Insurer Opacity and Ownership Structure

Transcription

1 Insurer Opacity and Ownership Structure Stanley R. Adamson, 1 David L. Eckles, 2 and K. Stephen Haggard 3 Abstract: We examine the differences in opacity among insurers based on differences in their ownership structures using split bond ratings as a proxy for opacity. Our analyses are consistent with mutual insurers being more opaque than stock insurers. These findings are consistent with the owners of mutual insurers being so dispersed and having managed the owner/manager conflict well enough that the opacity of mutual insurer operations is less important to their owners. On the other hand, owners of stock insurers demand relatively less opaque behavior on the part of managers. This finding has important implications for investors, rating agencies, and regulators charged with evaluating the financial condition of insurers. T INTRODUCTION his paper investigates the differences in opacity among insurers based on their ownership structure. The literature shows that financial institutions are more opaque than non financial firms, and that insurers tend to be among the most opaque of all industry classes (Morgan, 2002). Further research has shown that publicly traded property/casualty companies are relatively opaque, confirming Morgan s findings (Park, 2008). Our study incorporates the theories of ownership structure in the context of agency theory with the opacity studies cited above to determine whether there is a difference in opacity among insurers attributable to differences in ownership form (mutual versus stock). We define opacity as the uncertainty that even the most sophisticated investors face in accurately assessing the fundamental value of a firm (Jones, Lee, and Yeager, 2012: 383). Such uncertainty might result when a firm chooses to withhold information from investors, which creates infor 1 Associate Professor, Missouri State University, StanleyAdamson@MissouriState.edu 2 Associate Professor, University of Georgia, deckles@uga.edu 3 Associate Professor, Missouri State University, SHaggard@MissouriState.edu 93 Journal of Insurance Issues, 2014, 37 (2): Copyright 2014 by the Western Risk and Insurance Association. All rights reserved.

2 94 ADAMSON, ECKLES, AND HAGGARD mation asymmetry. Such firm controllable opacity would be punished in an efficient market. However, even full disclosure, as provided to insurance company regulators and investors through reports and audits, might not eliminate opacity if such disclosure is not credible or if investors interpret the information in contradictory ways. Furthermore, the inherent complexity of a business and the nature of its underlying assets can also contribute to its opacity. Our study tests for opacity differences among insurers using split bond ratings. Split bond ratings are instances of disagreement between Standard and Poor s (S&P) and Moody s bond ratings and can be used as a proxy for opacity (Morgan, 2002). The rationale for using such differences as an opacity proxy is that raters will disagree more often when debt issuer opacity prevents ratings firms from precisely determining the default probability of issuers. LITERATURE REVIEW As our paper considers insurer opacity with respect to ownership structures, we highlight below the relevant literature for both insurer opacity and insurer ownership structure, noting the relevant overlaps. Opacity Opacity among insurers is not surprising. The insurer underwrites insurance policies whose actuarial loss estimation is both uncertain and unknown to outsiders (bond raters or investors in the case of stock insurers). Also, the loss reserve estimate comprises the largest part of the firm s liabilities and disclosure of such details and proprietary methodologies used in arriving at this estimate might imperil the firm s competitive advantage in the marketplace. Yet because of the inherent opacity brought about by this necessary practice, substantial managerial discretion is employed in setting these most critical reserves (Petroni, 1992). Insurers also have large financial assets that can often be transformed quickly and efficiently (Merton and Perold, 1993). Sometimes a large portion of such assets are highly liquid due to the need to quickly convert assets to cash in order to meet business demands. Myers and Rajan (1998) discuss the duality of such liquid assets. While liquid assets should be highly transparent, the ability to convert such assets to completely different positions quickly and efficiently brings about another source of potential opacity for insurers. Zhang, Cox, and Van Ness (2009) find that insurers underwriting more opaque lines of business are subject to higher adverse selection costs. 4 Yet

3 INSURER OPACITY AND OWNERSHIP STRUCTURE 95 they find little evidence of a relationship between asset opacity and adverse selection. Using one year of data, Pottier and Sommer (2006) find more difficulty in rating (i.e., greater opacity among) property and casualty insurers exhibiting the following characteristics: smaller size, stock ownership form, a history of reserving errors, less reinsurance, greater holdings of stocks and low grade bonds, and greater geographic diversification. We differ from Pottier and Sommer (2006) in at least three important ways. First, we examine both property/liability and the life/health segments of the insurance industry. Second, we utilize a panel dataset (as opposed to only one year of data) to test our hypothesis. And third, we examine split ratings on bond issuances by insurers as the proxy for opacity. Pottier and Sommer (2006) use split financial strength ratings in their analysis. A final area that might create opacity among insurers is their relatively larger holdings of privately placed bonds. 5 Like banks, insurers must collect a great deal of information concerning these borrowers. Additionally, such private information leads to information asymmetry, as outsiders do not possess the same information. Indeed, Park (2008) finds that, in addition to risky underwriting activities, the ownership of certain assets (including privately placed bonds) appears to be a source of opacity. Ownership Structure There are two major organizational forms in the insurance industry stock firms and mutual firms. These two forms have distinct characteristics; in particular, in the way they manage the three important functions of an insurer: (1) the managerial function, (2) the owner/risk bearing function, and (3) the customer/policyholder function. Each ownership structure combines these functions in different ways. In the mutual insurer, the owner/risk bearing and customer/policyholder functions are merged. This merger should give the mutual insurer a competitive advantage in controlling the incentive problems between policyholders and owners as they are 4 Zhang, Cox, and Van Ness (2009) use various definitions of opaque lines. With the exception of accident/health, they consider all life insurance based lines to be transparent. Longtailed property/casualty lines such as medical malpractice, workers compensation, auto liability, products liability, other liability, boiler and machinery, aircraft, ocean marine, farmowners multi peril, homeowners multiperil, commercial multi peril, international, and reinsurance are considered opaque. 5 According to the National Association of Insurance Commissioners, life insurers in 2012 held $726 billion in privately placed bonds (accounting for 28% of their bond portfolio). Property and casualty insurers had approximately 8% of their bond portfolio in private placements ($79 billion) in To give an idea as to the scale, the total volume of new private bond placements in 2012 was $54 billion (see placements.htm).

4 96 ADAMSON, ECKLES, AND HAGGARD one and the same. However, this form of ownership has an incentive problem that is worsened between owners and managers as compared to stock firms, as ownership is widely dispersed among all policyholders/ owners. As long as these owners are satisfied that they will be indemnified in the case of a loss, they have little interest in the transparency or opacity of the firms financial reporting. Thus, if the costs of controlling management in mutual firms are higher than in stock firms, one might expect mutual firms to dominate in lines of business where management exercises little discretion as compared to stock companies (Mayers and Smith, 1988). This is referred to as the managerial discretion hypothesis and is confirmed in the research of Mayers and Smith. With regard to opacity and ownership structure, one might assume that mutual insurers would be more opaque than stock insurers because of the lack of controls and the costs of overseeing managers day to day activities. The stock options and other incentives that align managers interests with owners in stock insurers are absent in the mutual ownership structure. In addition, especially for publicly held stock insurers, reporting requirements and disclosures are more rigorous than those required of mutual firms. Finally, the market discipline faced by publicly traded firms would suggest that stock firms would be less opaque than mutual firms. Indeed, the prior literature has shown that mutual firms and stock firms are significantly different in many ways. This difference is first identified by Mayers and Smith (1981) and confirmed by numerous empirical studies. For example, Mayers and Smith (1988) find significant differences in both geographical and line of business concentrations related to forms of business ownership. In 1992, the same authors find life insurer CEO compensation to be higher in stock firms than in mutual firms. Kim, Mayers and Smith (1996) document a strong relationship between ownership structure and the choice of distribution systems. Pottier and Sommer (1997) find systematic differences in both operational and financial factors with regard to life insurer ownership structure consistent with the managerial discretion hypothesis. In addition to operational significance, ownership structure also plays a role in the insurers risk taking behavior. Lamm Tennant and Starks (1993) find that stock insurers tend to do more business in higher risk states than do mutuals. The stock insurers also participate in higher risk lines of business to a greater extent than do mutual insurers. Lee, Mayers, and Smith (1997) find that stock insurers take more investment risks after the enactment of state guaranty funds while mutual insurers do not. Finally, Cole et al. (2011) find that insurers risktaking behavior is significantly different according to ownership structure. These studies are consistent with greater separation of ownership and control at mutual insurers being associated with less risk taking.

5 INSURER OPACITY AND OWNERSHIP STRUCTURE 97 HYPOTHESIS Insurers operate a complicated business. Managers (i.e., underwriters and actuaries) must, ex ante, determine the riskiness of, and appropriate pricing for, customers. Depending on the line of business, this process can be exceedingly difficult. More importantly (for our purposes), the difficulty in determining the appropriate riskiness of an insured gives rise to sophisticated and complicated tools utilized by insurers (e.g., modeling and insurance applications). Even after this initial screening, ex post, managers (i.e., claims managers and actuaries) must determine the amount of losses sustained in a book of business. For some lines, like fire insurance, determining these losses is rather straightforward. For others, like medical malpractice liability, the losses incurred in a policy year might not be realized or settled for several years. To this end, in their annual statements (statutory and publicly available), insurers report actual losses as well as estimated losses. Taken together, the entire risk bearing function of an insurer is rather opaque to third party evaluators (e.g., regulators, investors, and bond rating firms). We hypothesize that, ceteris paribus, the opacity of insurers will vary between stock and mutual forms of ownership. Mayers and Smith (1981) discuss the differences between these two forms of ownership. In particular, Mayers and Smith (1981) hypothesize that the different ownership forms are used to control managers. Specifically, managerial stock ownership aligns the incentives between owners and managers. To this end, owners of stock firms will presumably prefer less opaque processes, particularly with regards to financial reporting. Otherwise, if managers can sufficiently hide results behind more opaque financial reporting, the alignment of incentives offered by the stock ownership form is weakened. On the other hand, the mutual form of ownership creates atomistic ownership of the firm, weakening the control owners have over managers. To the extent that policyholders (owners) are satisfied that they will be indemnified in the event of a loss, the opacity of financial reporting is of lesser interest to a mutual firm owner. Taken together, we expect that insurers organized as stock firms will be less opaque. However, as Pottier and Sommer (2006) point out, it is possible that the managerial discretion associated with stock insurers may lead them to be more operationally complex and thus more difficult to evaluate (p. 155). Thus, the question of which insurer ownership form is more opaque remains one to be answered by our empirical tests. In order to examine the opacity differences between stock and mutual firms, we must control for business mix. With respect to insurer operations both before and after losses (i.e., the underwriting/ratemaking and claims

6 98 ADAMSON, ECKLES, AND HAGGARD settling/reserving phases), business mix impacts the opacity of insurer financial reporting. Short tail lines of business (i.e., those risks with relatively short terms to ultimate conclusion (e.g., auto property damage risks) are relatively easy to underwrite and relatively easy to close (i.e., observe and settle claims). Long tail lines (e.g., medical malpractice liability) are the opposite. The underwriting process is, relatively, quite stringent and claims can be open (or even unknown) for several years. Thus, from a purely operational standpoint, firms operating in more long tailed lines are likely to have more opaque financial reporting. DATA Data used for this study come from two sources. First, insurer firm data come from statutory filings to the National Association of Insurance Commissioners (NAIC) between 1994 and Both property/liability and life/health companies are included. Data from the property/liability and life/health files are combined for firms with operations in both, though we also test samples of pure property/casualty firms and pure life/health firms. Further, since the life/health data are not reported on a combined basis, we aggregate individual firm data to form group level results for both property/liability and life/health firms to maintain consistency. Our sample, therefore, includes group level and single unaffiliated insurers. Our second data source is the Securities Data Company (SDC) Platinum database, from which we obtain observations of debt issues over the same period. We start with 2,511 insurance related debt issues. Limiting our sample to issues that have ratings from both Moody s and S&P reduces our sample to 1,673 observations. Appendix A details the numeric coding of ratings following Park (2008).We collect information about the issue as well as information about the issuer, including the name of the issuer s parent firm. If a ticker symbol is present for the issuer or parent, we conclude that the issuer is a stock firm. We match the issuers from these observations with the NAIC data using the firm s group name. There are no unique identifiers universally shared between the two databases, increasing the complexity of merging these datasets. We first identify the company and date of debt issue in the SDC database, then search the NAIC dataset for the year of given debt issue. We match the names of the debt issuer with insurance company names in the NAIC database. When questions of proper matching persist, we perform internet searches to match 6 The 1994 financial statements are the first to include risk based capital (RBC) values, a variable used in our study.

7 INSURER OPACITY AND OWNERSHIP STRUCTURE 99 the debt issue with the proper insurer. While this process leads to successful identification in many cases, it does not lead to definitive matches in every case. If a match cannot be verified with certainty, the debt issue is dropped from the study. After matching the NAIC data with the SDC data, 1,104 observations remain. Appendix B details construction of our variables, which is inspired by Park (2008). RESULTS We examine life/health firms, property/casualty firms, and a combined sample of these firms to investigate the opacity differences between mutual and stock insurance companies. Table 1 presents descriptive statistics for our three samples, each divided into mutual and stock firms. Of our final combined sample of 1,104 debt issue observations, 125 are for mutual firms and 979 are for stock firms. Most variables of interest are available for all firms. However, six stock firm observations are missing data for RBCr, the ratio of risk based capital to surplus. YTM, the yield to maturity for the debt issue at the time of issuance, is missing for 56 mutual firm observations and 420 stock firm observations. Of the 308 debt issue observations in our life/health sample, 58 are from mutual firms and 250 from stock firms. Our sample for property/casualty firm debt issues is smaller at 205, only 10 of which come from mutual firms and 195 of which come from stock firms. Significant differences exist in our variables of interest between mutual firm observations and stock firm observations in all three samples. The frequency of a split bond rating is significantly different in all three samples between mutual and stock firms. In the combined sample, mutual firm bond issues have a percent greater probability of experiencing a split bond rating (t stat = 3.58) than stock firm bond issues, consistent with our hypothesis that mutual firms are more opaque. Similarly, in the property/ casualty sample, mutual firm bond issues have a percent greater probability of experiencing such a rating (t stat = 2.00). The exception to this pattern is for the life/health sample, where stock firm bond issues have a percent greater probability of receiving a split rating. Examining the absolute difference between numerical bond ratings provides a similar pattern, though the difference is no longer significant for the property/ casualty sample. Mutual firms hold significantly less liquid assets in the combined sample, but the difference is not significant in either of the two subsamples. The mutual firms in our combined sample hold significantly more private bonds as a proportion of total assets, as do the mutual firms of our life/ health subsample. For property/casualty companies, the difference is not

8 100 ADAMSON, ECKLES, AND HAGGARD Table 1. Descriptive Statistics 3 Panel A. Combined sample. Assets and capital in millions ($). Mutual Insurer Issues Variable N Minimum Mean Median Maximum Std Dev SPLIT ABSDIFF GROUP LIQUID ATOVER CAPITAL PRIVATE OPAQUE ASSET RESERVE PS RBCr REINSURANCE LONGTAIL PCPROP LBHERF GEOHERF AVGRTG CONVERTIBLE SUBORDINATED YTM Table continues statistically significant. We observe opposing signs on differences between mutual and stock firms between the two subsamples for the proportion of opaque assets and the average bond rating a debt issue receives. The mutual firms in all three samples have significantly greater reserves for

9 INSURER OPACITY AND OWNERSHIP STRUCTURE 101 Table 1. Panel A (continued) Stock Insurer Issues Variable N Minimum Mean Median Maximum Std Dev SPLIT ABSDIFF GROUP LIQUID ATOVER CAPITAL PRIVATE OPAQUE ASSET RESERVE PS RBCr REINSURANCE LONGTAIL PCPROP LBHERF GEOHERF AVGRTG CONVERTIBLE SUBORDINATED YTM This table presents descriptive statistics for both mutual and stock insurers. Variables are defined in Appendix B. Table continues losses as a proportion of their total liabilities, though this difference is not significant for property/casualty companies. Mutual firms in the combined sample and the life/health sample also have a significantly higher LONGTAIL (proportion of premiums written in long tailed lines), indicating that they engage in longer tail lines of

10 102 ADAMSON, ECKLES, AND HAGGARD Table 1. (continued) Panel B. Combined sample differences between mutual firms and stock firms. Assets and capital in millions ($). Comparison Variable Mutual Stock Equal Variances a t value SPLIT Yes 3.58*** ABSDIFF Yes 2.97*** GROUP No 2.02** LIQUID No 6.77*** ATOVER No 0.05 CAPITAL No 3.03*** PRIVATE Yes 6.80*** OPAQUE No 3.16*** ASSET No 2.50** RESERVE No 2.05** PS No 0.85 RBCr Yes 1.76* REINSURANCE No 2.11** LONGTAIL No 6.23*** PCPROP No 5.81*** LBHERF No 4.53*** GEOHERF No 2.21** AVGRTG No 0.85 CONVERTIBLE No 4.52*** SUBORDINATED No 3.01*** YTM No 2.61*** a If Equal Variances = yes, then Pooled test is used. Otherwise, Satterthwaite is used. *, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively. Table continues

11 INSURER OPACITY AND OWNERSHIP STRUCTURE 103 Table 1. (continued) Panel C. Life/health sample. Assets and capital in millions ($). Mutual Insurer Issues Variable N Minimum Mean Median Maximum Std Dev SPLIT ABSDIFF GROUP LIQUID CAPITAL ATOVER PRIVATE OPAQUE ASSET RESERVE PS RBCr REINSURANCE LONGTAILLH LBHERF GEOHERF AVGRTG CONVERTIBLE SUBORDINATED YTM Table continues business as a proportion of their overall business than do stock firms. 7,8 Stock firms are more concentrated from both a lines of business perspec 7 This result is almost certainly driven by the predominance of the mutual ownership form among life insurers. The difference in LONGTAIL between stock and mutual observations is not significant for property/casualty firms.

12 104 ADAMSON, ECKLES, AND HAGGARD Table 1. Panel C (continued) Stock Insurer Issues Variable N Minimum Mean Median Maximum Std Dev SPLIT ABSDIFF GROUP LIQUID CAPITAL ATOVER PRIVATE OPAQUE ASSET RESERVE PS RBCr REINSURANCE LONGTAILLH LBHERF GEOHERF AVGRTG CONVERTIBLE SUBORDINATED YTM Table continues tive (LBHERF) and a geographic perspective (GEOHERF), though these differences are not statistically significant for PC firms. The average rating on debt issues in our overall sample is not statistically significantly different between the two types of firms, though we do note that Moody s is 8 We use the definitions utilized in various studies (e.g., Phillips, Cummins, and Allen, 1998; and Zhang, Cox, and Van Ness, 2009).

13 INSURER OPACITY AND OWNERSHIP STRUCTURE 105 Table 1. (continued) Panel D. Life/health sample differences between mutual firms and stock firms. Assets and capital in millions ($). Comparison Variable Mutual Stock Equal Variances a t value SPLIT Yes 3.12*** ABSDIFF Yes 2.96*** LIQUID No 0.39 CAPITAL No 4.94*** ATOVER No 2.66*** PRIVATE No 3.08*** OPAQUE No 2.03** ASSET No 1.78* RESERVE Yes 6.00*** PS No 7.91*** RBCr Yes 0.76 REINSURANCE No 4.28*** LONGTAILLH No 5.39*** LBHERF No 4.67*** GEOHERF No 1.71* AVGRTG No 3.20*** CONVERTIBLE No 2.01** SUBORDINATED No 1.74* YTM No 0.09 a If Equal Variances = ye,s then Pooled test is used. Otherwise, Satterthwaite is used. *, **, and *** represent significance at the 10%, 5%, and 1% levels respectively. Table continues typically one half ratings notch harsher (0.44, t stat = 15.58) than their counterparts at S&P. Stock firms are engaged in significantly more property and casualty business as a proportion of their overall business than mutual firms. Regarding the debt issues themselves, no mutual firm issues con

14 106 ADAMSON, ECKLES, AND HAGGARD Table 1. (continued) Panel E. Property/casualty sample. Assets and capital in millions ($). Mutual Insurer Issues Variable N Minimum Mean Median Maximum Std Dev SPLIT ABSDIFF GROUP LIQUID CAPITAL ATOVER PRIVATE OPAQUE ASSET RESERVE PS RBCr REINSURANCE LONGTAILPC LBHERF GEOHERF AVGRTG CONVERTIBLE SUBORDINATED YTM Table continues vertible or subordinated debt, though stock firms do. The yield to maturity on mutual firm debt in the overall sample is approximately 55 basis points higher (t stat = 2.61) than the yield to maturity on bonds issued by stock firms. These economically and statistically significant differences between mutual firms and stock firms for measures which might be related to opacity necessitate a multivariate analysis to disentangle the sources of the

15 INSURER OPACITY AND OWNERSHIP STRUCTURE 107 Table 1. Panel E (continued) Stock Insurer Issues Variable N Minimum Mean Median Maximum Std Dev SPLIT ABSDIFF GROUP LIQUID CAPITAL ATOVER PRIVATE OPAQUE ASSET RESERVE PS RBCr REINSURANCE LONGTAILPC LBHERF GEOHERF AVGRTG CONVERTIBLE SUBORDINATED YTM Table continues opacity differences between mutual firms and stock firms. Given the significant differences between life/health companies and property/casualty companies discussed previously, we also examine these two subtypes of firms in separate multivariate analyses. Before proceeding to multivariate analysis, we perform a bivariate analysis to examine relations between variables of interest. Table 2 presents this correlation analysis for our combined sample and our two subsamples.

16 108 ADAMSON, ECKLES, AND HAGGARD Table 1. (continued) Panel F. Property/casualty sample differences between mutual firms and stock firms. Assets and capital in millions ($). Comparison Variable Mutual Stock Equal Variances a t value SPLIT Yes 2.00** ABSDIFF Yes 0.59 GROUP No 2.45** LIQUID No 1.02 CAPITAL No 4.27*** ATOVER No 1.86 PRIVATE Yes 0.39 OPAQUE No 6.38*** ASSET No 3.77*** RESERVE No 0.02 PS Yes 0.84 RBCr Yes 0.86 REINSURANCE No 0.66 LONGTAILPC Yes 0.61 LBHERF Yes 1.47 GEOHERF Yes 0.01 AVGRTG Yes 1.68* CONVERTIBLE No 2.69*** SUBORDINATED No 1.00 YTM Yes 3.09* a If Equal Variances = yes, then Pooled test is used. Otherwise, Satterthwaite is used. *, **, and *** represent significance at the 10%, 5%, and 1% levels, respectively. Given our hypothesis, we are primarily interested in STOCK (indicator variable equal to one if a firm is a stock firm), SPLIT (indicator variable equal to one if bond ratings differ), and ABSDIFF (the absolute difference of numeric ratings for bond issues). Consistent with our hypothesis that mutual firms are more opaque than stock firms, the STOCK and SPLIT

17 INSURER OPACITY AND OWNERSHIP STRUCTURE 109 indicators for the combined sample are significantly negatively correlated ( 0.107, p = ). Similarly, STOCK is significantly negatively correlated to ABSDIFF ( 0.089, p = 0.003) in the same sample. This pattern also holds for the property/casualty sample (though the ABSDIFF correlation loses significance), but not the life/health sample, which displays positive and significant correlations between STOCK and the two ratings disagreement variables. Moving our attention to SPLIT, we see the purely mechanical positive correlation with ABSDIFF, as ABSDIFF is always zero when SPLIT is zero, and always positive when SPLIT is positive. SPLIT is positively correlated to the proportion of opaque assets (OPAQUE) in all three samples (though not significantly so for life/health firms), supporting the suitability of SPLIT as a proxy for opacity. SPLIT is significantly negatively correlated to the size of the firm measured by total assets (ASSET) in all three samples, though the relation is not significant for the property/ casualty sample. Larger companies are likely more transparent to bond rating agencies simply due to their size and more frequent participation in the public debt markets. Correlations for ABSDIFF are largely similar to those for SPLIT, with some notable exceptions. The ratio of risk based capital to surplus (RBCr) is positively and significantly related to ABSDIFF (0.072, p = ) in the combined sample, but only marginally correlated with SPLIT (0.048, p = ). The subordination of a bond issue in the combined sample is significantly related to ABSDIFF (0.064, p = ), but unrelated to SPLIT. Multivariate analysis is required to disentangle the relations among these variables. Given the discrete nature of our opacity proxies, we use two types of probit analysis. For SPLIT, a dichotomous variable, we use a regular probit analysis to develop a regression model that predicts the probability of a debt issue receiving a split rating based on the ownership form of the firm (mutual versus stock), the characteristics of the firm s assets, the firm s business, and the debt issue. The results of this analysis are presented in Table 3. For ABSDIFF, which can take on integer values between 0 and 4, we perform an ordered probit, the results of which are presented in Table 4. We perform both sets of analyses on the combined sample, the life/health sample, and the property/casualty sample. For both sets of analyses, our first regression model contains only the STOCK indicator as a covariate. Then, we progressively add in sets of related variables as additional predictors. For the second regression model, we include additional predictors that describe the assets of the issuing firm. For the third regression model, we include predictors that describe the business of the issuing firm using variables based on the firm s premiums. For the fourth regression model, we include indicator variables which denote whether the debt issue is convertible and whether the debt issue is

18 110 ADAMSON, ECKLES, AND HAGGARD Table 2. STOCK SPLIT ABSDIFF GROUP LIQUID CAPITAL ATOVER PRIVATE OPAQUE ASSET Panel A. Combined sample. SPLIT ABSDIFF <.0001 GROUP < LIQUID < <.0001 CAPITAL <.0001 < <.0001 ATOVER < PRIVATE <.0001 < < OPAQUE <.0001 < < <.0001 ASSET <.0001 < <.0001 < <.0001 <.0001 RESERVE < < PS < < RBCr <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 REINSURANCE LONGTAIL <.0001 <.0001 < <.0001 <.0001 <.0001 <.0001 <.0001 <.0001 PCPROP <.0001 <.0001 < < <.0001 <.0001 <.0001 LBHERF <.0001 < <.0001 <.0001 <.0001 GEOHERF < < <.0001 AVGRTG <.0001 < < < <.0001 CONVERTIBLE SUBORDINATED YTM <.0001 < < <.0001 a This table presents correlations between variables of interest. Variables are defined in Appendix B.

19 INSURER OPACITY AND OWNERSHIP STRUCTURE 111 Correlations a RESERVE PS RBCr REINSUR ANCE LONG TAIL PCPROP LBHERF GEO HERF CONVER SUBORDIN AVGRTG TIBLE ATED < < < <.0001 < <.0001 < < < <.0001 < < < < <.0001 < < < < < <.0001 <

20 112 ADAMSON, ECKLES, AND HAGGARD Table 2. STOCK SPLIT ABSDIFF LIQUID CAPITAL ATOVER PRIVATE OPAQUE ASSET Panel B. Life/health sample. SPLIT ABSDIFF <.0001 LIQUID CAPITAL < ATOVER PRIVATE OPAQUE <.0001 ASSET <.0001 < RESERVE < < <.0001 < PS < <.0001 < <.0001 < RBCr REINSURANCE LONGTAILLH < <.0001 < LBHERF <.0001 < <.0001 <.0001 <.0001 GEOHERF < AVGRTG <.0001 < < <.0001 CONVERTIBLE < SUBORDINATED YTM subordinated. Finally, we include the yield to maturity of the debt issue. We introduce this variable in a regression model separate from the other debt issue characteristics because including it substantially reduces the

21 INSURER OPACITY AND OWNERSHIP STRUCTURE 113 (continued) RESERVE PS RBCr REINSUR ANCE LONG TAILLH LBHERF GEOHERF AVGRTG CON VERTIBLE SUB ORDINATED < <.0001 < < <.0001 < <.0001 < <.0001 <.0001 < < <.0001 < < <.0001 < < < sample size. Both sets of analyses use errors clustered by firm and fixed effects for years to account for systematic changes which affect all firms.

22 114 ADAMSON, ECKLES, AND HAGGARD Table 2. STOCK SPLIT ABSDIFF GROUP LIQUID CAPITAL ATOVER PRIVATE OPAQUE Panel C. Property/casualty sample. SPLIT ABSDIFF <.0001 GROUP < LIQUID <.0001 CAPITAL ATOVER < PRIVATE < OPAQUE < ASSET < RESERVE < PS < < < RBCr <.0001 < REINSURANCE LONGTAILPC <.0001 < LBHERF < < <.0001 GEOHERF < < AVGRTG <.0001 < <.0001 CONVERTIBLE SUBORDINATE D YTM

23 INSURER OPACITY AND OWNERSHIP STRUCTURE 115 (continued) ASSET RESERVE PS RBCr REINSUR ANCE LONG TAILPC LBHERF GEO HERF AVG RTG CONVERT IBLE SUBORDIN ATED < <.0001 <.0001 <.0001 < <.0001 <.0001 < < < <.0001 < <.0001 < < < <.0001 <

24 116 ADAMSON, ECKLES, AND HAGGARD Table 3. Probit Regressions a Panel A. Combined sample. (1) (2) (3) (4) (5) I.V. Coefficient p Coefficient p Coefficient p Coefficient p Coefficient p STOCK GROUP <.0001 CAPITAL LIQUID ATOVER PRIVATE OPAQUE ASSET RESERVE PS RBCr REINSURANCE LONGTAILC PCPROP LBHERF GEOHERF AVGRTG CONVERTIBLE SUBORDINATED YTM

25 INSURER OPACITY AND OWNERSHIP STRUCTURE 117 N R Max rescaled R Likelihood ratio < < < <.0001 Score < < < <.0001 Wald < < <.0001 Percent correct Int. only Int. & I.V. Int. only Int. & I.V. Int. only Int. & I.V. Int. only Int. & I.V. Int. only Int. & I.V. AIC SC a This table presents estimated coefficients for probit regressions with SPLIT as the dependent variable. Variables are defined in Appendix B.

26 118 ADAMSON, ECKLES, AND HAGGARD Table 3. (continued) Panel B. Life/health sample. (1) (2) (3) (4) (5) I.V. Coefficient p Coefficient p Coefficient p Coefficient p Coefficient p STOCK CAPITAL LIQUID ATOVER PRIVATE OPAQUE ASSET RESERVE PS RBCr REINSURANCE LONGTAILLH LBHERF GEOHERF AVGRTG CONVERTIBLE SUBORDINATED <.0001 YTM

27 INSURER OPACITY AND OWNERSHIP STRUCTURE 119 N R Max rescaled R Likelihood ratio Score Wald <.0001 Percent correct Int. only Int. & I.V. Int. only Int. & I.V. Int. only Int. & I.V. Int. only Int. & I.V. Int. only Int. & I.V. AIC SC

28 120 ADAMSON, ECKLES, AND HAGGARD Table 3. (continued) Panel C. Property/casualty sample. (1) (2) (3) (4) (5) I.V. Coefficient p Coefficient p Coefficient p Coefficient p Coefficient p STOCK GROUP CAPITAL LIQUID ATOVER PRIVATE OPAQUE ASSET RESERVE PS RBCr REINSURANCE LONGTAILPC LBHERF GEOHERF AVGRTG CONVERTIBLE SUBORDINATED < <.0001 YTM