS 14. Our sample starts in 1998 to ensure homogeneity of data. 12 Firms that disappeared from the sample during the analysis period for any reason such as mergers or bankruptcy. 13 Bureau of Economic Analysis data are taken from the official website: http://www.bea.gov/national/index.htm. Worldscope and Datastream data are obtained from the Thomson ONE package by Thomson Reuters. The Orbis database is provided by Bureau van Dijk. 14 These sample proportions are similar to prior diversification studies such as Villalonga (2004b). 15 Those observations outside the range of three standard deviations from each variable sample mean.

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

321

Table 1 Distribution of observations in the sample (SIC codes classification). This Table shows the distribution of the firm-year observations between unisegment and diversified firm subsamples before and after applying Berger and Ofek's (1995) sample selection criteria. The central columns contain the number of observations dropped from the initial sample due to the report of any business segment in the financial sector or failure to meet the other Berger and Ofek criteria. Initial sample (obs.)

Final sample (obs.)

Year

Uni

Multi

Total

Financial industry

Does not meet other criteria

Uni

Multi

Total

1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Total %

3337 3467 3451 3462 3636 3836 4049 4225 4346 4409 4258 4233 4181 4161 3837 3048 2468 64,404 75.53%

1362 1501 1600 1593 1420 1311 1229 1177 1119 1046 980 948 909 922 1069 1303 1373 20,862 24.47%

4699 4968 5051 5055 5056 5147 5278 5402 5465 5455 5238 5181 5090 5083 4906 4351 3841 85,266 100.00%

1105 1163 1211 1220 1275 1328 1309 1313 1304 1306 1262 1261 1240 1240 1218 1101 1039 20,895

2045 2161 2141 2136 2095 2139 2206 2282 2231 2146 1928 2067 1932 1874 1737 1375 924 33,419

949 961 984 967 1031 1073 1185 1251 1382 1467 1542 1408 1470 1512 1390 1249 1163 20,984 67.79%

600 683 715 732 655 607 578 556 548 536 506 445 448 457 561 626 715 9968 32.21%

1549 1644 1699 1699 1686 1680 1763 1807 1930 2003 2048 1853 1918 1969 1951 1875 1878 30,952 100.00%

impact diversification value outcomes. If so, diversification variables would be correlated with the error term in the diversificationvalue models, and ordinary least squares (OLS) estimators would not prove consistent. The Heckman two-stage method considers this self-selection bias as an omitted variable problem and corrects for it. The first step of the Heckman estimation involves a probit analysis to model the firm's propensity to diversify (selection equation). It enables us to obtain self-selection correction in the form of the inverse of Mill's ratio (λ), which will be included at the second stage to correct for selection bias (Greene, 2003). The resulting estimators of this latter equation would thus reflect the net effect of the diversification strategy on the dependent variable once sample selectivity has been corrected. We specify the following selection equation, which includes a number of variables traditionally considered as drivers of the diversification decision (Campa and Kedia, 2002; Villalonga, 2004b; Dastidar, 2009): dumDIV i;t ¼ γ0 þ γ1 LTAi;t þ γ 2 EBITsalesi;t þ γ3 CAPEXsalesi;t þ γ4 PNDIV i;t þ γ 5 PSDIV i;t þ þγ 6 ChangeGDPi;t þ γ7 CONTRACTIONi;t þ ηi;t

ð8Þ

where ηi,t is an error term. The dependent variable in our selection equation is a diversification dummy (dumDIV), which equals one if the firm operates in two or more different 4-digit SIC industries, and zero otherwise. Following prior research, we assume the diversification decision can be driven by the following characteristics: • At the firm-level: firm size, estimated by the natural logarithm of the book value of total assets (LTA); profitability, approximated by the ratio of earnings before interest and taxes EBIT to total sales (EBITsales); and the firm's level of investment in current operations, proxied by capital expenditures to total sales ratio (CAPEXsales). • At the industry-level: industry attractiveness, based on both the fraction of companies in the firm's core industry that are diversified (PNDIV) and the proportion of the firm's core industry sales accounted for by diversifiers (PSDIV).16 • At the macro-economic level: economic cycle attractiveness, approximated by the real growth rates of gross domestic product, calculated as the GDP per cent change based on 2005 dollars (changeGDP); and the number of months in the year the U.S. economy was in recession (CONTRACTION). At the second stage of the Heckman procedure, our main models (outcome equations) are estimated by OLS. First, as a preliminary analysis to test the validity of our index, we relate both INTER and INTRA to a firm's excess value: EXVALi;t ¼ α þ β1 INTERi;t þ β2 INTRAi;t þ β3 LTAi;t þ β4 LDTAi;t þ β5 EBITsalesi;t þ þβ 6 CAPEXsalesit þ β7 LTA2it þ β8 λit þ β9 dumINDUSTRY it þ þβ10 dumYEARrit þ ν it

ð9Þ

where i identifies each firm, t indicates the year of observation (from 1 to 17), α and βp are the coefficients to be estimated, and νit represents the random disturbance for each observation. The dependent variable is excess value (EXVAL), calculated following the

16

We calculate these two proxies at the 4-digit SIC level.

322

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

Berger and Ofek (1995) imputed value approach, based on comparing the firm's market value against the estimated value the firm would have were all its divisions to operate as individual entities (imputed value).17 We then estimate Eq. (10) to test our hypothesis regarding the effect of the diversification pattern on a firm's excess value, where the explanatory variable is our proposed index DIVPAT⁎: 

EXVALi;t ¼ α þ β1 DIVPAT i;t þ β2 LTAi;t þ β3 LDTAi;t þ β4 EBITsalesi;t þ þβ5 CAPEXsalesi;t þ β6 LTA2i;t þ β7 λi;t þ β8 dumINDUSTRY i;t þ þβ 9 dumYEARi;t þ ν i;t

ð10Þ

In line with prior research (Berger and Ofek, 1995; Campa and Kedia, 2002), we control both Eqs. (9) and (10) for several firmcharacteristics likely to impact excess value: firm size (LTA) and its squared term (LTA2), financial leverage (proxied by the ratio of long-term debt to total assets, LDTA), profitability (EBITsales), and level of investment (CAPEXsales). Following Santaló and Becerra (2008), we also incorporate the industry effect at the 2-digit SIC code level (dumINDUSTRY). Additionally, we control for the year effect (dumYEAR) and self-selection (λ). The estimated coefficient associated with the λ term is a key point in the analysis. A significant λ coefficient will mean that the correlation between the residuals of the selection equation and the outcome equation cannot be assumed to be zero, confirming the existence of selectivity. 3.2.2. 2SLS with instrumental variables Controlling endogeneity on a level that is more specific to our research, we estimate our main models by 2SLS with instrumental variables where the diversification pattern index is explicitly considered as endogenous. 2SLS has been widely used to deal with potential endogeneity, as shown in prior literature (e.g. Bai and Elyasiani, 2013; Çolak, 2010; Estélyi and Nisar, 2016; He, 2012). This estimation procedure requires identifying a valid instrumental variable, which is correlated with the diversification pattern index (relevance condition), but uncorrelated with the error term (exclusion restriction). We create an instrumental variable for a firm's diversification pattern by using a firm's patent activity. More specifically, we use the dummy variable dumRELPATENT, which equals one if a firm's number of patents is above its corresponding core (3-digit SIC) industry median, and zero otherwise. This is a plausible instrumental variable because firms granted more patents are better protected against potential pre-emptive threats affecting underdeveloped investments in new diversifying businesses, thus making firms more likely to opt for a GO diversification pattern. Despite such a potential correlation between patent activity and a firm's diversification pattern choice, the instrumental variable is unlikely to directly impact excess value of diversification. To explore this, we perform a difference of means test between excess values of companies with a number of patents higher than their industry medians (dumRELPATENT = 1) versus excess values of companies with a number of patents lower than their industry medians (dumRELPATENT = 0). The results reveal that the mean excess value is higher for firm-year observations with a higher patent activity (0.0260) compared to those displaying low patent activity (0.0134), although the difference (0.0126) is not statistically significant (p-value = 0.3377). Consequently, we find no evidence that excess values differ depending on patent activity, which meets the exclusion restriction imposed for a valid instrument. Additionally, we include dumPATENT (which takes the value one for patent firms and zero for non-patent firms) as an additional instrumental variable to control for the existence of patents within the firm. To implement 2SLS, we conduct the following first stage regression of diversification pattern on control variables and instrumental variables: DIVPAT

 i;t

¼ γ0 þ γ1 dumPATENT i;t þ γ 2 dumRELPATENT i;t þ γ 3 LTAi;t þ γ 4 LDTAi;t þ þγ 5 EBITsalesi;t þ γ6 CAPEXsalesi;t þ γ7 LTA2i;t þ γ8 dumINDUSTRY i;t þ þγ 9 dumYEARi;t þ νi;t

ð11Þ

In the second stage, we employ the fitted values of the diversification pattern proxies from the first stage as our main independent variable and we estimate the core models (Eqs. (9) and (10)) stated earlier.18 At the bottom of the results tables, we conduct different tests to examine the suitability of our 2SLS estimations. The Durbin-WuHausman (DWH) statistic tests the endogenous relation between the diversification pattern index and excess values (Greene, 2003). The underlying assumption of this test is that OLS estimates are more efficient than 2SLS ones in the absence of endogeneity. In addition, we evaluate the power of our chosen instrument by performing a weak instrumental variable (IV) F-statistic test, which contrasts whether the instruments (dumPATENT and dumRELPATENT) are statistically and significantly different from zero when predicting a firm's diversification pattern. 17 If the excess value is negative, a discount will emerge, with diversification proving to be a value-destroying strategy. In contrast, a positive excess value will imply that the diversifier trades at a premium over its single-segment counterparts. A diversification strategy thus contributes towards enhancing a firm's value. Excess value is calculated by dividing the enterprise's value by its imputed value, and then taking the natural logarithm of this ratio. See Berger and Ofek (1995) for more details. Following Campa and Kedia's (2002) study, we compute a firm's market value (MV) as the sum of market value of equity, long-term debt, short-term debt, and preferred stock. 18 The dependent variable of the first stage in 2SLS regressions corresponds with the diversification pattern proxy included as the endogenous explanatory variable in the subsequent stage. Accordingly, whereas in the first stage of the estimation of Eq. (9) the dependent variables are INTER and INTRA alternatively, the first-stage dependent variable in Eq. (10) is DIVPAT⁎. Logically, the self-selection correction λi,t variable is omitted in 2SLS estimations.

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

323

Table 2 provides a summary of the descriptive statistics of the variables involved in the analysis for the sample. Particularly noteworthy is the negative sign for the average excess value (−0.0684) reflecting a diversification discount. As regards the diversification pattern, companies display a balanced average position, which is not strongly inclined towards either of the two extreme strategies. 4. Performance of the diversification pattern proxies: some preliminary analyses Several preliminary analyses were conducted to examine the usefulness and consistency of our proposed measurements to proxy for different investment diversification patterns.19 First, we compare our proxies (INTER, INTRA, and DIVPAT⁎) with a commonly accepted measure of concentration such as the Herfindahl index (HERF) for a number of base cases. Second, we analyse how our diversification index (DIVPAT⁎) and its two components (INTER and INTRA) relate to a firm's growth opportunities. 4.1. Diversification pattern proxies and HERF Table 3 presents a number of base cases with unlike diversification scopes (as measured by HERF) as well as heterogeneous patterns. In Columns (2) to (6), we show the distribution in percentages of total sales by business segments. For the sake of simplicity, a maximum of five segments is considered. First, these base cases serve to illustrate how INTER captures the degree of inequality in a firm's sales distribution, which is a relevant dimension vis-à-vis discriminating between AiP and GO strategies, whereas HERF reveals no differentiating power. Firms 1 to 5 are cases with different HERF but equal INTER. HERF ranges between 1 (for the unisegment company, Firm 1) to 0.2 (reported by the most diversified company, Firm 5). Although these five firms differ in their levels of diversification, they follow similar diversification behaviour, which is closer to the AiP pattern that shows a uniform distribution across business segments. In contrast to HERF, INTER does capture such a similarity between diversification strategies, thereby attributing the same INTER = 0 value to all five firms. By contrast, Firms 6 and 7 represent the opposite case: two companies with identical HERF, but dissimilar INTER values. According to HERF, Firms 6 and 7 are equally diversified (HERF = 0.520). However, their diversification strategy is completely different: Firm 6 pursues an AiP diversification strategy, reflected by a relevant presence in the two businesses in which it operates (40% and 60%), whereas Firm 7 follows a GO strategy by combining a strong participation in its core business (70%) and marginal participation (10%) in its remaining businesses. INTER captures such differences by reporting a value of 0.231 for Firm 6 (AiP pattern) and 0.808 for Firm 7 (GO pattern). Similarly, Firms 6 and 8 exhibit clear differences in their diversification models as AiP and GO, respectively. Such dissimilarity is captured by an increase in INTER, from 0.231 (Firm 6) to 0.750 (Firm 8); in turn, HERF decreases (from 0.520 to 0.4), indicating greater diversification. However, when comparing Firms 6 and 9, an increase in HERF is observed (from 0.520 to 0.660), indicating lower diversification, despite the change, here again, from an AiP diversification strategy (Firm 6, with INTER = 0.231) to a GO strategy (Firm 9, with INTER = 0.849). Finally, Firm 10 exemplifies the extreme case of a GO strategy with an INTER value equal to 0.989. However, only considering HERF might prove misleading in this case since this company would be similar to a unisegment company (HERF = 0.922). Second, the last two columns of Table 3 display how INTRA helps to correct the degree of inequality in a firm's sales distribution with the imputed sales for each segment according to the standards for each firm and its peers in that business. Firms 1, 4, and 5 are again cases with differing HERF values but the same AiP strategy as confirmed by all their segments showing above average commitment (INTRA = 1) and consequently, a same value in DIVPAT⁎ (DIVPAT⁎ = 0). In contrast, Firms 2 and 3 only have one of their respective segments with an AiP profile. As a result, INTRA counterbalances the zero value of INTER for Firms 2 and 3, leading to DIVPAT⁎ values higher than for Firms 1, 4, and 5. On the other hand, Firm 6 holds a relevant position in both of its two segments (INTRA = 1). As a consequence, despite having a more unequal distribution than Firm 2, the final value of DIVPAT* is lower (0.231 in Firm 6 versus 0.50 in Firm 2). Firms 7 to 9 allow us to see how INTRA counterbalances high levels of INTER. Whereas Firms 7 and 9 present a relevant position in only one of their segments (with INTRA being equal to 0.25 and 0.333, respectively), Firm 8 presents a relevant position in four of the five segments (INTRA = 0.800). The effect of INTRA is noticeable in the values of DIVPAT⁎, which are clearly higher for Firm 7 (DIVPAT⁎ = 1.102) and Firm 9 (DIVPAT⁎ = 1.081) than for Firm 8 (DIVPAT⁎ = 0.776). Finally, Firm 10 shows the highest value (DIVPAT⁎ = 1.272), as a result of presenting not only a great inequality among sales (INTER = 0.989), but also a typical foothold position in four of the five segments (INTRA = 0.200). 4.2. DIVPAT and a firm's growth opportunities To show how our proposed measurements indeed capture distinguishable patterns of diversification based on growth options, we conduct a number of difference-of-mean tests. Following prior literature, we use Tobin's Q (Cao et al., 2008) to proxy for the presence of growth options. We split our sample into quartiles and deciles based on Tobin's Q. We then compare the average value of the diversification pattern proxies in the first and last deciles (quartiles) – firm-year observations with high and low growth opportunities, respectively – for our diversification pattern proxies. Table 4 summarizes the results. As shown in Columns (1) to (2) of Table 4, results of the difference-of-means tests between the extreme subsamples (based on deciles) are statistically significant at the 1% level in most cases. On average, we find that firms with higher growth opportunities 19

We thank an anonymous referee for this suggestion.

324

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

Table 2 Summary statistics of variables for the full sample (1998–2014). This Table shows descriptive statistics of the variables involved in our models for the final sample of 27,955 firm-year observations of unisegment (18,651 firm-year observations) and multisegment companies (9304 firm-year observations). Some observations contain missing data for certain variables. EXVAL is the measure developed by Berger and Ofek (1995) to assess the value created by diversifying. INTER and INTRA are the twodimensional index component variables. DIVPAT⁎ is the diversification pattern index based on the Euclidean distance and with the extreme AiP diversification pattern of reference theoretically specified. Control variables are: LTA (size), EBITsales (profitability), CAPEXsales (level of investment in current operations), LDTA (financial leverage), PNDIV (fraction of firms in the firm's core industry that are diversified), PSDIV (the proportion of the firm's core industry sales accounted for by diversifiers), and changeGDP (real growth rates of gross domestic product), CONTRACTION (the number of months in the year the U.S. economy was in recession). Figures are expressed in million US$. Variable

N

Mean

Median

STD

Min.

Max.

1st quartile

3rd quartile

EXVAL EXVAL (without extremes) INTER INTRA DIVPAT⁎

24,257 19,926 27,955 27,955 27,955

−0.0684 0.0242 0.1892 0.4882 0.5963

0 0 0 0.5000 0.9187

1.0613 0.6420 0.3170 0.4596 0.5038

−4.0171 −1.3860 0 0 0

3.8082 1.3859 1 1 1.4141

−0.6212 −0.4310 0 0 0

0.5682 0.4841 0.3072 1 1

Control variables LTA EBITsales CAPEXsales LDTA PNDIV PSDIV changeGDP CONTRACTION

27,955 26,940 27,927 27,955 27,955 27,955 27,955 27,955

6.1162 0.0526 0.0685 0.1650 0.4474 0.5718 0.0214 1.6473

6.0226 0.0696 0.0325 0.1216 0.4286 0.5987 0.0240 0

1.9260 0.2101 0.1139 0.1711 0.2421 0.3082 0.0176 2.7953

0.9317 −1.4732 0 0 0 0 −0.0280 0

11.8794 1.4670 0.9946 0.7834 1 1 0.0470 9

4.6440 0.0113 0.0162 0.0016 0.2727 0.3457 0.0160 0

7.4653 0.1384 0.0663 0.2783 0.6000 0.8319 0.0330 3

have significantly greater values of both INTER and DIVPAT⁎, consistent with our expectations that these measurements directly proxy for options-based diversification strategies. Conversely, this same subsample displays lower values of INTRA on average, which supports the idea that a higher commitment in the portfolio in which the firm operates moves away from a pattern of growth options diversification. Results are robust to the use of quartiles of Tobin's Q to identify the groups of firm-year observations with the highest and lowest level of growth options (see Columns (3) and (4) of Table 4).20 5. Empirical findings 5.1. Propensity to diversify (first stages of Heckman and 2SLS) Table 5 contains the probit estimation for the selection equation (Eq. 8) as the first step in the Heckman method. Estimations in Columns (2) to (4) extend the probit specification (1) by incorporating lags and year dummies. Goodness-of-fit (pseudo-R2) ranges between 0.17 and 0.18, comparable to prior literature (e.g. Villalonga, 2004b). Among firm-characteristics, the level of investments (CAPEXsales) shows a negative coefficient, which is statistically significant in estimations without lags. This result suggests that companies with low investment levels are more prone to diversify. Firm size (LTA) has a positive and highly significant coefficient in the models where lagged variables are omitted, indicating that larger companies are more likely to incorporate multiple business units. Finally, firm profitability (EBITsales) is negative and statistically significant in all estimates, suggesting that less profitable enterprises are more liable to engage in this strategy. All of these results are in line with previous works such as Villalonga (2004b) and Dastidar (2009). Results concerning the effect of industry factors on the propensity to diversify are robust to the alternative estimations. Consistent with results in Campa and Kedia (2002) and Villalonga (2004b), we find evidence that a greater presence of diversified firms in the core industry positively impacts the decision to diversify. As far as macroeconomic variables are concerned, there is no significant impact on diversification likelihood, with CONTRACTION only borderline significant in Column (3). There is also weak evidence concerning the relevance of the changeGDP variable. In the probit specification in Columns (1) and (2), changeGDP is positively associated with the diversification decision, suggesting that companies are more likely to diversify during cycles of economic growth. However, this variable does not retain its statistical significance in the remaining specifications. In sum, we find that characteristics at the firm and industry levels are the main drivers of the diversification decision. Moving on to the second step of Heckman's approach and performing the estimations of our outcome equations, we take the specification of the selection equation in Column (1). In this way, we exclude lagged values of firm variables and time dummies that lack statistical significance in most cases, while minimising loss of observations for subsequent analyses. This probit ensures at least four exclusion restrictions since the variables PNDIV, PSDIV, changeGDP, and CONTRACTION are included in the selection equation but not in the outcome equations, thus mitigating potential collinearity problems. As far as the first stage estimates of 2SLS are concerned, Table 6 reports estimation results for Eq. (11) with the alternative proxies for the diversification pattern. As expected, dumPATENT and dumRELPATENT positively relate to a firm's GO diversification pattern as

20

These results also prove robust to the use of book-to-market ratio as a proxy for growth options (Da et al., 2012). Results are available upon request.

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

325

Table 3 Base cases. This Table presents a number of base cases with different diversification scopes (as measured by HERF) as well as heterogeneous patterns. HERF is the Herfindahl index. INTER and INTRA denote the inter-segment dimension and intra-segment dimension of the diversification pattern index, respectively. DIVPAT⁎ is the diversification pattern index based on the Euclidean distance and with the extreme AiP diversification pattern of reference theoretically specified.

Firm 1 Firm 2 Firm 3 Firm 4 Firm 5 Firm 6 Firm 7 Firm 8 Firm 9 Firm 10

Segment 1

Segment 2

100.00% 50.00% 33.33% 25.00% 20.00% 40.00% 10.00% 10.00% 10.00% 1.00%

50.00% 33.33% 25.00% 20.00% 60.00% 10.00% 10.00% 10.00% 1.00%

Segment 3

33.33% 25.00% 20.00% 10.00% 10.00% 80.00% 1.00%

Segment 4

Segment 5

25.00% 20.00%

20.00%

70.00% 10.00%

60.00%

1.00%

96.00%

HERF

INTER

INTRA

DIVPAT⁎

1 0.5 0.333 0.250 0.200 0.520 0.520 0.400 0.660 0.922

0.000 0.000 0.000 0.000 0.000 0.231 0.808 0.750 0.849 0.989

1 0.500 0.333 1 1 1 0.250 0.800 0.333 0.200

0.000 0.500 0.670 0.000 0.000 0.231 1.102 0.776 1.081 1.272

measured by a higher DIVPAT⁎ or INTER, or a lower INTRA. If we focus on the DIVPAT⁎ results displayed in Columns (3) and (6), we notice that dumPATENT displays both a higher statistical and economic significance, suggesting that the mere existence of patent protection (more than its level) encourages firms to engage in greater GO diversification. Consistent with real options arguments, we also find that bigger firms, which are more likely to have already exhausted their growth options by developing them into assets-inplace, are also more prone to engage in a GO pattern, probably as a way to open up new growth opportunities. More leveraged firms are also more inclined to choose this GO diversification, which could be explained by financial restrictions that prevent immediate funding of full-scale commitments. 5.2. Components of the diversification pattern index and firm value As a preliminary analysis, we test the impact of our index dimensions on a firm's value (Eq. (9)). The results obtained when using the alternative estimation procedures are displayed in Table 7. Evidence supports the need to control for endogeneity. First, the λ coefficient is strongly statistically significant in most regressions (except for Column (11) estimations), thus allowing us to reject the null hypothesis that the correlation between the residuals of the selection equation and the outcome equation is zero. This evidence confirms that our sample suffers from self-selection bias, thus justifying Heckman's approach. Second, the DWH test in all cases (pvalue = 0.0000) rejects the null hypothesis that the components INTER and INTRA of our diversification pattern index are exogenous, thereby supporting the use of instrumental variable techniques such as 2SLS. Table 7 presents the estimates with INTER and INTRA dimensions together. We find clear evidence of a negative effect of INTRA with excess value, which is statistically significant at the 1% level in all cases (p-value = 0.000). This result is consistent with our arguments since holding major commitments in many businesses is negatively related to GO diversification. We also find that INTER is positively associated with excess value when controlling for endogeneity. This result also supports our hypothesis since the greater the inequality in the distribution of business activity, the closer the firm's strategy is to GO diversification, despite the economic significance of INTER proving to be lower than the INTRA dimension when explaining excess value.21 The results are robust after dropping extreme excess value observations (above 1.386 or below −1.386) from the sample (Berger and Ofek, 1995) as shown in Columns (7) to (12) of Table 7. Moreover, as indicated by the Wald test or F-statistic reported at the bottom of the tables, variables are jointly significant above the 1% level. The weak IV partial F-test for weak instruments is statistically significant at the 1% level (p-value = 0.000) and thereby confirms the validity of the instrumental variables chosen (dumPATENT and dumRELPATENT). 5.3. Diversification pattern and firm value Columns (1) to (12) in Table 8 provide interesting insights into the relevance of the diversification pattern for explaining value outcome (Eq. (10)). Our research hypothesis receives strong support since our evidence shows that DIVPAT⁎ explains part of firms' excess values. Across all estimates, we find a significant (at the 1% level in all cases) and positive effect of DIVPAT⁎ on excess value, indicating that a strategy further away from the ‘extreme’ AiP diversification profile implies higher excess value. Similarly, as a firm's diversification strategy approaches a GO diversification strategy (as measured by a longer Euclidean distance and thus a higher DIVPAT⁎), diversification becomes a more value-enhancing strategy. These results suggest that there is a higher value generated by a diversification strategy aimed not only at exploiting but also at seeding new opportunities in further businesses. Results are robust to the addition of industry and year dummies as shown in Columns (4) to (6) of Table 8, and to the exclusion of extreme excess value observations from the sample as reported in Columns (7) to (12) of the same table. 21 It should also be noted that INTER loses its statistical significance in 2SLS regressions when taken together with INTRA. We discard any multicollinearity problems since the correlation between INTER and INTRA is −0.2439. For additional robustness checks, we evaluate how each dimension when individually considered impacts a firm's excess value. Table A.1 of the Appendix reports these individual estimates. Results prove robust, although when considered separately, INTER is statistically significant at the 5% level, thus confirming the relevance of inequality in the firm's level of participation in its businesses when accounting for diversification value outcomes.

326

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

Table 4 Difference of means results. This table summarizes the difference-of-means tests of the average value of the diversification pattern proxies between firm-year observations belonging to the first and last deciles (quartiles) of growth opportunities for the final sample of 27,955 firm-year observations. Tobin's Q proxies for growth opportunities. DIVPAT⁎ is the diversification pattern index based on the Euclidean distance and with the extreme AiP diversification pattern of reference theoretically specified. INTER and INTRA denote the inter-segment dimension and intra-segment dimension of the diversification pattern index, respectively.

INTER INTRA DIVPAT⁎

INTER INTRA DIVPAT⁎

Q (1st decile) (1)

Q (10th decile) (2)

Difference (1)–(2)

0.1808 0.4844 0.5936

0.1731 0.5811 0.5015

0.0077 −0.0967*** 0.0921***

Q (1st quartile) (3)

Q (4th quartile) (4)

Difference (3)–(4)

0.1790 0.4703 0.6080

0.1792 0.5542 0.5300

−0.0002 −0.0839*** 0.0780***

Finally, with regard to our control variables, LTA, LTA2, EBITsales, and CAPEXsales show statistical significance above the 1% level in all estimations, with all of them displaying a positive impact on excess value, as consistent with prior studies. The F-test and the Wald test indicate that variables display joint significance in all regressions. In the Heckman regressions, the λ coefficient contains statistical significance, which is even above 1% in certain cases, thus confirming the existence of self-selection bias in the sample. The DWB tests also confirm the endogeneity of the DIVPAT⁎ variable. In 2SLS estimates, the partial F test for weak instruments supports the validity of the instrumental variables. 5.4. Robustness analyses To check the robustness of our results, we conduct the following tests. First, we replace the theoretically defined, extreme AiP diversification profile by using a calibration sample based on the median scores of each dimension of our index for a subsample of companies with lower growth opportunities. Second, following prior literature, we examine the consistency of our empirical findings when controlling for a firm's diversification scope and growth opportunities. We then go further in testing our hypothesis and Table 5 Firms' propensity to diversify [first stage of Heckman's estimation] (Eq. 8). This table shows probit estimation results for the selection equation (Eq. 8) as the first stage of Heckman's procedure. The dependent variable takes the value one when the firm is diversified and zero otherwise. The pseudo-R square indicates the goodness of fit. Standard error is shown in parentheses under coefficients. ****, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively. First stage Heckman regressions Dependent variable: dumDIV

Constant Firm characteristics LTA EBITsales CAPEXsales

(1) baseline model

(2) with lags

(3) with dummy years

(4) with lags and dummy years

-2.7766*** (0.0504)

−2.7925*** (0.0566)

−2.9720*** (0.1533)

−2.9658*** (0.1663)

0.1237*** (0.0054) −0.1830*** (0.0477) −0.8245*** (0.0908)

−0.0280 (0.0483) −0.1288* (0.0750) −0.1803 (0.1943) 0.1546*** (0.0482) −0.0787 (0.0728) −0.7274*** (0.1921)

0.1275*** (0.0055) −0.1845*** (0.0479) −0.8334*** (0.0912)

−0.0227 (0.0485) −0.1304* (0.0752) −0.1455 (0.1952) 0.1537*** (0.0484) −0.0886 (0.0731) −0.7769*** (0.1937)

2.1368*** (0.0552) 0.6602*** (0.0413)

2.1532*** (0.0614) 0.6819*** (0.0462)

2.1433*** (0.0567) 0.6581*** (0.0416)

2.1715*** (0.0630) 0.6798*** (0.0464)

1.5920** (0.8277) 0.0040 (0.0051) NO 23,835 −11,089.892 0.1748

1.6857* (0.9413) 0.0041 (0.0056) NO 19,110 −8948.8359 0.1791

8.4731 (7.7022) −0.0396* (0.0212) YES 23,835 −11,075.639 0.1759

6.1425 (8.3062) −0.0368 (0.0229) YES 19,110 −8931.9605 0.1806

LTAt-1 EBITsalest-1 CAPEXsalest-1 Industry characteristics PNDIV PSDIV Macroeconomic characteristics changeGDP CONTRACTION Year dummies N. of obs. Log. Likelihood Pseudo-R2

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

327

Table 6 Firms' diversification pattern decision choice [first stage of 2SLS estimation] (Eq. 11). This Table shows estimation results for the first stage of 2SLS (Eq. 11). The dependent variable of the first stage in 2SLS regressions corresponds to the diversification pattern proxy included as the endogenous explanatory variable in the subsequent stage, either both INTER and INTRA (Eq. (9)) or DIVPAT (in the remaining equations). The F-test contrasts the null hypothesis of no joint significance of the explanatory variables. The adjusted-R square indicates the goodness of fit. Standard error is shown in parentheses under coefficients. ****, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively. First stage 2SLS regressions Dep. variable

INTRA (1)

INTER (2)

DIVPAT⁎ (3)

INTRA (4)

INTER (5)

DIVPAT⁎ (6)

Constant

1.0285*** (0.0282) −0.0832*** (0.0084) −0.0243** (0.0113) −0.1006*** (0.0088) −0.0266 (0.0190) 0.1772*** (0.0141) −0.5081*** (0.0262) 0.0041*** (0.0007) NO NO 23,285 270.74*** 0.0750

0.0725*** (0.0180) 0.0377*** (0.0054) 0.0554*** (0.0072) −0.0113** (0.0056) 0.0577*** (0.0121) 0.0130 (0.0090) −0.1192*** (0.0167) 0.0020*** (0.0004) NO NO 23,285 66.20*** 0.0192

0.0064 (0.0307) 0.0960*** (0.0091) 0.0487*** (0.0122) 0.0976*** (0.0096) 0.0573*** (0.0206) −0.1670*** (0.0154) 0.4394*** (0.0285) −0.0035*** (0.0007) NO NO 23,285 247.22*** 0.0689

1.4968*** (0.0299) −0.0411*** (0.0085) 0.0040 (0.0106) −0.1194*** (0.0084) −0.1109*** (0.0186) 0.1423*** (0.0133) 30.7096*** (0.0293) 0.0058*** (0.0006) YES YES 23,285 79.82*** 0.2025

−0.3532*** (0.0152) 0.0065 (0.0043) 0.0150*** (0.0054) −0.0002 (0.0042) 0.0196** (0.0094) 0.0245*** (0.0068) 0.0076 (0.0149) 0.0002 (0.0003) YES YES 23,285 266.64*** 0.4611

−0.6368*** (0.0309) 0.0458*** (0.0088) 0.0041 (0.0109) 0.1206*** (0.0087) 0.1188*** (0.0192) −0.1280*** (0.0138) 0.6992*** (0.0303) −0.0060*** (0.0007) YES YES 23,285 117.48*** 0.2728

dumPATENT dumRELPATENT LTA LDTA EBITsales CAPEXsales LTA2 Year dummies Industry dummies N. of obs. F-test Adjusted-R2

evaluate the robustness of our results when considering the risk of pre-emption. Finally, we control for the dynamic nature of the investment pattern by including several lags of our index. 5.4.1. Controlling for an alternative specification of the extreme AiP diversification profile Rather than predetermining the extreme AiP diversification profile theoretically, as detailed by Eq. (7), we re-estimate our diversification pattern index based on an empirically specified reference. To achieve this, we use a calibration sample (Venkatraman, 1989; Venkatraman and Prescott, 1990) comprised of the bottom 10% of firms according to growth opportunities.22 Growth opportunities are proxied by Tobin's Q (Cao et al., 2008). The AiP diversification reference point is then determined by the median scores along both index dimensions (INTER and INTRA) for this calibration sample. Table 9 reports a statistical summary of both dimensions for this subsample. To construct an alternative proxy for the diversification profile index, the vector of scores for the AiP diversification extreme profile is then replaced in Eq. (6) by the median scores of INTER and INTRA in the calibration sample determined by Tobin's Q. The redefined diversification pattern index is denoted by DIVPATQ.23 The results are displayed in Table 10. We find that our results are robust to the alternative specification of the reference pattern DIVPATQ as well as the elimination of extreme excess value observations.24 The statistical significance of the lambda (λi) coefficient in all Heckman estimates points to the existence of self-selection. The DWB test also supports the endogeneity of the diversification pattern DIVPATQ. The partial weak IV F-test also supports the validity of the instrumental variable. 5.4.2. Controlling for the diversification status We further check the robustness of previous empirical findings and redefine Eq. (10) to include the level of diversification as a control variable, approximated by the modified Herfindahl index (MHERF = 1-HERF)25: 

EXVALi;t ¼ α þ β1 DIVPAT i;t þ β2 MHERF i;t þ β3 LTAi;t þ β4 LDTAi;t þ þ β5 EBITsalesi;t þ β6 CAPEXsalesi;t þ β7 LTA2i;t þ þ β8 dumINDUSTRY i;t þ β9 dumYEARi;t þ νi;t

ð12Þ

Columns (1) to (5) in Table 11 report these additional sensitivity tests. The first two columns present estimation results for the core model of Eq. (12) only with control variables. In line with prior literature, our sample also shows a diversification discount. Columns 22

The most extreme cases of AiPD should imply the lowest growth opportunity values, as this pattern is primarily aimed at exercising options in one full-scale step. Following Vekatraman and Prescott (1990), in regressions where the index based on the calibration sample is used, both the bottom (calibration sample) and top 10% of firms according to their level of growth opportunities are excluded from the estimation sample. 24 Results are available in the Appendix (Table A.2). 25 Note that in this case, the greater the level of diversification, the higher the modified Herfindahl index. 23

328 Table 7 The dimensions of the index and Excess Value [Eq. (9)]. This Table shows the estimation results of Eq. (9). In the Heckman estimates, the selection equation (Column 1 in Table 5) models a firm's propensity to diversify, and the Inverse Mills Ratio ((λi) is included as an additional regressor to correct potential self-selection bias in the sample. EXVAL is regressed on INTER and INTRA. Firm size (LTA) and its squared term (LTA2), financial leverage (LDTA), profitability (EBITsales), level of investment (CAPEXsales), industry effect (dumINDUSTRY), and time effect (dumYEAR) are controlled in the estimations. The Wald and F tests contrast the null hypothesis of no joint significance of the explanatory variables. The Durbin-Wu-Hausman statistic tests for endogeneity of INTER and INTRA variables. The instrumental variable (IV) partial F-statistic tests for weak instruments. Standard error is shown in parentheses under coefficients. ****, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively. Model Excess Value = f(INTER, INTRA, control variables) Dependent variable: EXVAL

Dependent variable: EXVAL (without extremes)

INTER and INTRA

INTER INTRA Control variables LTA LDTA EBITsales CAPEXsales LTA2

INTER and INTRA with industry and time dummies

OLS (1)

Heckman (2)

2SLS (3)

OLS (4)

Heckman (5)

2SLS (6)

OLS (7)

Heckman (8)

2SLS (9)

OLS (10)

Heckman (11)

2SLS (12)

−2.1947*** (0.0574) −0.2240*** (0.0221) −0.4227*** (0.0141)

−2.17111*** (0.1285) 0.2466*** (0.0451) −0.7603*** (0.0451)

0.3524 (0.4734) 1.2750 (0.8318) −3.1292*** (0.3976)

−2.1280*** (0.0672) −0.0975** (0.0285) −0.4712*** (0.0145)

−2.5856*** (0.1394) 0.2846*** (0.0396) −0.7215*** (0.0445)

8.8307*** (2.7333) 8.2228 (5.3134) −5.8723*** (1.2056)

−0.7467*** (0.0430) −0.1761*** (0.0156) −0.2594*** (0.0098)

−0.8676*** (0.0982) 0.1619*** (0.0339) −0.3911*** (0.0339)

0.5892** (0.2940) 0.6193 (0.4881) −1.8017*** (0.2703)

−0.7499*** (0.0493) −0.0992** (0.0202) −0.2728*** (0.0102)

−1.2119*** (0.1057) 0.1720*** (0.0302) −0.3939*** (0.0340)

6.0357** (2.3905) 3.8474 (3.1092) −4.2574*** (1.2705)

0.6679*** (0.0185) 30.2167*** (0.0391) 0.6813*** (0.0296) 0.4716*** (0.0553) −0.0421*** (0.0014)

0.6206*** (0.0359) −0.1851*** (0.0796) 1.0568*** (0.0702) 0.9044*** (0.1336) −0.0385*** (0.0026) −0.1805*** (0.0300) NO NO 23,835 17,843 5992 1626.71***

0.3894*** (0.0580) −0.2690*** (0.0822) 1.1843*** (0.0981) −0.6501** (0.2820) −0.0335*** (0.0040)

0.6586*** (0.0184) −0.2544*** (0.0406) 0.7242*** (0.0293) 1.2472*** (0.0652) −0.0400*** (0.0014)

−0.0071 (0.1614) −0.9403*** (0.2159) 1.3245*** (0.2336) −2.6489*** (0.8889) −0.0101 (0.0086)

0.2506*** (0.0136) −0.2261*** (0.0271) 0.3611*** (0.0219) 0.1501*** (0.0383) −0.0148*** (0.0010)

0.2480*** (0.0137) −0.2365*** (0.0286) 0.3887*** (0.0220) 0.5752*** (0.0470) −0.0137*** (0.0010)

YES YES 23,285

NO NO 19,191

NO NO 19,191

YES YES 19,191

0.2268*** (0.0267) −0.0596 (0.0584) 0.5928*** (0.0539) 1.0601*** (0.1142) −0.0109*** (0.0019) −0.0361 (0.0224) YES YES 19,741 14,751 4990 1326.01***

−0.0875 (0.1169) −0.7902*** (0.2118) 1.0041 (0.2172) −2.0727** (0.8637) −0.0016 (0.0055)

YES YES 23,285

0.2191*** (0.0274) −0.0781 (0.0581) 0.5672*** (0.0555) 0.3676*** (0.0981) −0.0118*** (0.0020) −0.0753*** (0.0213) NO NO 19,741 14,751 4990 688.16***

0.1299*** (0.0345) −0.3017*** (0.0523) 0.6776*** (0.0710) −0.4519*** (0.1711) −0.0124*** (0.0023)

NO NO 23,285

0.6040*** (0.0347) −0.1224 (0.0788) 1.0914*** (0.0673) 1.9900*** (0.1487) −0.0352*** (0.0025) −0.1110*** (0.1394) YES YES 23,835 17,843 5992 2649.34***

Inverse Mills Ratio (λi) Industry dummies Year dummies No. of obs. No. censored obs. No. uncensored obs. Wald Chi2 χ2 F-statistic Adjusted-R2 Durbin-Wu-Hausman test p-value Weak IV F-statistic

INTER and INTRA

NO NO 23,285

1560.97***

790.49*** 0.1918

546.83***

94.80*** 0.2320

651.94***

329.67*** 0.1071

YES YES 19,191

215.93***

41.29*** 0.1360

0.000

0.000

0.000

0.000

INTER: 66.20*** INTRA: 270.74***

INTER: 266.64*** INTRA: 79.82***

INTER: 52.14*** INTRA: 187.34***

INTER: 215.63*** INTRA: 63.98***

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

Constant

INTER and INTRA with industry and time dummies

Table 8 Diversification pattern index (theoretically specified) and Excess Value [Eq. (10)]. This Table shows the estimation results of Eq. (10). In the Heckman estimates, the selection equation (Column 1 in Table 5) models a firm's propensity to diversify, and the Inverse Mills Ratio ((λi) is included as an additional regressor to correct potential self-selection bias in the sample. EXVAL is regressed on DIVPAT⁎, which is the diversification pattern index based on the Euclidean distance and with the extreme AiP diversification pattern of reference theoretically specified. Firm size (LTA) and its squared term (LTA2), financial leverage (LDTA), profitability (EBITsales), level of investment (CAPEXsales), industry effect (dumINDUSTRY), and time effect (dumYEAR) are controlled in the estimations. The Wald and F tests contrast the null hypothesis of no joint significance of the explanatory variables. The DurbinWu-Hausman statistic tests for endogeneity of the DIVPAT⁎ variable. The instrumental variable (IV) partial F-statistic tests for weak instruments. Standard error is shown in parentheses under coefficients. ****, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively. Model Excess Value = f(DIVPAT⁎, control variables) Dependent variable: EXVAL

Dependent variable: EXVAL (without extremes)

Constant DIVPAT⁎ Control variables LTA LDTA EBITsales CAPEXsales LTA2

Baseline index

Baseline index with industry and time dummies

OLS (1)

Heckman (2)

2SLS (3)

OLS (4)

Heckman (5)

2SLS (6)

OLS (7)

Heckman (8)

2SLS (9)

OLS (10)

Heckman (11)

2SLS (12)

−2.6736*** (0.0557) 0.3159*** (0.0127)

−3.1127*** (0.1262) 0.8043*** (0.0501)

−2.8043*** (0.1020) 3.2401*** (0.3226)

−2.5360*** (0.0639) 0.4193*** (0.0140)

−3.4912*** (0.1342) 0.8066*** (0.0495)

0.8314 (0.6126) 5.7077*** (0.9247)

−1.0348*** (0.0422) 0.1827*** (0.0089)

−1.3648*** (0.0973) 0.4704*** (0.0379)

−1.2373*** (0.0766) 1.8492*** (0.2388)

−0.9761*** (0.0474) 0.2400*** (0.0099)

−1.7204*** (0.1030) 0.4859*** (0.0379)

0.9291* (0.5296) 3.9010*** (0.9846)

0.6836*** (0.0186) −0.2384*** (0.0394) 0.6524*** (0.0297) 0.5793*** (0.0553) −0.0433*** (0.0014)

0.6329*** (0.0360) −0.1986** (0.0799) 1.0254*** (0.0701) 0.9233*** (0.1339) −0.0387*** (0.0026) −0.1577*** (0.0298) NO NO 23,835 17,843 5992 1582.60***

0.3737*** (0.0479) −0.2990*** (0.0717) 1.1864*** (0.0797) −0.6354*** (0.1671) −0.0323*** (0.0029)

0.6645*** (0.0185) −0.2544*** (0.0407) 0.7078*** (0.0293) 1.2876*** (0.0653) −0.0402*** (0.0014)

0.0039 (0.1255) −0.8063*** (0.1454) 1.4257*** (0.1479) −2.4092*** (0.6690) −0.0076*** (0.0069)

0.2599*** (0.0137) −0.2414*** (0.0273) 0.3396*** (0.0221) 0.2200*** (0.0384) −0.0155*** (0.0010)

0.2518*** (0.0137) −0.2359*** (0.0287) 0.3763*** (0.0220) 0.5954*** (0.0471) −0.0138*** (0.0010)

YES YES 23,285

NO NO 19,191

NO NO 19,191

YES YES 19,191

0.2295*** (0.0266) −0.0624 (0.0583) 0.5836*** (0.0536) 1.0584*** (0.1140) −0.0109*** (0.0019) −0.0332 (0.0222) YES YES 19,741 14,751 4990 1348.60***

−0.0826 (0.0980) −0.6985** (0.1487) 0.9811*** (0.1743) −1.7713** (0.6502) 0.0004 (0.0048)

YES YES 23,285

0.2228*** (0.0273) −0.0863 (0.0580) 0.5576*** (0.0553) 0.3641*** (0.0979) −0.0119*** (0.0020) −0.0715*** (0.0211) NO NO 19,741 14,751 4990 700.51***

0.1222*** (0.0303) −0.3269*** (0.0475) 0.6742*** (0.0605) −0.4256*** (0.1127) −0.0118*** (0.0018)

NO NO 23,285

0.6126*** (0.0347) −0.1237 (0.0789) 1.0673*** (0.0671) 2.0132*** (0.1487) −0.0353*** (0.0025) −0.0955*** (0.0312) YES YES 23,835 17,843 5992 2629.14***

Inverse Mills Ratio (λi) Industry dummies Year dummies No. of obs. No. censored obs. No. Uncensored obs. Wald Chi2 χ2 F-statistic Adjusted-R2 Durbin-Wu-Hausman test p-value Weak IV F-statistic

Baseline index with industry and time dummies

NO NO 23,285

1475.65***

855.93*** 0.1805

879.52***

93.37*** 0.2269 0.000 59.3241***

611.97***

329.92*** 0.0933 0.000 18.9761***

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

Baseline index

YES YES 19,191

307.42***

39.92*** 0.1305 0.0000 37.8206***

0.000 7.8559***

329

330

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

Table 9 Summary statistics of two-dimensional index component variables for the calibration sample (1998–2014). This Table displays the summary statistics of the two-dimensional index component variables (INTER and INTRA). INTER denotes the Gini Index; and INTRA is the ratio of a firm's segments displaying commitment ratios above or equal to one, over the total number of a firm's segments. Calibration sample defined by Tobin's Q: Bottom 10% of sample firms according to growth opportunities

INTER INTRA

N

Mean

Median

STD

2614 2614

0.1805 0.5850

0 1

0.3143 0.4510

(3) to (5) show results for the diversification pattern index DIVPAT⁎ when simultaneously adding the scope dimension (MHERF). Results concerning the pattern index are robust across all estimations; here again, bearing out that the closer a firm's diversification is to a GO pattern, the higher its excess value. These empirical findings are also robust to the alternative specification of the diversification pattern index DIVPATQ.26 Interestingly, once the pattern of diversification is accounted for in the regressions, the documented discount remains statistically significant, becoming a premium in Heckman estimates (Column (4)). To ensure comparability with prior research, we re-estimate Eq. (12) by using another standard measure of diversification widely used as a dummy variable indicating whether the firm is either focused or diversified (dumDIV). Our results remain similar, being robust across the alternative specifications of the diversification pattern index DIVPAT⁎ and DIVPATQ.27 Taken all together, our empirical findings reveal the diversification pattern as an important dimension that explains a firm's excess value, even after controlling for the diversification scope or the status of the firm either as focused or as diversified. Conflicting evidence regarding the impact of diversification on firm value may be partly explained by the fact that prior analyses mix the effects of different dimensions of diversification, namely, scope and investment pattern. These may require proper separate identification and measurement to investigate the overall impact of diversification more accurately. 5.4.3. Controlling for the diversification status and growth opportunities A further robustness check is to run our regressions controlling for a firm's growth opportunities together with diversification level (Eq. (13)). Given the potentially high correlation between Tobin's Q and excess value, we use an R&D based proxy for growth opportunities in the regressions. Following Grullon et al. (2012), growth opportunities are proxied by R&D intensity (RDA, defined as the ratio of annual R&D expenditures and beginning-of-year book assets)28: 

EXVALi;t ¼ α þ β1 DIVPAT i;t þ β2 MHERF i;t þ β3 RDAi;t þ β4 LTAi;t þ þ β5 LDTAi;t þ β6 EBITsalesi;t þ β7 CAPEXsalesi;t þ β8 LTA2i;t þ þ β9 dumINDUSTRY i;t þ β10 dumYEARi;t þ ν i;t

ð13Þ

Columns (6) to (8) of Table 11 summarise the results of Eq. (13). Our hypothesis also receives strong support after controlling for growth opportunities in the regressions. Across the alternative index specifications, the diversification pattern proves statistically significant (above the 1% level) to explain a firm's excess value. Its coefficient displays a positive sign, again suggesting that a further deviation (in terms of Euclidean distance) from an AiP strategy, and thereby a closer profile to GO diversification, is more value enhancing. Results are robust for DIVPATQ.29 5.4.4. Controlling for pre-emption risk To explore the possible effect of industry heterogeneity on a firm's pattern of diversification, we control for the threat of pre-emption. In highly competitive industries, sequential small-scale investments that characterise GO diversification may prove counter-valuable. Such an investment strategy may imply the loss of first-mover advantage, thereby reducing the growth option lifespan and its value. As a result, we expect a GO diversification strategy to be less valuable the higher the risk of pre-emption. Following Folta and Miller (2002), the risk of pre-emption is approximated by the number of rivals actively operating in the same product domain. We gather yearly data on the total number of U.S. firms by NAICS codes from the U.S. Census Bureau and then match NAICS codes with SIC codes.30 Our variable to proxy risk of pre-emption is PREEMPT, calculated as the natural logarithm of the number of firms operating in the same 2digit SIC code industry as the core business of the corresponding firm. We split our full sample into terciles/quartiles of PREEMPT levels and re-estimate Eq. (12). Table 12 displays the coefficients of the alternative estimations. Across all subsamples, the coefficient on the diversification pattern is positive and statistically significant at the 1% level in most cases, thus confirming our hypothesis that GO diversification leads to greater excess values. Interestingly, this coefficient is not only statistically significant but also smaller the higher the risk of pre-emption in the industry. For example, in Heckman estimates, the coefficient on DIVPAT⁎ decreases about 42% within the subsample of firm-years in the first tercile of PREEMPT (high pre-emption) compared to the third tercile (low pre-emption). Consistent with our arguments, a greater threat of pre-emption

26 27 28 29 30

Results are available in the Appendix (Columns (1) to (3) in Table A.3). Table A.4 of the Appendix contains these additional estimations. Nevertheless, we have checked that results remain similar when using Tobin's Q to proxy for growth options. These results are available upon request. Results are available in the Appendix (Columns (4) to (6) in Table A.3). Data are currently only available until 2013. As a result, estimates including PREEMPT are carried out with firm-year observations from 1998 to 2013.

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

331

Table 10 Diversification pattern index (based on calibration sample) and Excess Value [Eq. (10)]. This Table shows the estimation results of Eq. (10). In the Heckman estimates, the selection equation (Column 1 in Table 5) models a firm's propensity to diversify, and the Inverse Mills Ratio ((λi) is included as an additional regressor to correct potential self-selection bias in the sample. EXVAL is regressed on the alternative specification of the diversification pattern index based on a calibration sample of Tobin's Q: DIVPATQ. Firm size (LTA) and its squared term (LTA2), financial leverage (LDTA), profitability (EBITsales), level of investment (CAPEXsales), industry effect (dumINDUSTRY), and time effect (dumYEAR) are controlled in the estimations. The Wald and F tests contrast the null hypothesis of no joint significance of the explanatory variables. The Durbin-Wu-Hausman statistic tests for endogeneity of DIVPATQ. The instrumental variable (IV) partial F-statistic tests for weak instruments. Standard error is shown in parentheses under coefficients. ****, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively. Model EXVAL = f(DIVPATQ, control variables) Dependent variable: EXVAL

Constant DIVPATQ Control variables LTA LDTA EBITsales CAPEXsales LTA2

Index based on calibration sample of Tobin's Q

Index based on calibration sample of Tobin's Q with industry and time dummies

OLS (1)

Heckman (2)

2SLS (3)

OLS (4)

Heckman (5)

2SLS (6)

−2.6541*** (0.0564) 0.2949*** (0.0156)

−2.5283*** (0.1267) 0.4388*** (0.0442)

−3.2861*** (0.1393) 3.3248*** (0.4763)

−2.5319*** (0.0640) 0.3846*** (0.0166)

−3.0168*** (0.1369) 0.5950*** (0.0499)

−0.5712 (0.6237) 7.5389*** (2.1434)

0.6421*** (0.0185) 0.1127** (0.0388) 0.5024*** (0.0311) 0.4698*** (0.0535) −0.0400*** (0.0014)

0.5345*** (0.0358) 0.1861** (0.0781) 0.7352*** (0.0714) 0.8968*** (0.1281) −0.0316*** (0.0026) −0.1356*** (0.0288) NO NO 19,362 14,391 4971 1065.59***

0.4206*** (0.0473) 0.0316 (0.0685) 0.8153*** (0.0729) −0.2317 (0.1440) −0.0319*** (0.0028)

0.6112*** (0.0184) 0.1088*** (0.0401) 0.5369*** (0.0306) 1.1871*** (0.0631) −0.0362*** (0.0014)

−0.0108 (0.1959) −0.3929** (0.1999) 1.0176*** (0.1756) −1.8920** (0.9452) −0.0045 (0.0106)

NO NO 18,889

YES YES 18,889

0.5167*** (0.0345) 0.2681*** (0.0770) 0.7747*** (0.0684) 1.8171*** (0.1420) −0.0288*** (0.0025) −0.0848*** (0.0304) YES YES 19,362 14,391 4971 1940.59***

Inverse Mills Ratio (λi) Industry dummies Year dummies No. of obs. No. censored obs. No. uncensored obs. Wald Chi2 χ2 F-statistic Adjusted-R2 Durbin-Wu-Hausman test p-value Weak IV F-statistic

NO NO 18,889

1363.78***

717.99*** 0.1855

YES YES 18,889

511.30***

80.03*** 0.2364 0.000 30.4067***

0.000 6.1156***

by competitors detracts value from GO diversification. Our results concur with prior real options literature showing the detrimental effects of the risk of pre-emption on growth option value in a variety of contexts such as licensing (Jiang et al., 2009) or buyouts in equity partnerships (Folta and Miller, 2002). 5.4.5. Controlling for the dynamic effect A firm's diversification pattern may shape and evolve over time as a result of different investment decisions at multiple points in time. This suggests the possible influence of past values of the index on current diversification outcomes. In a dynamic context, diversification investment strategies, especially those closer to real options logic, are expected to be highly variable across time as growth options are continuously explored and exploited.31 Thus, we extend our core empirical model to a dynamic setting by incorporating the effect of lagged DIVPAT⁎: 



EXVALi;t ¼ α þ β1 DIVPAT i;t þ β2 DIVPAT i;t−1 þ β4 MHERF i;t þ β5 LTAi;t þ þ β6 LDTAi;t þ β7 EBITsalesi;t þ β8 CAPEXsalesi;t þ β9 LTA2i;t þ þ β10 dumINDUSTRY i;t þ β11 dumYEARi;t þ νi;t

ð14Þ

Table 13 reports the estimates of Eq. (14) with one lag of DIVPAT⁎ (denoted by DIVPAT⁎i,t-1). Our primary results remain similar. The evidence presented confirms that the diversification pattern is statistically significant vis-à-vis explaining excess value. Consistent 31 We estimated the volatility of each component of our index (INTER and INTRA) in terms of standard deviation over the sample years within each firm. We split the sample into two groups according to the values of our diversification pattern index in order to identify those firms with a diversification strategy closer to AiP diversification (below-median subsample) and those companies with a diversification strategy closer to GO diversification (above-median subsample). We find that volatility is higher in the above-median subsample, both in mean and median terms, confirming that GO diversification strategies tend to be more dynamic across time. Results are available upon request.

332

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

Table 11 Diversification pattern index and Excess Value [Eq. (12) and (13)]. Additional sensitivity tests controlling for diversification scope and growth opportunities. This Table shows the estimation results of Eqs. (12) and (13). In the Heckman estimates, the selection equation (Column 1 in Table 5) models a firm's propensity to diversify, and the Inverse Mills Ratio ((λi) is included as an additional regressor to correct potential self-selection bias in the sample. EXVAL is regressed on the diversification pattern index, degree of diversification and growth opportunities. DIVPAT⁎ denotes the diversification pattern index based on the Euclidean distance and with the extreme AiP diversification pattern of reference theoretically specified. MHERF represents the modified Herfindahl index and measures the degree of diversification. Growth opportunities are proxied by RDA (the ratio of annual R&D expenditures and beginning-of-year book assets). Firm size (LTA) and its squared term (LTA2), financial leverage (LDTA), profitability (EBITsales), level of investment (CAPEXsales), industry effect (dumINDUSTRY), and time effect (dumYEAR) are controlled in the estimations. The Wald and F tests contrast the null hypothesis of no joint significance of the explanatory variables. The Durbin-Wu-Hausman statistic tests for endogeneity of DIVPAT⁎. The instrumental variable (IV) partial F-statistic tests for weak instruments. Standard error is shown in parentheses under coefficients. ****, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively. Model EXVAL = f(DIVPAT⁎, DIV, GROWTH, control variables) Dependent variable: EXVAL Only controls

Constant DIVPAT⁎ Degree of diversification (DIV) MHERF

Full model with the baseline index and diversification

Full model with the baseline index, diversification and growth opportunities

OLS (1)

Heckman (2)

OLS (3)

Heckman (4)

2SLS (5)

OLS (6)

Heckman (7)

2SLS (8)

−2.8911*** (0.0656)

−3.0333*** (0.1360)

−2.6653*** (0.0647) 0.4434*** (0.0141)

−3.5874*** (0.1378) 0.8872*** (0.0561)

−0.0433 (0.4811) 5.5917*** (0.8860)

−2.8653*** (0.0849) 0.4339*** (0.0179)

−3.9714*** (0.1794) 0.9254*** (0.0718)

0.6560 (1.2816) 6.6511*** (2.2190)

−0.3741*** (0.0528)

−0.3457*** (0.0710)

−0.6140*** (0.0523)

0.2406*** (0.0788)

−3.3995*** (0.4978)

−0.6232*** (0.0650)

0.1264 (0.0995)

−4.3029*** (1.3271)

1.4889*** (0.1067)

1.4658*** (0.2708)

0.2071 (0.5566)

0.0097 (0.1216) −0.7686*** (0.1377) 1.3963*** (0.1405) −2.4223*** (0.6543) −0.0058 (0.0069)

0.6997*** (0.0234) −0.1984*** (0.0536) 0.7763*** (0.0363) 1.7829*** (0.1047) −0.0422*** (0.0018)

−0.2349 (0.3407) −0.6848*** (0.2355) 1.6721*** (0.3373) −3.5700* (1.9352) 0.0092 (0.0191)

YES YES 23,285

YES YES 15,302

0.7093*** (0.0433) −0.1824* (0.1044) 1.2334*** (0.0812) 2.3832*** (0.2390) −0.0414*** (0.0032) −0.1152*** (0.0401) YES YES 21,876 17,843 4033 2026.78***

Growth opportunities RDA Control variables LTA LDTA EBITsales CAPEXsales LTA2

0.7159*** (0.0187) −0.2079*** (0.0414) 0.6494*** (0.0298) 1.5703*** (0.0658) −0.0425*** (0.0015)

Inverse Mills Ratio (λi) Industry dummies Year dummies No. of obs. No. censored obs. No. uncensored obs. Wald Chi2 χ2 F-statistic Adjusted-R2 Durbin-Wu-Hausman test p-value Weak IV F-statistic

YES YES 23,285

79.03*** 0.1987

0.6444*** (0.0353) −0.0872 (0.0804) 1.0501*** (0.0685) 2.2099*** (0.1511) −0.03544*** (0.0026) 0.0118 (0.0310) YES YES 23,835 17,843 5992 2299.87***

0.6599*** (0.0184) −0.2523*** (0.0406) 0.7086*** (0.0292) 1.2537 *** (0.0652) −0.0396*** (0.0014)

YES YES 23,285

0.6092*** (0.0347) −0.1199 (0.0788) 1.0804*** (0.0672) 2.0143*** (0.1486) −0.0355*** (0.0025) −0.1048*** (0.0313) YES YES 23,835 17,843 5992 2642.00***

945.76***

94.51*** 0.2315

YES YES 15,302

526.36***

68.71*** 0.2492 0.000 19.7851***

0.000 4.4036**

with our hypothesis, we find that excess value is higher the greater the DIVPAT⁎. Results also support the dynamic nature of this relationship. When significant, the one-lagged index is positive and statistically significant at the 1% level. This suggests that an increase in this variable (and thereby, close to a GO strategy) will spark an increase in subsequent excess values. Interestingly, we find that the economic relevance decreases over time, thus suggesting that the effect of past diversification profiles attenuates over time. Nevertheless, the current value of the diversification pattern drives the main effect since it not only builds on the diversification history of the company but especially on the current configuration of the firm's growth options across its portfolio of businesses.

6. Conclusion This study investigates whether a firm's pattern of diversification accounts for part of the effect of this strategy on corporate value. We approximate each firm's diversification strategy relative to two extreme patterns (ranging from AiP diversification to GO diversification) and directly analyse the effect of this measure on diversification excess value. We perform our analysis on a sample of U.S. firms from 1998 to 2014, offering updated evidence on a post-1997 sample, after implementation of the new SFAS 131 reporting standard in the U.S. Our results confirm that the way firms diversify is by no means a trivial issue when determining diversification value outcomes.

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

333

Table 12 Industry pre-emption heterogeneity, Diversification pattern index, and Excess Value [Eq. (12)]. This Table shows the coefficients of the DIVPAT⁎ variable of the Heckman second stage estimations (Heckman's two-step estimator) and 2SLS estimations of Eq. (12). Regressions are performed separately for three different subsamples formed by terciles (quartiles) of PREEMPT levels. PREEMPT is the risk of pre-emption, calculated as the natural logarithm of the number of firms operating in the same 2-digit SIC code industry as the core business of the corresponding firm. Standard error is shown in parentheses under coefficients. ****, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively. INDUSTRY PRE-EMPTION HETEROGENEITY: Subsamples of firm-year observations by risk of pre-emption in the core industry Dependent variable: EXVAL DIVPAT⁎ estimation coefficient by PREEMPT terciles

1st tercile PREEMPT 2nd tercile PREEMPT 3rd tercile PREEMPT

Heckman

2SLS

0.5228*** (0.0264) 0.6125*** (0.0848) 0.9026*** (0.1053)

3.9055*** (0.8559) 9.7673** (4.1554) 4.4715*** (1.0684)

DIVPAT⁎ estimation coefficient by PREEMPT quartiles

1st quartile PREEMPT 2nd quartile PREEMPT 3rd quartile PREEMPT 4th quartile PREEMPT

Heckman

2SLS

0.5785*** (0.0314) 0.6776*** (0.0989) 0.7568*** (0.1091) 0.8101*** (0.1214)

3.1519*** (0.5582) -8.0537 (5.2593) 4.5602*** (1.0370) 4.9982*** (1.7497)

Some important contributions to the diversification literature emerge from this work. First, our results suggest that an additional dimension to the commonly studied dimensions of diversification (scope and relatedness) should be considered: namely, the diversification pattern that shapes the diversification-value linkage in a way that is consistent with real options logic. Our findings tie in with recent streams of research which advocate the endogenous nature of the diversification decision, thus making value creation or destruction contingent on firm-specific characteristics rather than on generic characteristics related to the strategy or to the firms undertaking it. We provide evidence that diversification strategy is neither inherently good nor bad. Rather, our findings suggest that when exploring the diversification puzzle, what seems important is not only how much to diversify (scope) and where (relatedness between businesses), but also ‘how’. Interestingly, our study sheds light on the need to explore further dimensions of diversification. In this study, we present the diversification pattern and show that it accounts for diversification excess value. Failing to consider the different dimensions of this strategy may have given rise to conflicting evidence in prior literature as a result of having mixed these different diversification dimensions, each of which has a different impact on a firm's value. Moreover, many papers call for the need to investigate further the validity of real options for strategic analysis in an effort to advance theory (Reuer and Tong, 2007). This study contributes to filling the gap in empirical works that apply the real options approach to strategy. We adopt this fresh approach in an effort to join a long-standing debate on the diversification-value relationship, which still awaits an answer, as evidenced by recent works (Elsas et al., 2010; Hoechle et al., 2012; Andreou et al., 2016; Kuppuswamy and Villalonga, 2016). We have shown that such a debate could benefit from linking real options based patterns of diversification and firm value. Finally, we develop and test a proxy for the diversification profiles based on how growth opportunities and assets-in-place are handled. To the best of our knowledge, this constitutes the first attempt to capture and measure these dimensions of corporate diversification. Our two-dimensional index follows the notion of profile deviation (Venkatraman, 1989), which has been applied in other areas of strategy, and joins prior research works that analyse the development of additional indices to better characterise each firm's diversification strategy such as Jacquemin and Berry's (1979) entropy index and Neffke and Henning's (2013) skill-relatedness index. This study has significant implications for business management since we show managers that the way in which a diversification strategy is implemented may prove to be very important vis-à-vis value creation. Accordingly, we advocate here for proactive managerial behaviour and stress the vital importance of combining expansion in a firm's core segments with simultaneously opening up fresh strategic options in new businesses. This evidence supports the basic premise by Williamson (2001) concerning the relevance of creating strategic options for the future. Our results reveal that the diversification discount is likely to be reduced when this strategy is designed to spread a firm's current capabilities beyond its core businesses by engaging in explorative participation in new industries. This investment logic emphasises the importance of getting ‘a foot in the door’ to access future investment opportunities while at the same time limiting downside risk by delaying the full commitment of resources. To end, we point to certain limitations in our research and to questions that remain open for future study. First, our sample only comprises U.S. firms. It might prove interesting to replicate the analysis on an international sample to check the consistency of our results. Second, our empirical analysis may inevitably suffer from the limitation of using segment data, which can sometimes be biased due to inconsistent segment reporting decisions or managerial self-interest reporting changes that do not correspond to an actual change in business operations (Villalonga, 2004a). In addition, we are only able to capture foothold investments to a certain extent since SFAS

334

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

Table 13 Dynamic models of Diversification pattern index and Excess Value controlling for diversification [Eq. (14)]. This Table shows the estimation results of Eq. (14). In the Heckman estimates, the selection equation (Column 1 in Table 5) models a firm's propensity to diversify, and the Inverse Mills Ratio ((λi) is included as an additional regressor to correct potential self-selection bias in the sample. EXVAL is regressed on the diversification pattern index (DIVPAT⁎), its lag, and the degree of diversification. DIVPAT⁎t-1 is the one-period lagged DIVPAT⁎. MHERF represents the modified Herfindahl index and measures the degree of diversification. Firm size (LTA) and its squared term (LTA2), financial leverage (LDTA), profitability (EBITsales), level of investment (CAPEXsales), industry effect (dumINDUSTRY), and time effect (dumYEAR) are included as control variables. The Wald and F tests contrast the null hypothesis of no joint significance of the explanatory variables. The Durbin-Wu-Hausman statistic tests for endogeneity of DIVPAT⁎ and DIVPAT⁎t-1 variables. The instrumental variable (IV) partial F-statistic tests for weak instruments. Standard error is shown in parentheses under coefficients. ****, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively. Model EXVAL = f(DIVPAT⁎, DIVPAT⁎ (−1), DIV, control variables) Dependent variable: EXVAL

Constant DIVPAT⁎ DIVPAT⁎t-1 Degree of diversification (DIV) MHERF Control variables LTA LDTA EBITsales CAPEXsales LTA2

OLS (1)

Heckman (2)

2SLS (3)

OLS (4)

Heckman (5)

2SLS (6)

−2.6099*** (0.0615) 0.3031*** (0.0183) 0.1363*** (0.0183)

−3.2968*** (0.1455) 0.8201*** (0.0645) 0.1714*** (0.0443)

−2.4797*** (0.1970) 2.9976 (3.1334) 1.5152 (2.7801)

−2.6275*** (0.0705) 0.3329*** (0.0183) 0.1700*** (0.0176)

−3.5547*** (0.1485) 0.7876*** (0.0633) 0.1712*** (0.0425)

−0.9526* (0.5544) 1.8254 (2.2460) 2.3716 (2.0334)

−0.6217*** (0.0391)

0.4344*** (0.0811)

−4.6255*** (0.9281)

−0.6035*** (0.0549)

0.3026*** (0.0832)

−2.5563*** (0.3561)

0.6510*** (0.0202) −0.2031*** (0.0423) 0.9018*** (0.0351) 0.3722*** (0.0651) −0.0408*** (0.0016)

0.5920*** (0.0393) −0.0870 (0.0865) 1.2981*** (0.0815) 0.7757*** (0.1554) −0.0372*** (0.0029) −0.1399*** (0.0307) NO NO 22,831 17,843 4988 1437.11***

0.1891* (0.0997) −0.1613* (0.0909) 1.4773*** (0.1213) −1.6976*** (0.4271) −0.0186*** (0.0061)

0.6334*** (0.0202) −0.2379*** (0.0440) 0.9493*** (0.0347) 1.1961*** (0.0762) −0.0382*** (0.0016)

0.1715* (0.0912) −0.5458*** (0.1006) 1.3814*** (0.1119) −1.2795** (0.5172) −0.0151*** (0.0055)

NO NO 19,000

YES YES 19,000

0.5741*** (0.0377) −0.0203 (0.0852) 1.2989*** (0.0777) 2.0197*** (0.1720) −0.0340*** (0.0027) −0.0945*** (0.0320) YES YES 22,831 17,843 4988 2386.51***

Inverse Mills Ratio (λi) Industry dummies Year dummies No. of obs. No. censored obs. No. uncensored obs. Wald Chi2 χ2 F-statistic Adjusted-R2 Durbin-Wu-Hausman test p-value Weak IV F-statistic

NO NO 19,000

952.78***

611.43*** 0.2045

YES YES 19,000

1512.87***

84.13*** 0.2471 0.000 DIVPAT⁎: 464.15*** DIVPAT⁎(−1): 363.64***

0.000 DIVPAT⁎: 104.33*** DIVPAT⁎(−1): 78.52***

131 requires firms to disclose segment data information for those segments accounting for at least 10% of a firm's total revenues, profit, or assets. Furthermore, it may prove interesting to examine in-depth a firm's motivations to engage in each diversification pattern, either AiP diversification or GO diversification. Prior studies, such as Folta and O'Brien (2004), suggest that certain factors connected to growth options value, such as industry uncertainty, may play a role. Further insights might serve to reveal the mechanisms that firms pursue when undertaking this strategy as well as their consistency with the value creation objective. Finally, our analysis suggests it may prove interesting to redefine and amplify our index to a dynamic context to control further for the impact of its evolution on value.

Acknowledgements Part of this research was conducted while Pilar Velasco was a Visiting PhD Student at the London Business School. Special thanks are due to her host, Professor Henri Servaes, for his kind hospitality and valuable discussions. The authors benefited from the helpful comments of Jeffry Netter (the editor), an anonymous referee, Luis Borge, Isabel Gómez, Philip Jaggs, Costas Markides, Krishna Paudyal, Pilar Pérez-Santana, Juan Antonio Rodríguez, Isabel Suárez; participants at the International Accounting & Finance Doctoral Symposium in Glasgow (2012), the ACEDE Conference in Cadiz (2012), Jornadas de Economía Industrial in IE Segovia (2013), the Finance Forum in IE Segovia (2013), the FMA European Conference in Maastricht (2014), and the IFABS Conference in Lisbon (2014); and the research seminars at the IE Business School in Madrid, Autonomous University of Barcelona, University of Salamanca, Public University of Navarre and CUNEF. Financial support was received from the Spanish Ministry of Education (FPU Programme), the Regional Government of Castilla y León (Ref. VA 260 U14), the Regional Government of Madrid and European Social Fund (Ref. EARLYFIN, S2015/HUM-3353), and the Spanish Ministry of Economy and Competitiveness (Ref. ECO2014-56102-P and ECO2016-77631-R).

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

335

Appendix A Table A.1 The dimensions of the index and Excess Value (Eq. (9) accounting for each dimension separately). This Table shows the estimation results of Eq. (9) considering the INTER and INTRA variables separately. In the Heckman estimates, the selection equation (Column 1 in Table 5) models a firm's propensity to diversify and the Inverse Mills Ratio ((λi) is included as an additional regressor to correct potential self-selection bias in the sample. EXVAL is regressed on INTER and INTRA separately. Firm size (LTA) and its squared term (LTA2), financial leverage (LDTA), profitability (EBITsales), level of investment (CAPEXsales), industry effect (dumINDUSTRY), and time effect (dumYEAR) are controlled in the estimations. The Wald and F tests contrast the null hypothesis of no joint significance of the explanatory variables. The Durbin-WuHausman statistic tests for endogeneity of INTER or INTRA variable. The instrumental variable (IV) partial F-statistic tests for weak instruments. Standard error is shown in parentheses under coefficients. ****, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively. Models: EXVAL = f(INTER, control variables) EXVAL = f(INTRA, control variables) Dependent variable: EXVAL INTER with industry and time dummies

Constant INTER

INTRA with industry and time dummies

OLS (1)

Heckman (2)

2SLS (3)

OLS (4)

Heckman (5)

2SLS (6)

−2.8072*** (0.0653) −0.0124 (0.0290)

−3.2122*** (0.1367) 0.1711*** (0.0402)

−1.1396 (0.8097) 4.8729** (2.3553)

−2.1015*** (0.0667)

−2.4819*** (0.1392)

5.6608*** (1.3566)

−0.4667*** (0.0144)

−0.6657*** (0.0441)

−5.6308*** (0.8961)

0.7180*** (0.0279) −0.3060*** (0.0770) 0.5270*** (0.0745) 1.5426*** (0.0997) −0.0442*** (0.0023)

0.6591*** (0.0184) −0.2559*** (0.0406) 0.7210*** (0.0293) 1.2497*** (0.0652) −0.0400*** (0.0014)

0.0208 (0.1202) −0.7560*** (0.1350) 1.4972*** (0.1539) −2.4133*** (0.6565) −0.0093 (0.0065)

YES YES 23,285

YES YES 23,285

0.6033*** (0.0348) −0.1151 (0.0791) 1.1220*** (0.0674) 2.0531*** (0.1490) −0.0351*** (0.0025) −0.0997*** (0.0314) YES YES 23,835 17,843 5992 2577.72***

INTRA Control variables LTA LDTA EBITsales CAPEXsales LTA2

0.7168*** (0.0188) −0.2104*** (0.0415) 0.6512*** (0.0298) 1.5808*** (0.0658) −0.0428*** (0.0015)

Inverse Mills Ratio (λi) Industry dummies Year dummies No. of obs. No. censored obs. No. uncensored obs. Wald Chi2 χ2 F-statistic Adjusted-R2 Durbin-Wu-Hausman test p-value Weak IV F-statistic

YES YES 23,285

0.6465*** (0.0353) −0.0770 (0.0804) 1.0483*** (0.0686) 2.2165*** (0.1511) −0.0358*** (0.0026) 0.0132 (0.0310) YES YES 23,835 17,843 5992 2292.16***

2617.50***

78.19*** 0.1970

YES YES 23,285

968.85***

95.88*** 0.2317 0.002 3.9010**

0.000 19.5765***

Table A.2 Diversification pattern index (based on calibration samples) and Excess Value (without extremes) [Eq. (10)]. This Table shows the estimation results of Eq. (10) after dropping extreme EXVAL observations (above 1.386 or below −1.386). In the Heckman estimates, the selection equation (Column 1 in Table 5) models a firm's propensity to diversify and the Inverse Mills Ratio ((λi) is included as an additional regressor to correct potential self-selection bias in the sample. EXVAL is regressed on the alternative specification of the diversification pattern index based on a calibration sample of Tobin's Q: DIVPATQ. Firm size (LTA) and its squared term (LTA2), financial leverage (LDTA), profitability (EBITsales), level of investment (CAPEXsales), industry effect (dumINDUSTRY), and time effect (dumYEAR) are controlled in the estimations. The Wald and F tests contrast the null hypothesis of no joint significance of the explanatory variables. The Durbin-Wu-Hausman statistic tests for endogeneity of DIVPATQ. The instrumental variable (IV) partial F-statistic tests for weak instruments. Standard error is shown in parentheses under coefficients. ****, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively. Model EXVAL = f(DIVPATQ, control variables) Dependent variable: EXVAL

Constant

Index based on calibration sample of Tobin's Q

Index based on calibration sample of Tobin's Q with industry and time dummies

OLS (1)

Heckman (2)

2SLS (3)

OLS (4)

Heckman (5)

2SLS (6)

−1.0596*** (0.0446)

−1.1031*** (0.1030)

−1.4593*** (0.0989)

−1.0384*** (0.0492)

−1.5449*** (0.1097)

−0.2820 (0.2933)

(continued on next page)

336

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

Table A.2 (continued) Model EXVAL = f(DIVPATQ, control variables) Dependent variable: EXVAL

DIVPATQ Control variables LTA LDTA EBITsales CAPEXsales LTA2

Index based on calibration sample of Tobin's Q

Index based on calibration sample of Tobin's Q with industry and time dummies

OLS (1)

Heckman (2)

2SLS (3)

OLS (4)

Heckman (5)

2SLS (6)

0.2091*** (0.0113)

0.3430*** (0.0342)

1.7412*** (0.2859)

0.2698*** (0.0121)

0.4344*** (0.0389)

3.9821*** (1.2962)

0.2515*** (0.0143) −0.1322*** (0.0278) 0.2658*** (0.0236) 0.1760*** (0.0384) −0.0147*** (0.0011)

0.1875*** (0.0289) −0.0145 (0.0596) 0.4643*** (0.0594) 0.3839*** (0.0993) −0.0093*** (0.0021) −0.0661*** (0.0216) NO NO 16,911 12,502 4409 488.21***

0.1616*** (0.0267) −0.1830*** (0.0416) 0.4371*** (0.0469) −0.1425* (0.0815) −0.0120*** (0.0016)

0.2374*** (0.0142) −0.1330*** (0.0289) 0.2919*** (0.0233) 0.6281*** (0.0468) −0.0124*** (0.0011)

−0.0089 (0.0935) −0.4085*** (0.1220) 0.6079*** (0.1258) −0.8621 (0.5341) −0.0012 (0.0048)

NO NO 16,438

YES YES 16,438

0.1936*** (0.0278) 0.0185 (0.0591) 0.4646*** (0.0571) 1.0379*** (0.1139) −0.0084*** (0.0020) −0.0280 (0.0228) YES YES 16,911 12,502 4409 1201.37***

Inverse Mills Ratio (λi) Industry dummies Year dummies No. of obs. No. censored obs. No. uncensored obs. Wald Chi2 χ2 F-statistic Adjusted-R2 Durbin-Wu-Hausman test p-value Weak IV F-statistic

NO NO 16,438

638.10***

269.88*** 0.0894

YES YES 16,438

359.72***

38.64*** 0.1449 0.000 27.2151***

0.000 4.7923***

Table A.3 Diversification pattern index DIVPATQ and Excess Value [Eq. (12) and (13)]. Additional sensitivity tests controlling for diversification scope and growth opportunities. This Table shows the estimation results of Eqs. (12) and (13). In the Heckman estimates, the selection equation (Column 1 in Table 5) models a firm's propensity to diversify and the Inverse Mills Ratio ((λi) is included as an additional regressor to correct potential self-selection bias in the sample. EXVAL is regressed on the diversification pattern index, degree of diversification and growth opportunities. DIVPATQ denotes the alternative specification of the diversification pattern index using a calibration sample based on Q. MHERF represents the modified Herfindahl index and measures the degree of diversification. Growth opportunities are proxied by RDA (the ratio of annual R&D expenditures and beginning-of-year book assets). Firm size (LTA) and its squared term (LTA2), financial leverage (LDTA), profitability (EBITsales), level of investment (CAPEXsales), industry effect (dumINDUSTRY), and time effect (dumYEAR) are controlled in the estimations. The Wald and F tests contrast the null hypothesis of no joint significance of the explanatory variables. The Durbin-Wu-Hausman statistic tests for endogeneity of DIVPATQ. The instrumental variable (IV) partial F-statistic tests for weak instruments. Standard error is shown in parentheses under coefficients. ****, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively. Model EXVAL = f(DIVPAT, DIV, GROWTH, control variables) Dependent variable: EXVAL Full model with the baseline index and diversification

Constant DIVPAT

Q

Degree of diversification (DIV) MHERF

Full model with the baseline index, diversification and growth opportunities

OLS (1)

Heckman (2)

2SLS (3)

OLS (4)

Heckman (5)

2SLS (6)

−2.6183*** (0.0651) 0.3942*** (0.0167)

−3.2026*** (0.1424) 0.7609*** (0.0614)

−1.021* (0.5470) 8.0132*** (2.3757)

−2.7368*** (0.0867) 0.3907*** (0.0212)

−3.5067*** (0.1852) 0.8650*** (0.0791)

−1.2152 (1.0230) 7.1855 (4.4121)

−0.3555*** (0.0516)

0.3939*** (0.0851)

−2.3129*** (0.6359)

−0.3413*** (0.0646)

0.4522*** (0.1076)

−2.4375* (1.3753)

0.9809*** (0.1121)

0.8197*** (0.2828)

0.1403 (0.6448)

0.6365*** (0.0236) 0.1548*** (0.0534)

0.5654*** (0.0434) 0.2444** (0.1026)

−0.0600 (0.4580) −0.2148 (0.2904)

Growth opportunities RDA Control variables LTA LDTA

0.6113*** (0.0184) 0.1098*** (0.0401)

0.5105*** (0.0344) 0.2718*** (0.0768)

−0.0459 (0.2146) −0.4154* (0.2148)

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

337

Table A.3 (continued) Model EXVAL = f(DIVPAT, DIV, GROWTH, control variables) Dependent variable: EXVAL Full model with the baseline index and diversification

EBITsales CAPEXsales LTA2

OLS (1)

Heckman (2)

2SLS (3)

OLS (4)

Heckman (5)

2SLS (6)

0.5360*** (0.0306) 1.1715*** (0.0630) −0.0359*** (0.0014)

0.7929*** (0.0684) 1.8199*** (0.1417) −0.0289*** (0.0025) −0.1008*** (0.0305) YES YES 19,362 14,391 4971 1969.44***

1.0390*** (0.1894) −2.1712** (1.0647) −0.0014 (0.0118)

0.5716*** (0.0386) 1.6514*** (0.1049) −0.0375*** (0.0018)

1.0948*** (0.3597) −2.0036 (2.3947) 0.0014 (0.0259)

YES YES 18,889

YES YES 12,217

0.9068*** (0.0848) 2.2078*** (0.2311) −0.0319*** (0.0031) −0.1152*** (0.0390) YES YES 17,695 14,391 3304 1466.77***

Inverse Mills Ratio (λi) Industry dummies Year dummies No. of obs. No. censored obs. No. uncensored obs. Wald Chi2 χ2 F-statistic Adjusted-R2 Durbin-Wu-Hausman test p-value Weak IV F-statistic

Full model with the baseline index, diversification and growth opportunities

YES YES 18,889

461.09***

79.79*** 0.2383

YES YES 12,217

405.13***

54.86*** 0.2485 0.000 5.5859***

0.000 1.3183

Table A.4 Diversification pattern index and Excess Value [Eq. (12)]. Additional sensitivity tests controlling for diversification scope (dumDIV). This Table shows the estimation results of Eq. (12) by using dumDIV to proxy for a firm's diversification status. EXVAL is regressed on the diversification pattern index, diversification status and growth opportunities. The diversification pattern is measured by two alternative specifications of our index: DIVPAT⁎ based on the Euclidean distance and with the extreme AiPD pattern of reference theoretically specified; and DIVPATQ based on a calibration sample using Q. dumDIV indicates a firm's diversification status: it is a dummy variable equalling 1 if the firm operates in two or more different 4-digit SIC industries, and zero otherwise. Firm size (LTA) and its squared term (LTA2), financial leverage (LDTA), profitability (EBITsales), level of investment (CAPEXsales), industry effect (dumINDUSTRY), and time effect (dumYEAR) are controlled in the estimations. The Wald and F tests contrast the null hypothesis of no joint significance of the explanatory variables. The Durbin-Wu-Hausman statistic tests for endogeneity of DIVPAT. The instrumental variable (IV) partial F-statistic tests for weak instruments. Standard error is shown in parentheses under coefficients. ****, **, and * denote statistical significance at the 1%, 5%, and 10% level, respectively. N.B.: It was not possible to report the Heckman estimates in this case. STATA drops the variable dumDIV from the second-stage estimations. Model EXVAL = f(DIVPAT, DIV, control variables) Dependent variable: EXVAL

Constant DIVPAT⁎

Full model with the baseline index

Full model with index based on calibration sample of Q

OLS (1)

2SLS (2)

OLS (3)

2SLS (4)

−2.7146*** (0.0645) 0.5019*** (0.0147)

−1.1741*** (0.2942) 5.3675*** (0.7939)

−2.6779*** (0.0650)

−2.4692*** (0.2281)

0.4470*** (0.0174)

8.1528*** (2.3401)

DIVPATQ Degree of diversification dumDIV Control variables LTA LDTA EBITsales CAPEXsales LTA2

−0.3923*** (0.0237)

−3.0353*** (0.4345)

−0.2693*** (0.0231)

−3.4119*** (0.9572)

0.6554*** (0.0184) −0.2469*** (0.0405) 0.7240*** (0.0292) 1.2155*** (0.0651) −0.0395*** (0.0014)

0.0564 (0.1070) −0.6456*** (0.1162) 1.4182*** (0.1327) −2.2820*** (0.5908) −0.0080 (0.0062)

0.6094*** (0.0183) 0.1149*** (0.0400) 0.5448*** (0.0305) 1.1470*** (0.0629) −0.0359*** (0.0014)

−0.0173 (0.2001) −0.3034 (0.1852) 1.1052*** (0.1988) −2.3246** (1.0751) −0.0021 (0.0113)

Inverse Mills Ratio (λi) (continued on next page)

338

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

Table A.4 (continued) Model EXVAL = f(DIVPAT, DIV, control variables) Dependent variable: EXVAL

Industry dummies Year dummies No. of obs. No. censored obs. No. uncensored obs. Wald Chi2 χ2 F-statistic Adjusted-R2 Durbin-Wu-Hausman test p-value Weak IV F-statistic

Full model with the baseline index

Full model with index based on calibration sample of Q

OLS (1)

2SLS (2)

OLS (3)

2SLS (4)

YES YES 23,285

YES YES 23,285

YES YES 18,889

YES YES 18,889

1125.52*** 96.94*** 0.2361

490.88*** 81.34*** 0.2419

0.000 22.7677***

0.000 5.9215***

References Ahuja, G., Novelli, E., 2016. Redirecting research efforts on the diversification-value linkage: the search for synergy. Acad. Manag. Ann. (forthcoming). Amihud, Y., Lev, B., 1981. Risk reduction as a managerial motive for conglomerate mergers. Bell J. Econ. 12 (2), 605–617. Andreou, P., Louca, C., Petrou, A., 2016. Organizational learning and corporate diversification performance. J. Bus. Res. 69 (9), 3270–3284. Andrés, P., Fuente, G., Velasco, P., 2014. Growth opportunities and the effect of corporate diversification on value. Span. Rev. Finan. Econ. 12 (2), 72–81. Andrés, P., Fuente, G., Velasco, P., 2016. Are real options a missing piece in the diversification-value puzzle? Int. Rev. Financ. Anal. 48, 261–271. Bai, G., Elyasiani, E., 2013. Bank stability and managerial compensation. J. Bank. Financ. 37 (3), 799–813. Berger, P.G., Ofek, E., 1995. Diversification's effect on firm value. J. Financ. Econ. 37 (1), 39–65. Bernardo, A., Chowdhry, B., 2002. Resources, real options, and corporate strategy. J. Financ. Econ. 63 (1), 211–234. Borghesi, R., Houston, J., Naranjo, A., 2007. Value, survival and the evolution of firm organizational structure. Financ. Manag. 36 (3), 5–31. Bowman, E.H., Hurry, D., Miller, A., 1992. Calls on high technology: Japanese exploration of venture capital investments in the United States. Strateg. Manag. J. 13 (2), 85–101. Campa, J.M., Kedia, S., 2002. Explaining the diversification discount. J. Financ. 57 (4), 1731–1762. Cao, C., Simin, T., Zhao, J., 2008. Can growth options explain the trend in idiosyncratic risk? Rev. Financ. Stud. 21 (6), 2599–2633. Çolak, G., 2010. Diversification, refocusing and firm value. Eur. Finan. Manag. 16 (3), 422–448. Da, Z., Guo, R., Jagannathan, R., 2012. CAPM for estimating the cost of equity capital: interpreting the empirical evidence. J. Financ. Econ. 103 (1), 204–220. Dastidar, P., 2009. International corporate diversification and performance: does firm self-selection matter? J. Int. Bus. Stud. 40, 71–85. Elsas, R., Hackethal, A., Holzhäuser, M., 2010. The anatomy of bank diversification. J. Bank. Financ. 34 (6), 1274–1287. Erdorf, S., Hartmann-Wendels, T., Heinrichs, N., Matz, M., 2013. Corporate diversification and firm value: a survey of recent evidence. Fin. Mkts. Portfolio Mgmt. 27 (2), 187–215. Estélyi, K., Nisar, T., 2016. Diverse boards: why do firms get foreign nationals on their boards? J. Corp. Finan. 39, 174–192. Fauver, L., Houston, J., Naranjo, A., 2003. Capital market development, international integration, legal systems, and the value of corporate diversification: a cross-country analysis. J. Financ. Quant. Anal. 38 (1), 135–157. Ferris, S.P., Sen, N., Lim, C.Y., Yeo, G., 2002. Corporate focus versus diversification: the role of growth opportunities and cash flow. J. Int. Financ. Mark. Inst. Money 12 (3), 231–252. Folta, T.B., Miller, K.D., 2002. Real options in equity partnerships. Strateg. Manag. J. 23 (1), 77–88. Folta, T.B., O'Brien, J., 2004. Entry in the presence of dueling options. Strateg. Manag. J. 25 (2), 121–138. Greene, W., 2003. In: Hall, Prentice (Ed.), Econometric Analysis. Fifth edition, United States. Grullon, G., Lyandres, E., Zhdanov, A., 2012. Real options, volatility, and stock returns. J. Financ. 67 (4), 1499–1537. He, X., 2012. Two sides of a coin: endogenous and exogenous effects of corporate diversification on firm value. Int. Rev. Finan. 12 (4), 375–397. Heckman, J., 1979. Sample selection bias as a specification error. Econometrica 47 (1), 153–161. Hirschman, A.O., 1964. The paternity of an index. Am. Econ. Rev. 54 (5), 761–770. Hoechle, D., Schmid, M., Walter, I., Yermack, D., 2012. How much of the diversification discount can be explained by poor corporate governance? J. Financ. Econ. 103 (1), 41–60. Holder, M., Zhao, A., 2015. Value exploration and materialization in diversification strategies. Rev. Quant. Finan. Acc. 45 (1), 175–213. Jacquemin, A., Berry, C., 1979. Entropy measure of diversification and corporate growth. J. Ind. Econ. 27 (4), 359–369. Jiang, M.S., Aulakh, P.S., Pan, Y., 2009. Licensing duration in foreign markets: a real options perspective. J. Int. Bus. Stud. 40 (4), 559–577. Klingebiel, R., Adner, R., 2015. Real options logic revisited: the performance effects of alternative resource allocation regimes. Acad. Manag. J. 58 (1), 221–241. Kogut, B., Kulatilaka, N., 1994. Options thinking and platform investments: investing in opportunity. Calif. Manag. Rev. 36 (2), 52–71. Kuppuswamy, V., Villalonga, B., 2016. Does diversification create value in the presence of external financing constraints? Evidence from the 2007–2009 financial crisis. Manag. Sci. 62 (4), 905–923. Laamanen, T., 1999. Option nature of company acquisition motivated by competence acquisition. Small Bus. Econ. 12 (2), 149–168. Martin, J.D., Sayrak, A., 2003. Corporate diversification and shareholder value: a survey of recent literature. J. Corp. Finan. 9 (1), 37–57. Miller, D.J., 2006. Technological diversity, related diversification, and firm performance. Strateg. Manag. J. 27 (7), 601–619. Mitton, T., Vorkink, K., 2010. Why do firms with diversification discounts have higher expected returns? J. Financ. Quant. Anal. 45 (6), 1367–1390. Neffke, F., Henning, M., 2013. Skill relatedness and firm diversification. Strateg. Manag. J. 34 (3), 297–316. Palich, L.E., Cardinal, L.B., Miller, C., 2000. Curvilinearity in the diversification-performance linkage: an examination of over three decades of research. Strateg. Manag. J. 21 (2), 155–174. Rajan, R., Servaes, H., Zingales, L., 2000. The cost of diversity: the diversification discount and inefficient investment. J. Financ. 55 (1), 35–80. Raynor, M.E., 2002. Diversification as real options and the implications of firm-specific risk and performance. Eng. Econ. 47 (4), 371–389. Reuer, J.J., Tong, T.W., 2007. How do real options matter? Empirical research on strategic investments and firm performance. In: Reuer, J.J., Tong, T.W. (Eds.), Real Options Theory. Advances in Strategic Management, first ed. Volume 24. Elsevier.

P. de Andrés et al. / Journal of Corporate Finance 43 (2017) 316–339

339

Rudolph, C., Schwetzler, B., 2013. Conglomerates on the rise again? A cross-regional study on the impact of the 2008–2009 financial crisis on the diversification discount. J. Corp. Finan. 22, 153–165. Santaló, J., Becerra, M., 2008. Competition from specialized firms and the diversification-performance linkage. J. Financ. 63 (2), 851–883. Servaes, H., 1996. The value of diversification during the conglomerate merger wave. J. Financ. 51 (4), 1201–1225. Stowe, J.D., Xing, X., 2006. Can growth opportunities explain the diversification discount? J. Corp. Finan. 12 (4), 783–796. Teplensky, J.D., Kimberly, J.R., Hillman, A.L., Schwartz, J.S., 1993. Scope, timing and strategic adjustment in emerging markets: manufacturer strategies and the case of MRI. Strateg. Manag. J. 14 (7), 505–527. Trigeorgis, L., Lambertides, N., 2014. The role of growth options in explaining stock returns. J. Financ. Quant. Anal. 49 (3), 749–771. Venkatraman, N., 1989. The concept of fit in strategy research: toward verbal and statistical correspondence. Acad. Manag. Rev. 14 (3), 423–444. Venkatraman, N., Prescott, J.E., 1990. Environment-strategy coalignment: an empirical test of its performance implications. Strateg. Manag. J. 11 (1), 1–23. Villalonga, B., 2004a. Diversification discount or premium? New evidence from BITS establishment-level data. J. Financ. 59 (2), 479–506. Villalonga, B., 2004b. Does diversification cause the “diversification discount”? Financ. Manag. 33 (2), 5–27. Williamson, P., 2001. Strategy as options on the future. In: Cusumano, M., Markides, C. (Eds.), Strategic Thinking for the Next Economy, MIT Sloan Management Review, Ed Vol. 7. Jossey-Bass A Wiley Company, Cambridge, MA, pp. 157–178 Chapter.