Separation of Ownership and Management - CiteSeerX

6 downloads 23133 Views 202KB Size Report
Email: [email protected] .... 2 BCAR is calculated by the A. M. Best Company. ..... Capital Adequacy Ratio and marketing system (i.e., independent agency ...
Separation of Ownership and Management: Implications for Risk-Taking Behavior

Cassandra R. Cole, PhD Associate Professor and Waters Fellow in Risk Management and Insurance College of Business Florida State University Tallahassee, FL 32306 Phone: 850-644-9283 Email: [email protected] Enya He, PhD (contact author) Assistant Professor College of Business University of North Texas Denton, TX 76203-5339 Phone: 940-565-3060 Email: [email protected] Kathleen A. McCullough, PhD Associate Professor and State Farm Professor of Risk Management/Insurance College of Business Florida State University Tallahassee, FL 32306 Phone: 850-644-8358 Email: [email protected] David W. Sommer, PhD Charles E. Cheever Chair of Risk Management Bill Greehey School of Business St. Mary's University San Antonio, TX 78228 Phone: (210) 431-8055 Email: [email protected]

2009

Separation of Ownership and Management: Implications for Risk-Taking Behavior

Abstract Issues associated with the relation between the separation of ownership and management and risk-taking behavior have been considered in an array of studies, with varying results. Due to the wide variety of ownership structures present, the property-casualty insurance industry provides an excellent setting to test the conflicting hypotheses related to separation of ownership from management and risk-taking behavior. Employing a large sample of property-liability companies over the period of 1996 to 2004, we empirically test the alternative hypotheses regarding the implications of separation of ownership from management for firms’ risk-taking behavior. The empirical tests include the ownership structures specified in prior research as well as a more detailed classification scheme. We find that each ownership structure is significantly different from every other ownership structure in terms of risk.

1. Introduction and Motivation In general, the insurance industry in the U.S. is dominated by two ownership structures: mutual companies and stock companies. Stock companies employ the standard corporate form, while mutuals merge the customer and owner functions. However, this simple distinction masks a more complex reality.

Within the category of stock insurers, tremendous variation in

ownership structure exists. At one extreme, stock ownership can be fully concentrated in the hands of management; and at the other extreme, the stock ownership can be widely dispersed when the stock company is ultimately owned by a mutual parent. These ownership structures manifest various degrees of separation of ownership from management.

Specifically, as

explained later in the paper, the degree of separation of ownership from management increases as one moves across the ownership structures, from companies closely-held by management, to companies closely-held by others (i.e., parties other than management), to widely-held stock insurers, to mutual-owned stock insurers, and finally to mutual insurers. Prior literature has shown that mutual and stock ownership structures are significantly different from each other, and that such differences have important implications for insurers’ distribution choices, CEO compensation, and other aspects of insurers’ operations [e.g., Mayers and Smith (1981), Mayers and Smith (1988), Mayers and Smith (1992), Kim, Mayers and Smith (1996), Pottier and Sommer (1997), and Regan and Tzeng (1999)]. The difference between mutual and stock ownership structures also has important implications for insurers’ risk-taking behavior, including decisions related to business mix and investment strategies [Lamm-Tennant and Starks (1993) and Lee, Mayers and Smith (1997)]. The results of existing studies are consistent with greater separation of ownership and management being associated with less risk taking.

1

More than one theory can be found in existing literature relating to the implications of separation of ownership and management for firms’ risk-taking behavior, especially when taking into account the entire spectrum of ownership structures. For example, Cummins and Sommer (1996) argue that risk taking is negatively associated with the degree of separation of ownership and management, implying that risk taking would be lower in widely-held companies than in closely-held companies.

However, Fama and Jensen (1983) predict that firms’ risk taking is

positively associated with the degree of separation of ownership and management, implying that risk taking is expected to be higher in widely-held companies than in closely-held companies. Understanding the relation between ownership and management separation and risktaking behavior is a salient issue for a variety of stakeholders. Not only does it impact both the owners and managers of the firm, but it is also important to policyholders and regulators. The risk-taking behavior of insurance companies affects the price and availability of coverage as well as the solvency of the insurers. In addition, a better understanding of this dynamic across a more complete spectrum of ownership structures is important for regulators tasked with ensuring insurer solvency. In the current study, we empirically test the alternative theories regarding the relation between separation of ownership and management and risk-taking by examining the implications of ownership structure for firm’s risk-taking behavior in the U.S. property-liability insurance industry, while controlling for other factors known to impact firm risk. Many prior studies in the insurance literature either contrast only mutual and stock companies [e.g., Lamm-Tennant and Starks (1993) and Lee et al. (1997)], or limit their samples to stock insurance companies only

2

[Cummins and Sommer (1996)]. 1

We replicate the classification schemes utilized in these

studies to allow for both a comparison to prior work as well as the ability to determine if the classification structure utilized impacts the results obtained. Further, we also examine a more detailed classification of ownership structures than previous studies based on a sample of all individual mutual and stock insurance companies for which data are available. Specifically, we consider an ownership classification scheme that includes mutual companies, mutual-owned stock companies, widely-held stock companies, stock companies closely-held by others, and stock companies closely-held by management. In addition to using a more refined ownership classification scheme, we add to the insurer risk-taking literature by utilizing a multi-dimensional summary risk measure, namely the Best’s Capital Adequacy Ratio (BCAR) 2 Most previous studies examining the risk-taking behavior of ..

insurance companies rely on a single dimensional measure of firm risk, such as operational risk or investment risk [e.g., Lamm-Tennant and Starks (1993) and Lee et al. (1997)].3 In doing so, these studies are able to capture only certain aspects of firm risk. In contrast, the risk measure employed in this paper, BCAR, is a much more comprehensive measure, as it combines various aspects of insurer risk into one single ratio. Specifically, BCAR incorporates five categories of firm risks (namely, asset, credit, underwriting, off-balance sheet, and interest rate risks) as well as qualitative measures (such as financial flexibility, reinsurance quality, and catastrophe loss exposure).

In addition, it reflects the insurer’s level of capitalization.

1

As such, this

Cummins and Sommer (1996) consider three ownership structures among stock companies (i.e., widely-held, closely-held by management and closely-held by others). They do not use a single summary risk measure, but instead model capital and portfolio risk separately in a simultaneous equations framework. 2 BCAR is calculated by the A. M. Best Company. Further discussion of this measure is provided later in this paper. 3 Downs and Sommer (1999) consider capital market measures of overall risk, but their analysis is necessarily limited to publicly traded firms.

3

comprehensive measure has great advantages over most other single risk measures employed in previous studies. The remainder of the paper is organized as follows: Section 2 reviews the relevant literature and develops the hypotheses. Section 3 describes the data and methodology. Section 4 summarizes the dependent and independent variables. Section 5 presents the results of the empirical analysis and Section 6 offers concluding remarks. 2. Literature Review and Hypotheses Development 2.1. Mutual vs. Stock Ownership Various ownership structures exist in the insurance industry.

In the absence of

contracting/transactions costs, ownership structure does not matter [Coase (1960)]. However, when contracting/transactions costs are not zero, “conflicts of interest arise whenever discretionary behavior is authorized with costs related to the amount of discretion and the effectiveness of the mechanism chosen to control the conflict” [Mayers and Smith (1981), p.425]. Such conflicts of interest are commonly known as agency problems. As one control mechanism of agency problems, ownership structure has important implications for firms’ operations. Mutual and stock ownership are the two most prominent ownership structures present in the insurance industry. As first outlined in Mayers and Smith (1981) and later empirically tested in papers such as Mayers and Smith (1988), Mayers and Smith (1992), Lamm-Tennant and Starks (1993), Kim et al. (1996), and Pottier and Sommer (1997), there are significant differences between the operations of stock insurers and mutual insurers. For example, Mayers and Smith (1992) find that stock firms are associated with higher CEO compensation than mutual firms. In addition, Kim et al. (1996) document a strong association between firms’

4

ownership structures and their distribution systems. Overall, these studies have shown that ownership structures have important implications for a wide variety of firm characteristics. Not only do differences between the mutual and stock ownership forms have important implications for insurers’ operations, but also they have important implications for insurers’ risktaking behavior. Two studies directly examining the difference in risk-taking between mutual and stock property-liability insurers are Lamm-Tennant and Starks (1993) and Lee et al. (1997). Measuring risk as the variance of the loss ratios, Lamm-Tennant and Starks (1993) find that stock insurers take greater risk than mutual insurers. In particular, their evidence suggests that “stock insurers write relatively more business than do mutual insurers in lines and states having higher risk” (p. 29). Using an event-study approach, Lee et al. (1997) examine the risk-taking behavior of companies by investigating insurers’ asset portfolio changes surrounding the enactment of guaranty funds. Using insurers’ investment in stocks, bonds, and other assets as measures of risk-taking behavior, they show a shift in investment allocation by stock insurers from less risky assets (i.e., bonds) to more risky assets (i.e., common stocks) after the establishment of the guaranty funds, but no such shift for mutual insurers. 2.2. Stock Ownership Classes While most of the literature discussing insurer ownership structure focuses only on the stock form versus the mutual form, some research has recognized the variety of ownership structures that exist within the stock category. First, a stock company can be classified as mutual-owned if it is ultimately owned by a mutual company. Alternatively, a stock company can be classified as stock-owned if its ultimate parent is a stock company. Prior studies have shown that mutual-owned stock insurers are quite different from stock-owned insurers. For example, Mayers and Smith (1994) find that the production allocation decisions of mutual-

5

owned stock insurers are more similar to those of mutual than stock insurers. Second, ownership also varies among stock-owned stock insurers.

Depending on the degree of ownership

concentration, stock-owned stock insurers can be further classified as widely-held stock insurers and closely-held stock insurers. For example, Mayers and Smith (1990) find strong evidence that widely-held stocks reinsure less than closely-held stocks. Third and last, among closelyheld stocks, further distinction can be drawn between stocks closely-held by management and stocks closely-held by others. Further distinction among stock ownership classes is important, as prior studies have shown that difference exists among these ownership classes with respect to insurers’ risk-taking behavior [Lee et al. (1997), Cummins and Sommer (1996), and Downs and Sommer (1999)]. For example, in addition to their study of mutual and stock insurers, Lee et al. (1997) also examine the asset portfolio change of mutual-owned stock insurers around the enactment of guaranty funds. The authors find that the mutual-owned stock companies in their sample exhibit risk-taking behaviors similar to those of other stock companies. This is contrary to the findings in Mayers and Smith (1994), which find that mutual-owned stock companies behave more like mutual companies. Given the conflicting evidence regarding this unique group, how mutualowned stock insurers behave remains an empirical question. Cummins and Sommer (1996) utilize the risky-debt model of the firm derived from the option-pricing theory in their examination of insurer risk-taking behavior. They find evidence supporting the hypothesis that capital and risk are positively related. In addition, Cummins and Sommer (1996) find that closely-held firms have lower capital and higher risk than publiclytraded firms, consistent with their notion that “owner-manager incentives are more closely aligned in these [closely-held firms] than in widely-held firms” (p. 1089).

6

Downs and Sommer (1999) extend the work of Lee et al. (1997) by addressing several additional issues. First, the authors employ a series of market-based risk measures, instead of the single dimensional asset risk measure used by Lee et al. (1997). 4 They also investigate the impact of insider ownership on insurers’ risk-taking behavior. Consistent with the risk-subsidy hypothesis of Lee et al. (1997), Downs and Sommer (1999) document a positive relation between risk-taking behavior and insider ownership for publicly traded property-liability companies. Before discussing the hypotheses examined in this study, we first summarize the relation between the various ownership categories and the level of separation of ownership and management. Mutuals are posited to have the greatest separation of ownership and management. Managers of mutuals cannot be monitored by institutional owners or other large blockholders, cannot be provided with stock-based incentive compensation, and do not have to contend with the threat of hostile takeover [Mayers, Shivdasani and Smith (1997)]. Thus, many of the common mechanisms used to assure that managers act in the best interest of owners are not present in mutuals. Next in the ordering of ownership/management separation are mutual-owned stocks. By definition, a mutual-owned company is a stock company whose ownership ultimately rests with the policyholders of the mutual parent. Mayers and Smith (1994) suggest that the problems of controlling the managers of a mutual-owned stock are similar to those of mutual companies, and thus owner-manager conflict in mutual-owned stocks is expected to be greater than that in stockowned stocks. From the perspective of owner/management separation, mutual-owned stock companies represent the most diffuse ownership structure among all stock ownership classes, as each policyholder of a typical mutual company has only one voting right.

4

Downs and Sommer (1999) employ nine different capital market measures of risk, including measures of total return risk, systematic risk, and nonsystematic risk.

7

Among the stock-owned stock firms, the greatest separation of ownership and management occurs in widely-held stocks, since this category exhibits the most diffusion of ownership.

Next are stocks closely-held by others.

For these firms, ownership is highly

concentrated, so owners have both the incentive and ability to closely monitor the management. Finally, the least separation of ownership and management exists in stock firms that are closelyheld by management, since for these firms, the owners and the managers are one and the same. 2.3. Hypotheses Development Based on the findings of previous studies, it is clear that the overall level of insurer risk varies with the organizational form of the insurers. In particular, the degree of the separation of ownership and management is shown to have important implications for insurers’ risk-taking behavior. However, as mentioned earlier, alternative theories exists regarding the implication of ownership and management separation for firms’ risk-taking behavior. The first theory, proposed by Cummins and Sommer (1996), argues that a higher degree of separation of ownership and management may be associated with lower firm risk. We refer to this as the incentive misalignment hypothesis. The rationale is that non-owner managers may be risk averse and thus hesitant to pursue risky strategies due to the fact that their human capital is invested in the firm, despite the fact that higher risk may be desirable to firms’ owners as it increases the market value of owners’ equity. Employing a sample of stock property-liability insurers, Cummins and Sommer (1996) find that publicly-traded firms have lower risk than closely-held firms, consistent with the incentive misalignment hypothesis. A similar argument is raised by Sanders, Strock and Travlos (1990) who find a positive relation between the percentage of a firm’s outstanding stock held by insiders and the level of risk of their sample banks. Also consistent with this theory, Downs and Sommer (1999) find a positive relation between risk-

8

taking by property-liability insurers and insider ownership, with higher insider ownership representing less separation of ownership and management. A second theory regarding ownership structure and risk-taking, which we refer to as the sub-optimal diversification hypothesis, is implied by Fama and Jensen (1983). Specifically, Fama and Jensen (1983) argue that unlike widely-held firms that are owned by many individuals whose wealth is likely to be diversified, closely-held companies are controlled by one or several individuals whose wealth is typically more concentrated in the closely-held venture. Such ownership concentration results in foregone diversification and limited alienability in ownership. This raises the cost of risk-bearing services and leads to less investment in risky projects by closely held corporations compared to firms with widely dispersed ownership [Fama and Jensen (1983), pp. 332-333]. This same argument would apply to closely-held firms compared to mutuals and mutual owned stocks. Drawing from this theory, we hypothesize that some of the relationships described above regarding ownership structure and risk-taking across the five ownership categories may be altered due to the concentration of wealth and thus sub-optimal diversification in closely-held firms.

In regards to stocks closely-held by management, the existence of the sub-optimal

diversification factor is expected to biases risk taking downward compared to the other ownership forms. This is because, under the sub-optimal hypothesis, the owners of these firms have a strong incentive to engage in less risky behavior, as both their financial capital and their human capital are concentrated in the firms they manage. Such downward influence would also exist for firms closely-held by others, but to a lesser extent, since the financial and human capital of their owners are not both concentrated in the same firm, as is the case for owners of firms closely-held by management. Finally, while likely much weaker, a downward influence could

9

also occur for widely-held stocks, since some widely-held stocks might exhibit some managerial wealth concentration due to extensive insider ownership. While the sub-optimal diversification hypothesis may not impact the relation between the separation of ownership and management and risk taking when comparing stocks and mutuals, it may change the ordering of risk taking among the more detailed ownership structures. Specifically, firms closely-held by management, which according to the incentive misalignment hypothesis should engage in the highest levels of risk-taking, may exhibit less risky behavior due to sub-optimal diversification. Likewise, firms closely-held by others might also engage in less risky behavior compared to that implied by the incentive misalignment hypothesis, though not as great a reduction as that for firms closely-held by management. Finally, it is possible that suboptimal diversification may lead to lower risk-taking among widely-held stocks compared to what would be implied by just the incentive misalignment hypothesis. 3. Data and Methodology The data used for this study are an extension of the dataset compiled by He and Sommer (2006a). In their study of board composition, they collected detailed ownership information (i.e., mutual, stock, and among stock companies, mutual-owned, widely-held, closely-held by others, and closely-held by management) for all property-liability insurance companies whose management information is available from A. M. Best Company’s Best Insurance Reports and whose ultimate ownership is determinable between 1996 and 2004. We obtain the firm-specific financial information from the NAIC Database. In addition, information including firms’ Best’s Capital Adequacy Ratio and marketing system (i.e., independent agency system or exclusive agency system) are collected from A. M. Best Company’s Best’s Key Rating Guide – P/C Edition.

10

Given the panel nature of our data, we conduct multiple tests to determine the best empirical approach. F-tests for the appropriateness of pooled-OLS models, as described in Baum (2006), indicate that pooled-OLS models are not appropriate for our data.

Further,

Hausman tests reject the hypotheses that random effects are more appropriate than fixed effects models for our data.

While testing indicates that the fixed-effects approach is the most

appropriate, one difficulty in properly estimating this model is that ownership structure and distribution systems rarely change over the sample period. Thus, the traditional fixed effects approach is not an appropriate methodology as the procedure can not precisely estimate time invariant or nearly time invariant variables [Wooldridge (2002), p. 286]. Instead, we employ the fixed effects vector decomposition (FEVD) methodology which was recently developed by Plumper and Troeger (2007) to specifically address the problem of performing fixed-effects analysis with time-invariant variables or nearly time invariant variables. This procedure involves three stages. In the first stage, a fixed-effects model is estimated without the time-invariant or nearly time invariant independent variables. In the second stage, the fixed-effects vector is decomposed into two parts, the portion that is attributable to the time-invariant or nearly time invariant variables and the error term. Specifically, the fixed-effects vector is regressed on the time-invariant or nearly time invariant variables. In the third and final stage, the time-invariant or nearly time invariant variables and the error term from the second stage are added to the original model from the first stage and it is re-estimated using pooled OLS.5 This procedure has

5

For a more detailed discussion of the consequences of the use of the traditional fixed effects methodology with time invariant or nearly time invariant variables, see Woolridge (2002). In addition, for a more detailed description of the FEVD procedure, see Plumper and Troeger (2007).

11

been used in several recently published articles by researchers facing similar data issues [e.g., Steinberg and Saideman (2008), Krogstrup and Wälti (2008), and Alemán (2008)].6

4. Discussion of Dependent and Explanatory Variables While the key point of interest is the relation of the insurer’s risk and the ownership form of the firm, we also control for the other factors known to impact firm risk. These factors include firm size, distribution system, the use of outside directors, the amount of reinsurance ceded, number of insurers licensed in the state of domicile, and business mix. The dependent and independent variables are discussed in detail below. 4.1. The Dependent Variable – Natural Log of BCAR Our measure of risk is the Best’s Capital Adequacy Ratio, often known as BCAR. 7 According to A. M. Best, BCAR is defined as the ratio of adjusted surplus over net required capital.

Adjusted surplus consists of the following components: reported surplus, equity

adjustments (i.e., unearned premiums, assets, loss reserves, and reinsurance), debt adjustments (i.e., surplus notes and debt service requirements), and other adjustments (i.e., potential catastrophe losses and future operating losses). Net required capital is calculated by applying risk charges to the following components: fixed-income securities, equity securities, interest rate, credit, loss and loss-adjustment-expense reserves, and off balance sheet items [A. M. Best (2003), p.1]. While BCAR is designed very similarly to the NAIC’s risk-based capital (RBC) calculation, it also includes several critical risk components not considered in NAIC’s RBC 6

For comparison purposes, results of the models obtained using OLS controlling for the time effects are included in the appendix. However, given the potential econometric issues created due to the inclusion of nearly time invariant variables, as well as the rejection of the appropriateness of the pooled-OLS framework by the F-test discussed above, we feel that great caution should be used in interpreting these results. The FEVD approach has a much stronger theoretical foundation in our situation, so we focus only on these results for the remainder of the article. 7 Since BCARs are not normally distributed in our sample, we use the natural logarithm instead of the original value of BCARs in order to reduce the skewness of the distribution.

12

calculation, such as total investment risk and credit risk (A. M. Best, 2003). In fact, Pottier and Sommer (2002) find that BCAR dramatically outperforms the NAIC’s RBC ratio in predicting insurer insolvencies. To reduce the possibility that results are driven by outliers, we truncate the sample by removing observations whose BCAR values are outside the 1st and 99th percentiles.8 Table 1 reports the descriptive statistics of BCAR by year. 9 Both the mean and median level of BCAR were lower in the first three years than the last five years for the sample period of 1996 to 2004, suggesting that property-liability companies on average are taking less risk, relative to their capital, in more recent years than they were five to eight years earlier. Such a trend also is reflected in Exhibit 1, which maps both the mean and median BCAR over the sample period.10 [Insert Table 1 Here] [Insert Exhibit 1 Here]

4.2. Explanatory Variables 4.2.1. Key Explanatory Variable(s) – Ownership Indicator(s) As noted earlier, the key explanatory variables of interest are the ownership indicator variables. To make our study more comparable to existing studies in this area, we first conduct

8

BCAR has a maximum value of 999.9. The higher the value of BCAR, the lower firm risk. Truncating at the 99th percentile of BCAR removes all observations with a BCAR of 999.9. 9 Note that there are fewer observations in 1999 than in any other year, due to the significantly higher number of missing values of BCAR in Best’s Key Rating Guide 1999. We have no explanation for this large number of missing values. We consulted two different copies of the 1999 Key Rating Guide file, and both contained the same missing values. As a robustness test, we run all of our empirical models dropping all 1999 observations. The results for the key explanatory variable(s) are not sensitive to whether or not data from 1999 are included. 10 Note that we also perform multiple robustness tests to determine if some irregularities observed among BCAR ratios had any impact on the results obtained. These results are discussed at the end of Section 5, “Empirical Results”.

13

our analysis with the same ownership classifications as these previous studies. We then expand the analysis to include the more refined ownership classifications unique to our study. Specifically, we first compare mutual and stock insurers as did Lamm-Tennant and Starks (1993) and Lee et al. (1997). For this part of the analysis, we include in the regression an indicator variable for stock insurers, with mutual insurers being the omitted category. We then compare mutual-owned stocks with stock-owned stocks, as did Mayers and Smith (1994) and Lee et al. (1997). In this portion, we include an indicator variable for stock-owned stock insurers, with mutual-owned stock insurers being the omitted category.

We then compare

widely-held stock insurers and closely-held stock insurers, similar to Cummins and Sommer (1996).11 For this part of the analysis, the ownership variable is an indicator variable for widelyheld stock insurers, with closely-held stock insurers being the omitted category. Lastly, we examine the full spectrum of ownership structures in our sample. In this case, we include in the regression a series of indicator variables equal to one if the insurer has that particular ownership structure and zero otherwise. The structures included are: mutual, mutualowned stock, widely-held stock, and stock closely-held by others, with stock closely-held by management being the omitted category. As we move from one side of the spectrum (i.e., stocks closely-held by management) to the other (i.e., mutual insurers), the degree of separation of ownership and management increases accordingly. Such detailed classification enables us to conduct a more complete analysis than previous studies.

11

Specifically, Cummins and Sommer (1996) consider three ownership structures among stock companies (i.e., widely-held, closely-held by management and closely-held by others).

14

4.2.2. Control Variables Drawing from prior research, we incorporate additional variables in the models to control for other factors that may be related to firm risk.12 These include firm size, an indicator for firms with independent agency marketing system, the fraction of outside directors on the company board of directors, reinsurance ceded as a fraction of direct premiums written, number of insurers licensed in the state of domicile, and line-of-business controls. Since larger firms are expected to be more diversified, these firms have a greater ability to take risks.13 As such, we would expect firm size to be positively related to risk. Firm size is measured as the natural logarithm of total assets. A dichotomous measure of the type of agency system used controls for the potential impact of the distribution system on firm risk-taking behavior. The variable is equal to one if the firm utilizes independent agents and zero otherwise. Given the advantages of independent agents in markets with more heterogeneous and/or complex risks [Regan and Tennyson (1996)], we expect the independent agency indicator to be positively related to firm risk. The board independence variable, defined as the percentage of non-officer directors, also is expected to impact firm risk-taking behavior. Specifically, outside directors are likely more independent from the officers of the firm and thus more effective monitors of a firm’s executives [Weisbach (1988)]. However, the impact of outside directors on firm risk is ambiguous. On one hand, the board of directors represents the interest of shareholders, and as a result, may encourage certain risk-taking behaviors with the ultimate goal being to maximize shareholder 12

It can be argued that since BCAR is such a comprehensive risk measure, including both quantitative and qualitative measures, most of these control variables would already be captured in the insurer’s BCAR. As noted by Pottier and Sommer (2002), the exception would be firm size, as the authors find that the impact of firm size is not fully reflected in the insurer’s BCAR. As a robustness test, we run all of the models including only the ownership variables and firm size as independent variables. The key results, as discussed at the end of Section 5, are qualitatively the same as those reported. 13 Note that BCAR should incorporate an insurer’s level of diversification. If this is the case, then it is possible that we would not expect to observe a relation between BCAR and firm size.

15

value. On the other hand, the board of directors is concerned with the insolvency risk of the firm and may want to limit the risk of the company, especially since insolvency can impact the reputation of the board members. As such, we offer no predictions for the relation between firm risk and the fraction of outside board of directors. Since reinsurance can be used to transfer particular exposures or lines of business, a firm can utilize reinsurance transactions to reduce its overall risk. As such, we include reinsurance ceded, measured as the volume of reinsurance ceded to affiliates and non-affiliates relative to direct business written plus reinsurance assumed from both affiliates and non-affiliates, in the model. If reinsurance is used to reduce risk, we would expect to see a positive relation between the reinsurance ceded variable and BCAR, assuming the impact of greater reinsurance is not offset by a reduced holding of capital. All states as well as the District of Columbia and Puerto Rico have established guaranty funds covering most lines of property-liability insurance. Guaranty funds are believed to create incentives for solvent insurance companies to monitor their competitors and report unsafe practices to state regulators, and thus provide monitoring of insurers [Lee et al. (1997)]. Such monitoring should curtail excessive risk-taking by insurers. In fact, Munch and Smallwood (1980) find that guaranty fund operations have reduced insolvencies, consistent with the notion that other companies to whom liability for insolvencies is shifted by creation of a guaranty fund are more efficient monitors of potentially weak companies than are consumers [p. 274]. This monitoring hypothesis implies a negative relation between number of insurers in the state of domicile and the degree of risk an insurer is taking. However, monitoring by peer insurers may not necessarily occur when an insurer takes a level of risk which is not enough to put the company near insolvency. In addition, insurers can often recover guaranty fund payments

16

through rate increases or tax offsets [Lee et al. (1997)]. These factors weaken the incentive of insurers to mutually monitor one another, and so we may not observe any relation between the number of insurers in a state and risk-taking behavior. Thus, the impact of the number of peer insurers in a state on an insurer’s risk-taking behavior remains an empirical question. Finally, we include variables measuring the percentage of total premiums written in each line of business in the model.14 Prior research has shown that different lines of business have different payout periods as well as variation in loss ratios [i.e., Mayers and Smith (1981), Mayers and Smith (1988), Lamm-Tennant and Starks (1993), and Pottier and Sommer (1997)]. As such, these variables are included to control for the impact of the type of business written by the insurer on overall risk. 5. Empirical Results Summary statistics for the variables utilized in the models are shown in Table 2. An examination of the pairwise correlations as well as the variance inflation factors generally does not indicate a problem with multicollinearity.15 Results of the variance inflation factors check can be found in the appendix. In addition, the Breusch-Pagan/Cook-Weisberg test indicates the presence of heteroskedasticity (p-value < .0001), and the Wooldridge test for autocorrelation indicates the presence of serial correlation (p-value < .0001).16 As such, robust standard errors are reported for all regressions.

14

The model includes the following lines: fire, allied lines, farmowners multiple peril, homeowners multiple peril, commercial multiple peril, mortgage guaranty, ocean marine, inland marine, financial guaranty, medical malpractice – occurrence, medical malpractice – claims-made, earthquake, group accident and health, credit accident and health (group and individual), other accidental health, workers’ compensation, other liability-occurrence, other liability-claims made, private passenger auto liability, commercial auto liability, auto physical damage, aircraft (all perils), fidelity, surety, burglary and theft, credit, international; with aggregate write-ins for other lines of business being the omitted category. 15

The only exception is that two variables, firm size and the number of insurers, have high variance inflation factors. To ensure that the results are not driven by these two variables, we rerun the models alternately dropping these two variables as a robustness check. The results are discussed at the end of this section. 16 To test for heteroskedasticity, we used the “estat hettest” command in STATA. To test for serial correlation, we used the “xtserial” command in STATA.

17

[Insert Table 2 Here] As seen in Panel A of Table 2, the predominant ownership type is widely-held stock companies (676 firms), followed by mutual companies (377), stock companies closely-held by management (157), and mutual-owned stock companies (147). Much less common are stock companies closely-held by others (31). Regression results are shown in Tables 3 and 4. For all regressions, the fixed effects vector decomposition (FEVD) method is used. Table 3 reports results of the models replicating the ownership categorizations utilized in prior literature. The first column shows the results in which only one ownership variable, indicating whether a firm is a stock or a mutual, is considered. As expected, stock insurers are associated with lower BCAR and thus higher risk than mutual insurers, consistent with prior findings in Lamm-Tennant and Starks (1993) and Lee et al. (1997). [Insert Table 3 Here] Among stock ownership classes, we also differentiate between mutual-owned stock insurers and stock-owned stock insurers. As shown in the second column of Table 3, stockowned stock insurers exhibit significantly lower BCAR than mutual-owned stock insurers, suggesting that stock-owned stock insurers are associated with higher risk than mutual-owned stock insurers.17 As noted earlier, among stock-owned stock insurers, we further distinguish between widely-held stock insurers and closely-held stock insurers. The results of this comparison are reported in the third column of Table 3. The results show that widely-held stock insurers exhibit

17

Whether mutual-owned stocks behave more like mutual insurers, as shown in Mayers and Smith (1994) will be addressed in the last part of the analysis when we examine all five ownership classes.

18

significantly lower BCARs than closely-held stock insurers, suggesting that widely-held stock insurers are engaging in riskier behavior than the latter. This is contrary to the findings in Cummins and Sommer (1996), but consistent with the argument by Fama and Jensen (1983) that ownership concentration in closely-held companies results in less risk taking as the wealth of these individuals is less diversified than in widely-held companies.18 The final part of our analysis employs a more refined classification scheme which includes five ownership classes, namely, mutuals, mutual-owned stocks, widely-held stocks, stocks closely-held by others, and stocks closely-held by management. The results are shown in Table 4. Overall, the results indicate that each ownership class is significantly different from all other ownership classes in regards to risk. Specifically, mutual insurers are associated with lower risk than all stock ownership classes except stocks closely-held by management. Interestingly, mutual-owned stock insurers are significantly different from both mutual insurers and all of the stock ownership classes.19 Among stock-owned stock insurers, as the degree of separation of ownership and control increases from stocks closely-held by management, to stocks closely-held by others, to widelyheld stocks, the level of overall firm risk also increases.

Finally, stocks closely-held by

management appear to be the least risky of all ownership classes under investigation. Taken together, these results are consistent with the downward influence on risk-taking behavior resulting from sub-optimal diversification. In addition, the finding that stocks closely-held by management are less risky than all other ownership structures is evidence that the influence of

18

The results of these same models estimated with a traditional pooled OLS estimation with time effects are included in the appendix. The results for all three models are generally consistent with Table 3. In terms of the variables of interest, there is the loss of significance of the widely-held stock indicator in the model in column 3. 19 When estimated as an OLS model, the mutual and mutual-owned stock variables are positive and significant, and not all ownership structures are significantly different from one another. Full results from this estimation can be found in the appendix, but as stated previously, these OLS results should be interpreted with caution.

19

sub-optimal diversification is especially strong when owners/managers have both their financial capital and human capital concentrated in the same firm. [Insert Table 4 Here] In addition to the organizational form variables, several control variables also are significant. For example, the number of insurers licensed in the state of domicile is significant and positive. This is consistent with the monitoring hypothesis of Lee et al. (1997) and serves as evidence that monitoring among peer insurers results in less risky behavior. The percent of reinsurance ceded variable is significant and positive, suggesting that insurers using more reinsurance are associated with lower overall firm risk. Finally, several line-of-business control variables are significant, suggesting that the type of business written impacts firm risk.20 To check the robustness of our results, we run several model variations.21 First, since the number of observations in year 1999 is significantly lower than in any other sample year due to the significantly higher volume of missing values for BCAR reported in Best’s Key Rating Guide 1999, we run the regressions reported in Table 3 and Table 4 excluding all observations of year 1999. The key results are qualitatively the same as those reported. Moreover, the mean (and median) value of BCAR is significantly higher for the period starting 1999 compared to pre-1999. To reduce the possibility that our results are driven by any possible structural break in BCAR, we also repeat all regressions with data from year 1999 and after and again, the key results are qualitatively the same as those reported. In addition, we run the models without the line of business controls and the key results are not sensitive to whether these line-of-business controls are included. 20

The results for the line of business control variables are not shown to conserve space. However, the results are available from the authors upon request. 21 We thank an anonymous referee for suggesting these additional tests. To save space, the results of these additional tests are not reported here. However, the results are available from the authors upon request.

20

As noted earlier, firm size and the number of insurers have high variance inflation factors. To ensure that this does not impact the results obtained, we run the models alternately excluding these two variables. The results are very similar to those reported, with one exception. When firm size is excluded from the regression, the comparison between mutuals and stocks closely-held by management is still significant, but the results indicate that mutuals are associated with lower risk than stocks closely-held by management. Finally, given that BCAR is such a comprehensive measure of risk, it could be argued that the control variables are already captured within the insurer’s BCAR. As such, we run the models including only the ownership variables and firm size. We leave size in the model because the results of Pottier and Sommer (2002) indicate that BCAR does not fully reflect the impact of size on insurer risk. The results for all of the ownership variables are consistent with those reported.

In addition, size is

significant and positive. 6. Conclusion Taking advantage of the large variation of ownership structures in the property-liability insurance industry, we examine the implications of separation of ownership and management for firms’ risk-taking behavior. The incentive misalignment hypothesis, derived from Cummins and Sommer (1996), argues that in firms with significant separation of ownership and management, managers may engage in risk averse behavior due to their undiversified human capital investment in the firm. This suggests a negative relation between the degree of separation of ownership and management and firm risk-taking behavior.

However, the sub-optimal

diversification hypothesis, derived from Fama and Jensen (1983), argues that ownership concentration in closely-held companies results in less risk taking as the wealth of these individuals are more concentrated (or less diversified). This implies that risk taking will be

21

higher in widely-held companies than in closely-held companies. Employing a large sample of property-liability companies over the period of 1996 to 2004, we empirically test these alternative theories using a detailed ownership classification scheme in an effort to examine the implication of ownership structure on risk-taking behavior in the U.S. property-liability insurance industry. In addition, the analysis is replicated employing the ownership classification schemes of prior research to determine if the classification scheme utilized impacts the results obtained and to offer a more direct comparison of our results to those in existing literature. Overall, our results show that ownership structures have a significant impact on firms’ risk-taking behavior. Specifically, utilizing a simplified ownership classification scheme, we find that stock insurers are more risky than mutuals.

However, when a more detailed

classification scheme is employed, the results indicate that stock insurers closely-held by management are the least risky firms, followed by mutuals, mutual-owned stocks, stocks closelyheld by others, and widely-held stocks.

These results are consistent with the downward

influence on risk-taking behavior resulting from sub-optimal diversification of the owners. The greatest impact is observed for stocks closely-held by management, as these are the firms in which the degree of sub-optimal diversification is likely the greatest. In an era marked by heightened concern about corporate responsibility and risk-taking incentives, these results are important to a variety of stakeholders. Among those most concerned are likely the regulators who are tasked with ensuring solvency in the industry.

22

References A. M. Best (2003). “Understanding BCAR.” A. M. Best Company, Inc., November 24, 2003 Alemán, J. (2008). "Labor market deregulation and industrial conflict in new democracies: a cross-national analysis." Political Studies 56(4): 830-856. Baum, Christopher F (2006). An Introduction to Modern Econometrics Using Stata, State Press, College Station, Texas. Coase, R. H. (1960). "The problem of social cost." Journal of Law and Economics 3(October): 144. Cummins, J. D. and D. W. Sommer (1996). "Capital and risk in property-liability insurance markets." Journal of Banking & Finance 20(6): 1069-1092. Downs, D. H. and D. W. Sommer (1999). "Monitoring, Ownership, and Risk-Taking: The Impact of Guaranty Funds." The Journal of Risk and Insurance 66(3): 477-497. Fama, E. F. and M. C. Jensen (1983). "Separation of Ownership and Control." Journal of Law and Economics 26(2, Corporations and Private Property: A Conference Sponsored by the Hoover Institution): 301-325. He, E. and D. W. Sommer (2006a). "Separation of Ownership and Control: Implications for Board Composition." University of Georgia Working Paper. Kim, W.-J., D. Mayers and C. W. Smith, Jr. (1996). "On the Choice of Insurance Distribution Systems." Journal of Risk & Insurance 63: 207-227. Krogstrup, S. and S. Wälti (2008). "Do fiscal rules cause budgetary outcomes?" Public Choice 136: 123-138. Lamm-Tennant, J. and L. T. Starks (1993). "Stock Versus Mutual Ownership Structures: The Risk Implications." Journal of Business 66(1): 29-46. Lee, S.-J., D. Mayers and J. C. W. Smith (1997). "Guaranty funds and risk-taking evidence from the insurance industry." Journal of Financial Economics 44(1): 3-24. Mayers, D., A. Shivdasani and C. W. Smith, Jr. (1997). "Board Composition and Corporate Control: Evidence from the Insurance Industry." Journal of Business 70(1): 33-62. Mayers, D. and C. W. Smith, Jr. (1981). "Contractual Provisions, Organizational Structure, and Conflict Control in Insurance Markets." Journal of Business 54(3): 407-434. Mayers, D. and C. W. Smith, Jr. (1988). "Ownership Structure across Lines of Property-Casualty Insurance." Journal of Law and Economics 31(2): 351-378. Mayers, D. and C. W. Smith, Jr. (1990). "On the Corporate Demand for Insurance: Evidence from the Reinsurance Market." The Journal of Business 63(1): 19-40. Mayers, D. and C. W. Smith, Jr. (1992). "Executive Compensation in the Life Insurance Industry." Journal of Business 65(1): 51-74. Mayers, D. and C. W. Smith, Jr. (1994). "Managerial Discretion, Regulation, and Stock Insurer Ownership Structure." Journal of Risk and Insurance 61(4): 638-655. Munch, P. and D. E. Smallwood (1980). "Solvency regulation in the property-liability insurance industry: empirical evidence." Bell Journal of Economics 11: 261-279. Plumper, T. and V. E. Troeger (2007). "Efficient Estimation of Time-Invariant and Rarely Changing Variables in Finite Sample Panel Analyses with Unit Fixed Effects." Political Analysis 15(2): 124-139. Pottier, S. W. and D. W. Sommer (1997). "Agency Theory and Life Insurer Ownership Structure." Journal of Risk and Insurance 64(3): 529-543.

23

Pottier, S. W. and D. W. Sommer (2002). "The Effectiveness of Public and Private Sector Summary Risk Measures in Predicting Insurer Insolvencies." Journal of Financial Services Research 21(1-2 (February 2002)): pp. 101-116. Regan, L. and S. Tennyson (1996). "Agent Discretion and the Choice of Insurance Marketing System." Journal of Law and Economics 39: 637-666. Regan, L. and L. Y. Tzeng (1999). "Organizational Form in the Property-Liability Insurance Industry." Journal of Risk and Insurance 66(2): 253-273. Sanders, A., E. Strock and N. G. Travlos (1990). "Ownership structure, deregulation, and bank risk taking." Journal of Finance 45: 643-654. Steinberg, D. A. and S. M. Saideman (2008). "Laissez Fear: Assessing the Impact of Government Involvement in the Economy on Ethnic Violence." International Studies Quarterly 52: 235-259. Weisbach, M. S. (1988). "Outside directors and CEO turnover." Journal of Financial Economics 20: 431-460. Wooldridge, J. M. (2002). Econometric analysis of cross section and panel data. Cambridge, Massachusetts, The MIT Press.

24

Table 1: Descriptive Statistics of BCAR by Year22

Year 1996 1997 1998 1999 2000 2001 2002 2003 2004 Total

# of Obs 1,187 1,176 1,150 651 999 966 964 933 865 8,891

Mean 145.04 121.93 128.73 220.05 227.51 215.60 210.89 216.31 224.21 184.62

Median 119.7 99 111.9 173.2 189.5 177.9 176.7 187.7 194 157.60

SD 90.41 85.83 79.64 129.25 125.43 120.45 117.21 108.24 110.76 115.00

Min 6.6 0.7 10 24.4 25.2 38.8 32.6 31.4 54.8 0.70

Exhibit 1: Trend in BCAR from 1996 to 2004

250.00

200.00

150.00 Mean Median 100.00

50.00

0.00 1996

22

1997

1998

1999

2000

2001

Higher BCAR represents lower firm risk.

25

2002

2003

2004

Max 866.6 874.3 867.7 908.2 915.3 832.7 913.4 932.6 906.7 932.60

Table 2: Summary Statistics Panel A: Summary Statistics of Ownership Classes Ownership Type Mutual Stock Mutual-Owned Stock Widely-Held Stock Closely-Held by Others Stock Closely-Held by Management

# of Firms 377 147 676 31 157

Panel B: Summary Statistics of Independent Variables23 Variable N Mean Std. Dev. Min Max Log of BCAR 8,891 5.06 0.55 -0.36 6.84 Mutual 8,891 0.29 0.45 0 1 Mutual-Owned Stock 8,891 0.12 0.32 0 1 Widely-Held Stock 8,891 0.47 0.50 0 1 Stock Closely-Held by Others 8,891 0.03 0.17 0 1 Firm Size 8,891 18.35 1.82 13.19 25.12 Number of Insurers Licensed in the State of Domicile 8,891 991.35 156.04 84 1248 Percentage of Outside Board Members 8,891 0.55 0.29 0 1 Independent Agency Indicator 8,891 0.70 0.46 0 1 Percentage of Reinsurance Ceded 8,891 0.40 0.29 0 1 The variables are defined as follows: stock mutual-owned = 1 if the stock company is ultimately owned by a mutual company; stock widely-held = 1 if the stock company itself or its ultimate parent is publicly traded; stock closely-held by others = 1 if the stock company is closely-held by parties other than management; stock closely-held by management = 1 if the stock company is closely-held by management and/or their relatives; mutual = 1 if mutual; firm size = natural log of total assets; percentage of outside board members = total number of outside directors divided by total number of directors on board, where outside directors are defined as non-officer directors; independent agency indicator = 1 if the company uses independent agency marketing system; percent of reinsurance ceded = reinsurance ceded to affiliates and non-affiliates divided by direct premiums written plus reinsurance assumed from affiliates and non-affiliates.

23

As noted in the discussion, line-of-business controls are included in the models. However, the summary information for these variables is not shown in the table to conserve space. This information is available from the authors upon request.

26

Table 3: FEVD Estimation of Firm Risk - Paired Ownership Classes Dependent variable = Natural log of BCAR ratios Mutual vs. Stock

Stock indicator

FEVD Estimates [Standard Errors] -0.278*** [0.001]

Mutual-owned Widely-held Stock vs. Stock- Stock vs. Closelyowned Stock held Stock FEVD Estimates FEVD Estimates [Standard Errors] [Standard Errors]

-0.183*** [0.002]

Stock-owned stock indicator

-0.330*** [0.002]

Widely-held stock indicator -0.009*** [0.002]

-0.059*** [0.002]

-0.105*** [0.003]

0.165 [0.101]

0.129 [0.092]

0.12 [0.102]

0.002*** [0.0001]

0.002*** [0.0001]

0.002*** [0.0001]

-0.002 [0.030]

-0.027 [0.026]

-0.025 [0.022]

% of reinsurance ceded

0.279*** [0.037]

0.209*** [0.035]

0.192*** [0.037]

Constant

-0.212*** [0.001]

0.723*** [0.002]

0.903*** [0.002]

Yes 8,891 0.62

Yes 6,336 0.60

Yes 5,306 0.60

Independent agency indicator Firm size # of insurers licensed in the state of domicile % of outside board members

Line of business controls Included Observations R-squared

* Significant at 10%; ** significant at 5%; *** significant at 1%; the sample is truncated at the 1st and 99th BCAR percentiles; and standard errors, as shown in brackets, are asymptotically robust to heteroskedasticity and serial correlation. The variables are defined as follows: stock indicator = 1 if stock and 0 otherwise; stock-owned stocks indicator = 1 if a stock company is ultimately owned by a stock parent and 0 otherwise; widely-held stocks indicator = 1 if the stock company itself or its ultimate parent is publicly traded and 0 otherwise; firm size = natural log of total assets; percentage of outside board members = total number of outside directors divided by total number of directors on board, where outside directors are defined as non-officer directors; independent agency indicator = 1 if the company uses independent agency marketing system; percent of reinsurance ceded = reinsurance ceded to affiliates and non-affiliates divided by direct premiums written plus reinsurance assumed from affiliates and non-affiliates; line of business controls are variables measuring the percentage of total premiums written in each of the following lines of business: fire, allied lines, farmowners multiple peril, homeowners multiple peril, commercial multiple peril, mortgage guaranty, ocean marine, inland marine, financial guaranty, medical malpractice – occurrence, medical malpractice – claims-made, earthquake, group accident and health, credit accident and health (group and individual), other accidental health, workers’ compensation, other liability-occurrence, other liability-claims made, private passenger auto liability, commercial auto liability, auto physical damage, aircraft (all perils), fidelity, surety, burglary and theft, credit, international.

27

Table 4: FEVD Estimation of Firm Risk – Full Sprectrum of Five Ownership Classes Dependent variable = Natural log of BCAR ratios FEVD Estimates [Standard Errors] -0.083*** [0.003]

Mutual indicator (ß1) Mutual-owned stock indicator (ß2)

-0.191*** [0.003]

Widely-held stock indicator (ß3)

-0.492*** [0.004]

Stock closely-held by others indicator (ß4)

-0.233*** [0.010]

Wald Test p -value: ß1 = ß2