Read WP04_9.pdf text version

DUPES OR INCOMPETENTS? AN EXAMINATION OF MANAGEMENT'S IMPACT ON FIRM DISTRESS J. Tyler Leverty + Martin F. Grace ++

August 2009 ABSTRACT This paper examines whether managers impact firm performance when it is arguably needed most--during times of distress. We conservatively define managerial ability as the manager's capacity to deploy the firm's resources. We verify the validity of our metric using a manager-firm matched panel data set which allows us to track managers (CEOs) across different firms over time. Manager fixed effects are highly related to our measure of managerial ability. We then document that the ability of managers inversely influences the amount of time a firm spends in distress, the likelihood of a firm's failure, and the cost of failure. These results overturn prior findings and suggest that managers of failed firms are less skilled than their counterparts. Keywords: Managerial Ability, Manager Fixed Effects, Financial Distress JEL Classification Numbers: M1, D2, G33

+ Corresponding Author: Assistant Professor, Department of Finance, Henry B. Tippie College of Business, University of Iowa, Iowa City, Iowa 52242-1994; Phone: 319-335-0963; E-Mail: [email protected] ++ James S. Kemper Professor, Department of Risk Management & Insurance, Georgia State University, PO Box 4035, Atlanta, Georgia 30302-4035; E-Mail: [email protected]

DUPES OR INCOMPETENTS? AN EXAMINATION OF MANAGEMENT'S IMPACT ON FIRM DISTRESS

ABSTRACT This paper examines whether managers impact firm performance when it is arguably needed most--during times of distress. We conservatively define managerial ability as the manager's capacity to deploy the firm's resources. We verify the validity of our metric using a manager-firm matched panel data set which allows us to track managers (CEOs) across different firms over time. Manager fixed effects are highly related to our measure of managerial ability. We then document that the ability of managers inversely influences the amount of time a firm spends in distress, the likelihood of a firm's failure, and the cost of failure. These results overturn prior findings and suggest that managers of failed firms are less skilled than their counterparts. Keywords: Managerial Ability, Manager Fixed Effects, Financial Distress JEL Classification Numbers: M1, D2, G33

I. INTRODUCTION The failure of a firm is often blamed on management. Managers of failed firms are viewed as possessing poor judgment and being less skilled (Cannella, Fraser, and Lee [1995]). Managers, though, rarely take responsibility for their firm's failure (see, for example, the Congressional testimony of Richard S. Fuld Jr. [2008], chief executive officer (CEO) of Lehman Brothers, or Martin Sullivan [2008], CEO of AIG).1 Interestingly, the academic literature largely supports the managers' opinions. Lang and Stulz [1992], John, Lang, and Netter [1992], and Khanna and Poulsen [1995] find that managers of failed firms are not less skilled than their contemporaries and they thus serve as scapegoats for their firm's failure. The belief that managers do not affect the performance of failed firms largely contrasts the views expressed in the business press, the pedagogy (and existence) of business schools, the compensation of top managers (Gabaix and Landier [2008]), and a large body of research examining the relationship between CEO changes and firm performance (see for example, Warner, Watts and Wruck [1989], Weisbach [1995], Perez-Gonzales [2006], or Bennedsen, et al. [2007]). The objective of this paper is to revisit managements' influence on firm financial distress. The bulk of existing studies implicitly assume all managers behave similarly. They rely on firm-, industry-, or market-level characteristics and not the characteristics or ability of managers. In this study we are interested in managerial ability and how it relates to firm distress. Our definition of ability is based on an actual economic outcome--the manager's efficient use of the firm's inputs and outputs in the production process (Farrell [1957]). We examine whether the CEO is able to minimize firm costs, maximize revenues, operate at an efficient scale, adopt the best practice technology, and use inputs according to their marginal productivities. We compute a firm's

1

Richard Fuld Jr. blamed many factors for the demise of Lehman Brothers: a crisis of confidence, naked short selling, and the "changed landscape of our financial system and regulatory regime". Martin Sullivan indicated the collapse of AIG was largely due to marked to market accounting rules.

1

efficiency in producing outputs compared to a hypothetical `best practice' firm having the same inputs. All deviations from the `best practice' firm, including the influence of agency problems and managerial incompetence, are measured as inefficiency.2 To verify the validity of our definition of managerial skill, we adopt the approach of a growing literature in finance and economics and follow chief executive officers (CEOs) across different firms over time to examine the influence of individual CEOs.3 This allows us to estimate manager fixed effects to determine the extent that firm efficiency is linked to managers, above and beyond firm, group, and year effects (Bertrand and Schoar [2003]). Testing for the contribution of managers on firms' outcomes is challenging as firm performance depends on the actions of the CEO as well as on random factors. In particular, firm performance will rise or fall with industry fortunes (Bertrand and Mullainathan [2001]). Since large movements in industry profits are likely to be beyond the control of a single CEO, it will be difficult to isolate the performance of individual CEOs from the concurrent performance of the industry in which they operate. We, therefore, focus on a single industry ­ property-liability insurance. By focusing on a single industry we are better able to measure firm growth opportunities and acknowledge differences in access to external capital markets as well as agency costs associated with various organizational forms, distribution channels, and products. There are many advantages to using the property-liability insurance industry to investigate management's impact on financial distress.4 First, property-liability insurers are in the business

2

There is a large management literature that studies managerial characteristics (see, for example, Hambrick and Mason [1984], Waldman, Ramirez, House, and Puranam [2001], or Agle, Nagarajan, Srinivasan, and Sonnenfeld [2006]). Our definition of managerial ability does not explicitly consider managerial charisma or leadership; however, it does account for when these characteristics allow managers to more effectively use the firm's resources. 3 For example, Bertrand and Schoar [2003], Malmendier and Tate [2005], Pérez-González [2006], Bennedsen, Nielsen, Pérez-González and Wolfenzon [2007]. Bennedsen, Pérez-González, and Wolfenzon [2007], or Billett and Qian [2008]. 4 We do not want readers to be bogged down by the institutional aspects of this industry. All the control variables we use in the study are common in the insurance literature and are described fully in the Data Appendix.

2

of taking risk and distress is relatively frequent and severe (Bohn and Hall [1999], Hall [2000], Grace, Klein, Phillips [2009]).5 Second, even though insurers, along with utilities and banks, are often excluded from studies because they are regulated, the regulators' close monitoring of these firms' financial condition provides a weaker definition of distress ­ regulatory scrutiny ­ than is used in prior studies.6 Third, a unique data source allows us to estimate of the total cost of resolving company failures (Hall [2000], Grace, Klein, Phillips [2009]). Fourth, the relatively large amount of organizational structure heterogeneity in the industry provides easier decomposition of the manager fixed effects. Fifth, there is an established measure of firm efficiency in the property-liability literature (Cummins and Weiss [2001]) that is strongly linked to traditional and market measures of performance (Cummins, et al. [2008], and Leverty and Grace [2009]). Sixth, we do not have to restrict our study to publicly traded firms as data on all insurance companies, public and private, is readily available from the National Association of Insurance Commissioners (NAIC) Annual Statement database. Finally, we have the names of all the CEO's heading these companies and are able to track them over time and across companies. We seek to determine whether CEOs can influence firm performance when it is arguably most needed and likely to matter--during times of financial distress. First, we examine whether high quality managers are able to remove their firms from regulatory scrutiny sooner than lower quality managers. Second, we explore whether management quality significantly reduces the

For our sample, the average cost of insolvency is roughly $1.02 for every dollar of pre-insolvency assets. To put the cost of resolving property-liability insurer insolvencies in perspective, the cost to resolve bank insolvencies averaged around 30 percent for a sample of banks that failed during 1985-1989 (James, 1991) and 20 percent for banks that failed in the late 1990's (Kaufmann [2001]). 6 Strict definitions of financial distress are the norm in previous studies. For example, Andrade and Kaplan [1998] define a firm to be in financial distress if one of the following four events occurs: (1) EBDITA/Interest Expense is less than one; (2) the company announced it attempted debt restructuring due to its inability to meet debt obligations; (3) the company defaulted on debt; or (4) the company filed for Chapter 11 proceedings. In contrast, our definition of distress is based on an ex ante likelihood of failure as opposed to an ex post measure. Prior to formal regulatory proceedings against a distressed insurer, regulators put the firm under regulatory scrutiny where it is subject to closer inspection and perhaps a formal examination.

5

3

likelihood that a firm becomes insolvent. Finally, we examine whether managerial ability influences the cost of insolvency. This paper makes a unique contribution to a growing literature on the impact of CEOs on the firm. Our results show that manager fixed effects are an empirically and economically important determinant of firm efficiency. Moreover, the proportion of CEO fixed effects that is statistically related to firm efficiency is always greater than the proportion of group fixed effects and frequently greater than the proportion of firm fixed effects.7 These findings provide evidence that managers play an economically and statistically important role in firms. Further, by correlating our estimated CEO fixed effects across different measures of economic efficiency, we identify some prevailing patterns in managerial ability. Most importantly we find evidence that ability varies across CEOs. In contrast to previous research (Lang and Stulz [1992], John, Lang, and Netter [1992], and Khanna and Poulsen [1995]), we find that mismanagement negatively impacts firm performance during times of distress. Our results indicate that superior managers are able to remove their firms from regulatory scrutiny sooner than relatively inferior managers. We also find that more efficient managers reduce the likelihood that their firms become insolvent. In addition, managerial quality decreases the cost of failure. Overall, our results cast doubt of the belief that the managers of failed firms are not less skilled than their contemporaries. The remainder of the paper is organized as follows. Section II presents the data sources. Section III details our measure of managerial ability. Sections IV, V, and VI examine the impact of managerial ability on the duration of regulatory scrutiny, the likelihood of insolvency, and the cost of insolvency, respectively. Section VII concludes.

7

Many insurance firms are organized as a group of firms under common ownership. State Farm, for example, has seventeen separately capitalized companies within its group. We measure efficiency at the firm level but account for common ownership with a group indicator.

4

II. DATA DESCRIPTION The variables used in this paper are extracted from multiple sources. The majority of the data comes from the 1989-2000 National Association of Insurance Commissioners (NAIC) Property-Casualty Annual Statement Database. The NAIC database contains the yearly regulatory filings of approximately 2,300 property-liability insurance companies domiciled in the United States. The demographics page of the regulatory annual statement provides information on the identity of the CEO from 1992 to 1999. We cross-reference this data with the CEOs listed in Best's Insurance Reports. The remainder of the annual statement contains financial data and data on premiums, losses, and expenses categorized by state and by 26 insurance lines. The NAIC annual statements are the primary source of data for the construction of the managerial ability metrics. In addition to the regulatory annual statements we also obtain input price data from the U.S. Bureau of Labor Statistics and Ibbotson Associates. Information on the insurance firm's ownership structure (stock or mutual) and group membership is extracted from the A. M. Best Company's Key Rating Guide. We categorize firms according to the ownership structure of their ultimate owner (Mayers and Smith [1994]). Even though many individual insurers are organized into groups, we examine insurers at the individual insurer level as it is less subject to bias and noise from mergers and acquisitions than the analysis of groups.8 It also allows us to track CEO changes across affiliated individual insurance companies, which provides additional heterogeneity in our subsample of manager-firm switches. The list of insolvent insurers, defined as formal regulatory action taken against the firm in the form of proceedings for conservation of assets, rehabilitation, receivership, or liquidation, comes from the NAIC's Report on Receiverships (various years), the Status of Single-State and

8

The majority of mergers and acquisitions in the property-liability insurance industry involve insurance groups buying and selling individual insurance companies. The NAIC annual statements are compiled at the individual company level regardless of the company's group affiliations.

5

Multi-State Insolvencies (various years), and the A.M. Best Company's list of all propertyliability insurers that failed from 1969-2001 (A.M. Best [2004]). The cost of liquidating an insurer comes from the Assessment and Financial Information Report published by the National Conference of Insurance Guaranty Funds (NCIGF [2003]). The NCIGF report records the cumulative payments, recoveries, and net cost through 2003 for each insolvency that triggered a guaranty fund assessment since 1969. Information on the definition and construction of all variables used in the paper is available in the Data Appendix.

III. MANAGERIAL ABILITY The purpose of this section is to provide a definition of managerial ability and to document its effectiveness. We provide the background for the construction of the ability metric and its relationship to other performance measures. The eventual goal is to then see whether firms with better managers reduce their time in financial distress. III.A. Definition Our definition of managerial ability focuses on the ability of the manager to marshal the firm's resources efficiently. Ideally two firms with similar characteristics and opportunity sets should have the same level of production, Y*. However, in reality some firms will not be as successful as others because of agency problems and differences in the ability of managers. As a result a firm may be at a production level Y, which is less than Y*. Firm inefficiency is the difference between Y* and Y. To measure efficiency as a firm's deviation from Y*, we need a credible benchmark of Y*. To avoid an unequal comparison of companies with different opportunities and characteristics, the benchmark needs to hold constant the firm's opportunity set and characteristics. Traditional measures of efficiency (for example, return on assets) are constrained to a single input and output

6

and therefore are unable to control for differences among firms in input and output mix. Frontier efficiency methods, in contrast, provide a mechanism to benchmark Y* and control for differences in input usage and output production in multi-input, multi-output firms using a rigorous approach derived from micro-economic theory (Aigner, Lovell, and Schmidt [1977] and Charnes, Cooper, and Rhodes [1978]). Frontier efficiency methods form a "best practice" frontier function for each firm that provides the maximum output for any given combination of inputs. This frontier function serves as the benchmark hypothetical value Y* that a firm could obtain if it were to match the production performance of its best-performing peer(s). A firm's shortfall from the frontier is a measure of inefficiency. We focus on cost, revenue, scale, technical, and allocative efficiencies. Cost efficiency is the ratio of the minimum required costs to the actual costs utilized to produce a given level of output. A firm is considered fully efficient if its actual input usage equals optimal input usage for given output quantities and input prices. A firm is inefficient if actual input usage exceeds optimal input usage. Revenue efficiency is the ratio of the revenues of a given firm to the revenues of a fully efficient firm with the same input vector and output prices. It is important to estimate both cost and revenue efficiency, since the objective of the firm is profit maximization. Thus to be completely efficient the firm must be both cost efficient and revenue efficient. Cost efficiency is composed of allocative efficiency and technical efficiency. Allocative inefficiency results from a firm's use of a suboptimal combination of inputs in producing a given level of output. Technical inefficiency results from not operating with the best-practice technology, i.e. the use of excessive resources to produce a given output. Technical efficiency can be further decomposed into pure technical efficiency and scale efficiency. Pure technical efficiency is the proportion by which the firm could reduce its input usage by adopting the best

7

technology represented by the variable returns to scale frontier. A firm operating on the variable returns to scale frontier is scale inefficient because it is not operating on the constant returns to scale frontier. Hence scale efficiency is measured as the ratio from the constant returns to scale frontier to the variable returns to scale frontier. We estimate frontier efficiency using a mathematical programming approach, Data Envelopment Analysis (DEA).9 DEA has been widely used to measure the efficiency of property-liability insurers (see Cummins and Weiss [2001]).10 Although DEA was traditionally viewed as a strictly non-parametric methodology, research has shown that it can be interpreted as a maximum likelihood procedure (Banker [1993]). In addition, the DEA estimator is consistent and converges faster than other estimators (Grosskopf [1996]). As such, the asymptotic distribution of the DEA estimators is identical to the true distribution of the efficiency. DEA efficiency estimates, however, are biased upward in finite samples (Simar and Wilson [1998]). To correct the upward bias of our efficiency estimates, we implement the bootstrapping procedure of Simar and Wilson [2000] with 1000 bootstrap replications. We estimate efficient production, cost, and revenue frontiers giving measures of technical, allocative, cost, and revenue efficiency for each firm in each year of our sample.11 Given a certain level of inputs and outputs, DEA compares each firm to its `best practice' peers and

There are two methods for estimating frontier functions ­ a regression approach and a mathematical programming approach. The regression approach assumes a production function and measures efficiency based on both random and firm-specific (in)efficiency components. It requires assumptions to be made for the production function (for example, Cobb-Douglas or translog), the distribution of the random error component, as well as the distribution of the firm-specific inefficiency component. Some recent research relies upon translog (Maksimovic and Phillips [2001]) or log-linear Cobb-Douglas (Schoar [2002]) production functions. The regression model, however, is subject to specification error unless the components required for the analysis are precisely known. Data Envelopment Analysis, in contrast, does not require any assumptions regarding the production function or error term distribution and is thus less susceptible to specification errors. Banker and Natarajan [2008] show that DEA-based procedures generally outperform regression methods since it is often the case that no a priori knowledge exists about the form of the production function or the distributions of the error and efficiency components. 10 For parsimony, DEA is not discussed in detail. A description of DEA is provided in Cooper, et al. [2000]. 11 An advantage of DEA is that it provides a convenient way to decompose cost efficiency into its pure technical, scale, and allocative components. Separate estimates of the sources of inefficiency are important for determining where managers are weak and strong.

9

8

provides an efficiency score from zero to one. A firm is classified as fully efficient (efficiency of 1.0) if it lies on the frontier and inefficient (0 < efficiency < 1) if its outputs can be produced more efficiently by another set of firms. In accordance with the majority of the recent literature on financial institutions, we adopt a modified version of the value-added approach to identify insurer outputs (Berger and Humphrey [1992] and Cummins and Weiss [2001]). Leverty and Grace [2009] examine other approaches to measuring insurance output and find that the value-added approach is the most consistent with the economic realities of the insurance market. The value-added approach employs as important outputs all categories that have substantial value-added, as judged by operating cost allocations (Berger and Humphrey [1992]). Operating cost allocations identify three principal services provided by P/L insurers: risk-pooling and risk-bearing, "real" insurance services, and financial intermediation (Cummins and Weiss [2001]). The proxy for the quantity of risk-pooling and real insurance services for property-liability insurers is the present value of real losses incurred, which are the losses that are expected to be paid as a result of providing insurance coverage. Since the risks and types of services provided differ between the main types of insurance, we separate lines of insurance with similar characteristics into categories: personal lines short-tail losses, personal lines long-tail losses, commercial lines short-tail losses, and commercial lines long-tail losses.12 Output prices are defined as the difference of premiums earned and the present value of losses incurred divided by the present value of losses incurred. This is the markup of prices over expected losses. The value-added approach captures the quantity of intermediation output using the average real

The line of business definitions are described in Phillips, Cummins, and Allen [1998]. The tail length refers to the length of the loss cash flow stream. Long tail lines, like liability lines, take longer for losses to be fully realized.

12

9

invested assets of a firm (Cummins and Weiss [2001]). The price of the intermediation output is measured by the expected rate of return on the insurer's assets. The inputs of the firm are classified into five categories: administrative labor, agent labor, business services and materials (including physical capital), financial equity capital and policyholder-supplied debt capital.13 The quantity of an input is defined as the current dollar expenditures associated with the particular input from the regulatory annual statement divided by its current price, which we obtain from the U.S. Department of Labor and the Bureau of Labor Statistics. The construction of the inputs and outputs is the standard in the literature. Additional details are available in the Data Appendix. Table I shows the means and standard deviations of these variables. III.B. Validity of our Managerial Ability Definition Empirical research documents a strong relationship between property-liability insurer efficiency and traditional and market measures of performance. For example, Cummins, et al. [2008] find that efficiency measures are directly related to the market value performance of publicly traded insurers. In turn, Leverty and Grace [2009] find that efficiency measures are closely related to traditional measures of firm performance, such as return on assets, return on equity, and the expense ratio (the sum of loss adjustment expenses, other underwriting expenses, and investment expenses to net premiums written).14 Our sample is not different. Pure technical, scale, allocative, and revenue efficiency have a Pearson (Spearman) correlation with return on assets of 0.264 (0.299), 0.457 (0.465), 0.420 (0.439), and 0.295 (0.310), and the correlation with the expense ratio is -0.129 (-0.221), -0.122 (-0.203), -0.261 (-0.254), and -0.101 (-0.154), respectively. Thus, our measures of managerial ability are correlated with traditional

See Cummins and Weiss [2001] for a comprehensive explanation of the inputs used in the value-added approach. The expense to premium ratio is a common metric of insurer efficiency (see for example, Joskow [1973], Cummins and VanDerhei [1979]).

14 13

10

performance metrics, but they also appear to be sufficiently different from conventional metrics to demonstrate the potential added value of using a multi-dimensional measure that automatically controls for differences among firms in input and output mix. To further validate our measure of managerial ability, we determine whether there are systematic differences in the efficiency of top managers, controlling for relevant observable firm- and group-level characteristics. Since there may be lasting differences in practices across firms and groups due to some unobservable factors (which may be correlated with the manager fixed effects), we separate manager fixed effects from firm and group fixed effects. Specifically, we construct a manager-firm matched panel data set that allows us to track the same top managers across different firms from 1992 to 1999. While many individual insurers are subsidiaries or belong to a group, we examine CEOs at the individual insurer level.15 This provides more heterogeneity in our subsample of manager-firm switches. Similar to Bertrand and Schoar [2003] we confine our analysis to the subset for which at least one CEO can be observed in at least one other insurer.16 We keep all firm observations satisfying this condition. The subsample includes over 800 individual insurance firms and over 330 individual CEOs who are identified in a least two different firms.17 Table II displays means and standard deviations for efficiency and other corporate variables of interest. Columns two and three report the mean and standard deviation for the manager-firm matched sample (i.e., the sample in which at least one manager is observed in

Each group of affiliated insurers utilizes a different organizational structure. In some groups, the CEO of the group is also the acting CEO of each individual affiliated insurer. In other groups, separate CEOs are appointed at the individual insurer level. 16 Bertrand and Schoar [2003] require that the manager be in each firm for at least three years so that the manager has a real impact on a given company. We do not have the same restriction. Ignoring the three-year requirement makes finding statistically strong results more difficult. Thus, to the extent we find results, they are statistically robust. 17 The ratio of CEOs to insurance firms may seem high; however, we capture CEO transitions within groups of affiliated insurance companies.

15

11

multiple firms from 1992 to 1999). For comparison the equivalent summary statistics for the full sample of firms are shown in columns four and five. Limiting the sample to firms where at least one CEO switch is observed yields larger firms and a higher proportion of stock companies. The average firm in the manager-firm matched sample has greater pure technical efficiency and lower scale, allocative, cost and revenue efficiency. For the full sample of firms, average cost efficiency is 35.2 percent, suggesting that, on average, the industry could have reduced costs by 64.8 percent if all managers operated on the production frontier and chose the cost minimizing input bundles. Decomposing cost efficiency into its components, we find that allocative inefficiency is the primary source of cost inefficiency. Average allocative efficiency is 59.1 percent, while pure technical efficiency and scale efficiency average 66.7 and 89.8 percent, respectively. Average revenue efficiency is 23.3 percent. The industry, on average, could have increased revenues by 76.7 percent if managers operated on the production frontier and picked the revenue maximizing output bundles. To determine whether individual managers influence firm performance, we investigate how much of the variance in firm performance can be attributed to CEO-specific effects. For each efficiency variable, we estimate the following regression: (1)

yit = t + i + g + X it + CEO + it ,

where yit is the natural logarithm of one of the efficiency variables,18 t are year fixed effects, i are firm fixed effects, g are group fixed effects, X it is a vector of time-varying firm characteristics, CEO are CEO fixed effects, and it is an error term.19 When estimating the equation, we account for serial correlation by allowing for clustering of the error term at the firm

Since the universe of firms determines the production, cost, and revenue frontiers, we use the efficiency variables calculated with the full set of firms. 19 Banker and Natarajan [2008] determine that OLS is an appropriate and robust technique to evaluate the impact of external variables on efficiency.

18

12

level. We estimate the equation for only those managers who switch companies. This is a conservative approach that mandates efficiency be correlated across two (or more) firms when the same CEO is present. This allows us to assess whether there is any evidence that firm efficiency systematically changes with the identity of the CEOs in these firms. Because the manager-firm matched sample is observed to be significantly different from the full sample (Table II), we use a standard Heckman two-stage regression methodology to control for potential sample selection bias. Table III reports the proportion of fixed effects which are statistically significant at the 10 percent level, the F-tests, and the adjusted R2s from the estimation of equation (1) for the different efficiency variables. For each efficiency variable, we report the results of four regressions. The first includes firm and year fixed effects. This model serves as the benchmark specification that allows us to document the change in adjusted R2 when we add the CEO fixed effects. The second contains CEO and year fixed effects. The third consists of firm, CEO and year fixed effects. The final regression contains firm, group, CEO, and year fixed effects. All regressions include the natural logarithm of total assets and a mutual indicator variable for those companies which are owned by policyholders rather than shareholders. Overall, the results in Table III indicate that CEO-specific effects are important both economically and statistically for firm efficiency. The F-tests on the CEO fixed effects are large and always reject the null hypothesis that all the manager fixed effects are zero. Furthermore, adding CEOs to the benchmark specifications (row three for each dependent variable) increases the adjusted R2s. The magnitude of the increase is not large but it is important given that the fit of the benchmark models is already quite high. Moreover, as we describe in detail below, analysis of variance shows that manager fixed effects explain more variation in efficiency than

13

firm fixed effects. In addition, a large proportion of CEO fixed effects are statistically significant. In many of the regressions that include firm and CEO fixed effects (row three), the proportion of significant CEO fixed effects is greater than those for the firm fixed effects. While our efficiency measures cannot be fully assigned to managers, we find that managers do appear to influence firm efficiency. It, therefore, seems that the efficiency variables are measuring, at least to some extent, the underlying concept of managerial ability.20 To further explore the relative economic importance of managerial differences on firm efficiency, we use analysis of variance. We first obtain partial sum of squares (also called Type III sum of squares) for each factor (firm, group, and manager fixed effects). We then divide the partial sum for each factor by the total Type III partial sum of squares over all the factors for a particular model. This normalization procedure allows us to understand the relative importance of each factor in determining efficiency. When firm, group, and manager fixed effects are included in the regressions, the normalized partial sum of squares for manager fixed effects is 50.9, 44.4, 53.5, 57.0, and 52.5 percent for pure technical, scale, allocative, cost, and revenue efficiency, respectively. For firm fixed effects it is 43.0, 51.9, 41.5, 37.2, and 42.1. Thus, for four of the five dependent variables, manager fixed effects explain more variation in efficiency than firm fixed effects. We also evaluate the magnitude of the observed differences between managers. Table IV reports the median, standard deviation, twenty-fifth percentile, and seventy-fifth percentile of the fixed effects for each of the regressions estimated in row 5 of Table III. The table allows us to observe the efficiency of a CEO in the top quartile of the efficiency fixed effects distribution

We also perform the regressions using the rate of return on assets as the dependent variable. The adjusted R2 of the benchmark specification was 0.5623, which is much lower than those for the efficiency variables. In the regression that includes the full set of fixed effects, 69 percent of the CEO fixed effects are significant relative to 48 percent for each of the firm and group fixed effects.

20

14

relative to a CEO in the bottom quartile. Each of the fixed effects is weighted by the inverse of its standard error to account for estimation error (Bertrand and Schoar [2003]). The results show that the variation in the magnitude of the CEO fixed effects is economically large. The difference between a CEO at the twenty-fifth percentile of the distribution of cost efficiency and one at the seventy-fifth percentile is 2.07. Some managers are also better than others at maximizing firm revenue. The difference between a top quartile CEO and a bottom quartile CEO for revenue efficiency is 0.89. In sum, there is a wide degree of heterogeneity in the ability of managers and the efficiency of a firm may improve or worsen with a different CEO.21 We next investigate whether there is a relationship between the various types of efficiency. Is a manager who is skilled at maximizing firm revenue also good at minimizing costs? Is a CEO adept at implementing the best practice technology also capable of operating at the correct scale? To determine the answer to these types of questions, we examine the correlation between the efficiency CEO effects. We take each of the separate efficiency CEO fixed effects and estimate the following regressions: (2) F .E.( j )i = + F .E.( k )i + i for j k ,

where i indexes CEOs and j and k indicate a type of efficiency fixed effect, F .E.( ) . The right-

hand-side variable is an estimated coefficient and will thus lead to a downward bias in an OLS estimation of . To ameliorate this problem, we account for the measurement error in the independent variable by using GLS to estimate equation (2). Specifically, we weigh each observation by the inverse of the standard error of the right-hand-side variable, which we obtain from the regressions estimated in row 5 of Table III (Bertrand and Schoar [2003]).

21

We also examine the distribution of CEO fixed-effects for the rate of return on assets. The difference between a top quartile CEO and a bottom quartile CEO for return on assets is 1.158. The average rate of return on assets for the manager-firm matched sample is 0.059.

15

Table V reports the results of these regressions. The average R2 of these regressions is approximately 33.3 percent. The R2 ranges from a low of 10.2 percent to a high of 76.6 percent. One broad pattern in the ability of managers surfaces from these results--there are good managers and there are bad managers. In particular, managers skilled at adopting the best practice technology are also adept at operating at the efficient scale, using an optimal combination of inputs, and maximizing revenue.22 The results are interesting in that they show that there are systematic differences in the skills of managers.23

IV. Duration of Regulatory Scrutiny

The previous section documents the validity of efficiency as a measure of managerial ability. Now we want to use this metric to determine whether managers influence firm performance during financial distress. Prior to any formal regulatory proceedings against a distressed insurer, the firm is put under regulatory scrutiny, where the insurer is subject to closer inspection and perhaps a formal examination. Many companies successfully remove themselves from regulatory oversight and avoid formal regulatory proceedings. The question we seek to address is whether managerial quality influences a firm's ability to extricate itself from regulatory scrutiny, and if so, to what extent? Before we can answer this question, we must first formulate an operational definition of regulatory scrutiny. While regulators do not reveal which insurers are subject to regulatory review, they do reveal the insurers subject to formal regulatory intervention. We also know the

The CEO fixed effects for return on assets also exhibit a direct relationship with the efficiency measures. We also examine the pair-wise correlations of the manager-fixed effects. The findings are similar. In addition, we conduct a factor analysis for all of the fixed effects. Using a varimax rotation we identify only one eigenvector with an eigenvalue greater than one. We also investigate the correspondence of manager fixed-effects across the different measures of performance. Specifically, we determine the proportion of manager fixed-effects identified by one efficiency measure as being in the top (bottom) quartile that were also identified in the top (bottom) quartile by another performance measure. The results confirm a direct relationship between the efficiency fixed effects.

23

22

16

basis of the analytical tools state insurance regulators use to monitor insurer financial condition. Using this information and the assumption that regulators cost-effectively allocate their limited resources, we can identify the insurers that are most likely to be subject to regulatory scrutiny. We estimate each firm's probability of insolvency using the NAIC's solvency screening mechanisms and the identity of firms subject to formal regulatory intervention. The NAIC uses two solvency screening mechanisms, the Insurance Regulatory Information System (IRIS) and the Financial Analysis and Surveillance Tracking (FAST) system.24 The objective of IRIS and FAST is to select those companies that merit the highest priority in the allocation of the regulators' resources. Each consists of a set of ratios that focuses on an insurer's leverage, liquidity, operations, profitability, and asset quality.25 In addition to IRIS and FAST ratios, we also control for insurer organizational structure (a mutual firm indicator variable) and size (the natural logarithm of total assets). We define a firm as being under regulatory scrutiny if it has a probability of insolvency that is greater than a "probability cutoff point" which recognizes the cost-effective allocation of regulators limited resources. The cost-effective allocation is based on the relative cost of misclassifying a failing firm (Type I error) and misclassifying a solvent firm (Type II error). The cost of misclassifying a failing firm is the total guarantee fund assessment due to a firm's failure.26 The cost of misclassifying a solvent firm is the opportunity cost of the regulators' formal examination of the firm. We use a 40:1 relative cost ratio, which is roughly the ratio of total insurer payments to New York's guaranty fund to total funds reimbursed to the New York

The NAIC also implemented risk based capital (RBC) in 1994. We do not discuss the RBC system in this paper because there is ample research showing its inability to predict insolvency (see for example, Grace, Harrington, and Klein [1998]). 25 IRIS consists of twelve financial ratios. Grace, et al. [1998] list 25 FAST ratios. There is no guarantee the NAIC still uses these same FAST ratios. Ratios that are common to both the IRIS and FAST systems are used only once. 26 The state's guarantee fund protects policyholders against an insurer's insolvency much like the FDIC protects depositors from bank failures. Each state establishes a guaranty association to cover an insolvent insurer's financial obligations, within statutory limits, to the insurer's policyholders and claimants in the state.

24

17

Department of Insurance for its examinations of insurers.27 New York is generally considered to have the most rigorous regulatory system and it is the only state that requires ex-ante guarantee fund assessments (Meier [1988] and Cummins and Sommer [1996]). Table VI shows the descriptive statistics for the firms under regulatory scrutiny. The 40:1 relative cost ratio provides a time-varying estimate of the firms under regulatory scrutiny. The mean probability cut-off point is 2.6 percent. The mean probability of insolvency for firms under scrutiny is 7.1 percent, while it is 0.9 percent for those not under scrutiny. Firms under scrutiny are significantly smaller, less geographically and operationally diversified, more likely to be stock companies, and have higher leverage compared to non-scrutiny firms (p<0.01). To examine how managerial ability influences an insurer's duration under regulatory scrutiny, we estimate a hazard model. The hazard model captures a firm each time it enters regulatory scrutiny during the sample period.28 A firm enters regulatory scrutiny when it has a probability of insolvency above the probability cut-off point and exits when it is below the cutoff point. The mean (median) duration of regulatory scrutiny is 1.373 (1.000). The maximum is seven years. As firms have a non-zero probability of exiting regulatory scrutiny after our sample ends in 2000, we adjust the likelihood function for right censoring. The hazard function is estimated using three distributional assumptions: lognormal, log-logistic, and Weibull.29 The loglogistic distribution provides the best fit.30 For parsimony, only the results of the log-logistic

Source: Annual Report of the Superintendent of the Insurance to the New York Legislature, various years. We also perform the analysis on a sample that captures only the last time a firm enters regulatory scrutiny during the sample period. The results are qualitatively equivalent to those reported. 29 We also estimate the model using two other techniques. The first is an OLS regression in which the dependent variable is the natural logarithm of duration under scrutiny. The second is a Cox proportional hazard model. The results of these regressions are qualitatively similar to those reported. 30 The Akaike Information Criterion (AIC) for the fully specified model with the components of cost efficiency is 1164.3, 1299.8, and 1842.1 using the log-logistic, lognormal, and Weibull distributions, respectively. With revenue efficiency, the AIC is 1164.2, 1302.0, and 1855.8 for the log-logistic, lognormal, and Weibull.

28

27

18

regression model are reported; however, the results are robust to all three distributional assumptions. Table VII presents the results. A positive (negative) coefficient signifies that a covariate increases (decreases) the length of regulatory scrutiny. We find that firms with better managers spend less time under regulatory scrutiny. Managers that implement a "better-practice" production technology are able to accelerate their firm's removal by 8 to 11 percent (p<0.05). Managers capable of operating at the correct scale decrease their firm's duration by 11 to 16 percent (p<0.10). Further, revenue efficient firms spend approximately 8 percent less time under regulatory scrutiny (p<0.01). In general, managerial quality influences the firm's ability to remove itself from the regulatory radar screen. We also regress the duration under regulatory scrutiny on the CEO fixed effects of each efficiency variable. We weigh each observation by the inverse of the standard error of the righthand-side variable, which we obtain from the regressions estimated in row 5 of Table III (Bertrand and Schoar [2003]). The results [not shown] reveal a significant negative relationship between managerial ability and the length of time a firm spends under regulatory watch. Replacing the worst CEO in terms of pure technical efficiency with the median CEO reduces the duration under regulatory scrutiny by more than 8 percent. Substituting a CEO at the twenty-fifth percentile of the distribution of revenue efficiency with one at the seventy-fifth percentile decreases the time in regulatory scrutiny by over 4 percent. The CEO fixed effect sample further indicates that managerial skill plays a role in removing a firm from regulatory scrutiny. Table VIII displays firm characteristics at entrance into and exit from regulatory scrutiny. Firms emerge from scrutiny more efficient than they were prior to scrutiny. Pure technical, cost,

19

and revenue efficiency all increase from entrance to exit. Firms also exit significantly larger, with a greater proportion of assets in liquid investments, and with lower leverage.

V. Likelihood of Insolvency

An alternative way at looking how management deals with distress is to investigate whether more efficient managers are able to reduce the likelihood that their firm becomes insolvent. Following the property-liability insurer insolvency literature (for example, Cummins, Grace, and Phillips [1999]), we estimate the following model: (3)

rst I it = + X it + X itfc + it

where for insurer i and data year t: I it is the unobserved propensity to fail, is an intercept

rst term, X it is a vector of time-varying regulatory solvency tools, X itfc is a vector of time-varying

firm characteristics, and it is an error term. We estimate equation (3) using a discrete-time hazard model (Shumway [2001]). The dependent variable is equal to one if the insurer is insolvent in either year t+1 or t+2. We use all insurers for which we have data. In an effort to include as many insolvent observations as possible in the analysis, we include insurers who report data two years prior to formal regulatory action if they do not report in the year prior to the year of regulatory action. We identify 301 property-liability insurers that fail between 1990 and 2002. In equation (3), the firm characteristics are the natural logarithm of total assets and a mutual indicator. The regulatory solvency tools are the financial ratios in the NAIC's IRIS and FAST systems. Ratios shared by the two systems are used once. We also include state of domicile and year fixed effects in some specifications.

20

The estimation results of the likelihood of insolvency are located in Table IX.31 Columns one and four are baseline specifications. They do not contain measures of managerial ability, so that we can determine whether incorporating managerial quality improves insolvency prediction. Pure technical efficiency is significantly negative, indicating that managers capable of implementing the best practice technology manage firms that are less likely to fail (p<0.05). Allocative efficiency, a measure of the manager's ability to utilize the right combination of inputs, is also negative (p<0.01). Thus, managers who use the correct combination of inputs are less likely to see their firms become insolvent. We also investigate the classification rates of the insolvent firms. An assessment of classification accuracy is made using a yearly optimal cutoff point for a 40:1 relative cost ratio between Type I error and Type II error.32 As described above, this ratio is estimated from New York data on the cost of misclassifying failing firms (total guarantee fund assessments) to the cost of misclassifying solvent firms (formal examinations). The inclusion of our managerial quality variables improves the average classification rate of insolvent firms by 11.5 percent compared to the baseline specification (column 2 of Table IX).33 We also regress the measured probability of insolvency (from Table IX column 3) on the CEO fixed effects of each efficiency variable. We weigh each observation by the inverse of the

In addition to the discrete-time hazard model, we also perform yearly logistic regressions. The results are qualitatively similar. 32 To reduce the upward bias that occurs when the same sample is used for estimation and prediction purposes, the approximate jackknife method (Pregibon [1981]) is used to calculate the predicted probabilities of insolvency for the error trade-offs. In addition to using the approximate jackknife method to calculate the predicted probabilities of insolvency, we randomly withheld single insolvent firms for the year-by-year analysis and groups of insolvent insurers for the panel analysis. The results are robust to the exclusion of these firms. 33 The inclusion of the components of cost efficiency also yields a significantly greater Receiver Operating Characteristic (ROC) area index than the baseline specifications. The goal of the ROC analysis is to provide a statistical test of whether a given model outperforms an alternative model in categorizing observations into two mutually exclusive groups for various Type II error rates (see Metz, Wang, and Kronman [1984] and Cummins, Grace and Phillips [1999]). A model that perfectly categorizes the insolvent and solvent companies will have an ROC area index equal to 1.0, while a model with no discriminatory power will result in an area index of 0.50. The areas under two ROC curves are compared using a nonparametric Chi-square test based on the theory of the MannWhitney U-statistic (see DeLong, DeLong, and Clarke-Pearson [1988]).

31

21

standard error of the right-hand-side variable, which we obtain from the regressions estimated in row 5 of Table III (Bertrand and Schoar [2003]). The results suggest that manager-specific ability makes an economically and statistically significant impact on a firm's likelihood of insolvency. Replacing the worst CEO in terms of pure technical efficiency with the median CEO reduces the probability of insolvency by roughly 51 percent. Moreover, substituting a CEO at the twenty-fifth percentile of the fixed effect distribution of allocative efficiency with one at the seventy-fifth percentile decreases the likelihood of insolvency by over 40 percent.

VI. Cost of Insolvency

We also examine whether managerial ability influences the cost of insolvency. Previous studies have examined the differences in the cost of resolving insolvencies across insurers (Hall [2000] and Grace, Klein, and Phillips [2009]). Our study differs from the previous research, in that while controlling for the influence of the incentive structure on regulators (Hall [2000]) and managers (Lee, Mayers, and Smith [1997] and Grace, Klein, and Phillips [2009]), we directly examine whether the quality of management affects the differences in the costs of insolvency. The relative cost of insolvency is measured as the ratio of cumulative net guaranty association assessments from the insolvency as of 2003 to the assets of the firm prior to the regulator taking formal regulatory action. A limitation of this measure is that we only have estimates for firms with claims covered by guaranty associations.34 In addition, the cost of the insolvency has to exceed the funds that can be collected from selling the firm's assets. Therefore, we do not directly observe the net costs when the assets of the insurer are sufficient to pay the

The guarantee associations are state specific funds that assess solvent insurers to pay the claims of an insolvent insurer. Most state guarantee funds cover personal lines of insurance (like auto and homeowners), but there is less homogeneity amongst the states regarding coverage for commercial lines of insurance. There is also heterogeneity in the deductibles and coverage limits applied to covered lines.

34

22

covered insurance claims. For that reason the underlying linear regression of the latent variable is: (4)

y i* = + X im a + X i fc + X i fo r + i

where y i* is the latent resolution cost variable for insurer i (it is equal to the ratio of net cumulative guaranty assessments by 2003 to insurer i's total assets in the year prior to formal regulatory action), X im a is vector of managerial ability variables for firm i, X i fc is a vector of firm characteristic variables for firm i, and X i for is a vector of regulatory forbearance variables. All independent variables are recorded in the year of the firm's entrance into regulatory scrutiny during the 1989 to 2000 period.35 The random error term is i . The observed variable yi is equal to yi* whenever yi* is greater than 0 and is equal to 0 otherwise. Consequently, the observed variable is censored at 0. We use Tobit estimation techniques to account for the censoring of the dependent variable. Table X presents summary statistics of the characteristics of U.S. property-liability insolvencies. On average, there are 2.76 years between a firm's entrance into regulatory scrutiny and formal regulatory action against the insurer, the first event year (FEY). The median interval is 3 years. For firms that access the guaranty fund system, the average cost to resolve the insolvency is roughly $1.02 for every dollar of pre-insolvency assets. The average cost for the five most expensive insolvencies is $9.68 and it is an astounding $27.92 for the most expensive. The skewness in our dependent variable (8.23 for the 111 firms that access the guaranty fund system and 9.34 for all firms in our sample) creates some econometric challenges in estimating equation (4). Tobit maximum likelihood estimates yield inconsistent estimates when

While it is possible that a firm can enter regulatory scrutiny multiple times during our sample period, we only record variables from the final time a firm enters regulatory scrutiny. We also repeat the analysis using the firm's first entrance. Our results are virtually unchanged.

35

23

the disturbances are non-normal (Arabmazar and Schmidt [1982]). To control for the extreme skewness of the dependent variable, we adopt two additional econometric strategies: (1) we drop all observations above the 95th percentile from the sample (this decreases the skewness of our dependent variable to 1.51); and (2) we estimate the equation under the assumption that the disturbances are drawn from a logistic distribution, a heavy-tailed distribution that will better cope with the skewness of the dependent variable.36 Table XI contains the estimates of these regressions. The results are reasonably stable across the estimation methodologies. The estimates reveal that higher skilled managers are associated with lower costs of failure. Managers that produce at the appropriate scale are associated with a significantly lower cost of insolvency (column 2; p<0.01). Managers capable of implementing the best practice technology are also linked to lower costs of resolution. In fact, pure technical efficiency is the most consistent and robust factor in reducing net guarantee fund assessments across all the estimation strategies (p<0.05). The results suggest a ten percent improvement in pure technical efficiency yields a 4.8 to 9.5 percent reduction in the cost of insolvency. At the mean level, for those firms that access the guaranty fund, this indicates a reduction in cost of $0.049 to $0.097 for every dollar of pre-insolvency assets. Revenue efficiency is also negatively related to the cost of insolvency. On net, managerial ability, at the time the firm is placed into regulatory scrutiny, plays a significant role in reducing the cost of resolving insurer insolvencies.37

36

We also use two other approaches. First, we trim the dependent variable values above the 95th percentile, which reduces the skewness of the remaining 105 non-zero observations to 1.72. Second, instead of using maximum likelihood procedures we use Powell's semiparametric estimator for censored data ­ the censored least absolute deviation (CLAD) model (Powell [1984] [1986]), which provides unbiased, consistent estimates that are robust to non-normality and heteroskadaticity. Our results are robust to these two alternative approaches. 37 There are only eleven CEOs from the manager-firm matched sample represented in the cost of insolvency sample; therefore, we were not able to conduct statistically meaningful regressions for the CEO fixed effect subsample.

24

VII. CONCLUSION

Lang and Stulz [1992], John, Lang, and Netter [1992], and Khanna and Poulsen [1995] all find that while managers are not less skilled than their contemporaries they do serve as scapegoats for their firms' failures. In contrast, we find a correlation between inefficient management and the likelihood of failure. We also find evidence that superior managers are able to remove their firms from regulatory scrutiny sooner than inferior managers. Adding managerial quality to the standard insolvency monitoring tools increases the average classification rate of insolvent firms in standard insolvency models by 11.5%. Furthermore, we discover that managerial quality decreases the ultimate cost of insolvency. Overall, we find that managers of failed firms are less skilled than their peers and the consequence of their incompetence is economically significant. Our definition of management ability is basic. Does the manager do well at minimizing cost, maximizing revenue, operating at the correct scale, and using its inputs according to their marginal productivities? Managerial quality, however, probability consists of much more than just this efficiency component. That being said, this paper does provide strong evidence that a certain type of mismanagement influences firm performance during times of distress.

25

REFERENCES

Agle, B.R., N. Nagarajan, D. Srinivasan, and J.A. Sonnenfeld, "Does CEO Charisma Matter? An Empirical Analysis of the Relationships Among Organizational Performance, Environmental Uncertainty, and Top Management Team Perceptions of Charisma," Academy of Management Journal, 49 (2006), 161-174. Aigner, D.J., C.A.K. Lovell, and P. Schmidt, "Formulation and Estimation of Stochastic Frontier Production Function Models," Journal of Econometrics, 6 (1977), 21-37. Andrade, G., and S. Kaplan, "How Costly is Financial (Not Economic) Distress? Evidence from Highly Leveraged Transactions that Became Distressed," Journal of Finance, 53 (1998). Arabmazar, A., and P. Schmidt, "An Investigation of the Robustness of the Tobit Estimator to Non-normality," Econometrica, 50 (1982), 1055-1063. Banker, R., "Maximum Likelihood, Consistency, and Data Envelopment Analysis: A Statistical Foundation," Management Science, 39 (1993), 1265-1273. Banker, R., and R. Natarajan, "Evaluating contextual variables affecting productivity using data envelopment analysis," Operations Research, 56 (2008), 48-58. Bennedsen, M., K.M. Nielsen, F. Pérez-González, and D. Wolfenzon, "Inside the Family Firm: The Role of Families in Succession Decisions and Performance," Quarterly Journal of Economics, 122 (2007), 647-691. Bennedsen, M., F. Pérez-González, and D. Wolfenzon, "Do CEOs Matter?," NBER Working Paper, 2007). Berger, A., and D. Humphrey, "Measurement and efficiency issues in commercial banking," in Z. Griliches, ed., Output Measurement in the Services Sector (Chicago, IL.: University of Chicago Press, 1992). Bertrand, M., and S. Mullainathan, "Are CEOs Rewarded for Luck? The Ones Without Principals Are," Quarterly Journal of Economics, 116 (2001), 901-932. Bertrand, M., and A. Schoar, "Managing with Style: The Effect of Managers on Firm Policies," Quarterly Journal of Economics, 118 (2003), 1169- 1208. Best, A.M., Best's Insolvency Study ­ Property-Casualty U.S. Insurers 1969-2002 (Oldwick, New Jersey: A.M. Best Company, Inc., 2004). Billett, M., and Y. Qian, "Are Overconfident Managers Born or Made? Evidence of SelfAttribution Bias from Frequent Acquirers," Management Science, 54 (2008), 1037-1051. Bohn, J.G., and B.J. Hall, "The Moral Hazard of Insuring the Insurers," in K.A. Froot, ed., The Financing of Catastrophe Risk (Chicago: The University of Chicago Press, 1999). Cannella, A.A., D.R. Fraser, and D.S. Lee, "Firm Failure and Managerial Labor Markets: Evidence for Texas Banking," Journal of Financial Economics, 38 (1995), 185-210. Charnes, A., W. Cooper, and E. Rhodes, "Measuring the Efficiency of Decision-Making Units," European Journal of Operational Research, 2 (1978), 429-444. Cooper, W.W., L.M. Seiford, and K. Tone, Data Envelopment Analysis (Norwell, MA: Kluwer Academic Publishers, 2000). Cummins, J.D., "Multi-Period Discounted Cash Flow Ratemaking Models in Property-Liability Insurance," Journal of Risk and Insurance, 57 (1990), 79-109. Cummins, J.D., M.F. Grace, J.T. Leverty, and R.D. Phillips, "The Missing Link: Is Book Value Efficiency Recognized By the Market," Working Paper, University of Iowa, (2008). Cummins, J. D., M. F. Grace, and R. D. Phillips, "Regulatory Solvency Prediction in PropertyLiability Insurance: Risk-Based Capital, Audit Ratios, and Cash Flow Simulation," Journal of Risk and Insurance, 66 (1999), 417-458.

26

Cummins, J. D., and D.W. Sommer, "Capital and Risk in Property-Liability Insurance Markets," Journal of Banking and Finance, 20 (1996), 1069-1092. Cummins, J.D., and J. VanDerhie, "A Note on the Relative Efficiency of Property-Liability Insurance Distribution Systems," Bell Journal of Economics, 10 (1979), 709-719. Cummins, J.D., and M. Weiss, "Analyzing firm performance in the insurance industry using frontier efficiency and productivity models," in G. Dionne, ed., Handbook of Insurance (Boston, MA: Kluwer Academic Publishers, 2001). DeLong, E.R., D.M. DeLong, and D.L. Clarke-Pearson, "Comparing the Areas under Two or More Correlated Receiver Operating Characteristic Curves: A Nonparametric Approach," Biometrics, 44 (1988), 837-845. Farrell, M.J., "The Measurement of Productive Efficiency," Journal of the Royal Statistical Society A, 120 (1957), 253-281. Fuld, R.S., "Statement of Richard S. Fuld, Jr. Before The Committee on Oversight and Government Reform," United States House of Representatives (Washington, DC: 2008). Gabaix, X., and A. Landier, "Why Has CEO Pay Increased So Much?," Quarterly Journal of Economics, 123 (2008), 49-100. Grace, M.F., S.E. Harrington, and R.W. Klein, "Risk-Based Capital and Solvency Screening in Property-Liability Insurance: Hypotheses and Empirical Tests," Journal of Risk and Insurance, (1998), 213-243. Grace, M.F., R.W. Klein, and R.D. Phillips, "Insurance Company Failures: Why Do They Cost So Much?," Working Paper, Georgia State University, 2009). Grosskopf, S., "Statistical Inference and Nonparametric Efficiency: A Selective Survey," Journal of Productivity Analysis, 7 (1996), 161-176. Hall, B.J., "Regulatory Free Cash Flow and the High Cost of Insurance Company Failures," Journal of Risk and Insurance, 67 (2000), 415-438. Hambrick, D.C., and P.A. Mason, "Upper Echelons: The Organization as a Reflection of its Top Managers," Academy of Management Review, 9 (1984), 193­206. James, C., "The Losses Realized in Bank Failures," Journal of Finance, 46 (1991), 1223-1242. John, K., L. Lang, and J. Netter, "The Voluntary Restructuring of Large Firms in Response to Performance Decline," Journal of Finance, 47 (1992), 891-917. Joskow, P.L., "Cartels, Competition and Regulation in the Property-Liability Insurance Industry," Bell Journal of Economics and Management Science, 4 (1973), 375-427. Kaufmann, G.G., "Prepared testimony of George G. Kaufmann for the U.S. Senate Committee on Banking, Housing and Urban Affairs," http://www.senate.gov/~banking/01_09hrg/ 091101 /kaufman.htm2001). Khanna, N., and A.B. Poulsen, "Managers of Financially Distressed Firms: Villains or Scapegoats?," Journal of Finance, 50 (1995), 919-940. Klein, R. W., "Insurance Regulation in Transition," Journal of Risk and Insurance, 62 (1995), 363-404. Lang, L., and R. Stulz, "Contagion and Competitive Intra-Industry Effects of Bankruptcy Announcements," Journal of Financial Economics, 32 (1992), 45-60. Lee, S.J., D. Mayers, and C.W. Smith, "Guaranty Funds and Risk-Taking: Evidence from the Insurance Industry," Journal of Financial Economics, 44 (1997), 3-24. Leverty, J.T., and M.F. Grace, "The Robustness of Output Measures in Property-Liability Insurance Efficiency Studies," Journal of Banking and Finance, Forthcoming (2009).

27

Maksimovic, V., and G. Phillips, "The Market for Corporate Assets: Who Engages in Mergers and Asset Sales and Are There Efficiency Gains?," Journal of Finance, 56 (2001), 2019­ 2065. Malmendier, U.M., and G. Tate, "CEO Overconfidence and Corporate Investment," Journal of Finance, 60 (2005), 2661-2700. Mayers, D., and C. W. Smith, "Managerial Discretion, Regulation, and Stock Insurer Ownership Structure," Journal of Risk and Insurance, 61 (1994), 638-655. Meier, K.J., The Political Economy of Regulation: The Case of Insurance (Albany, NY: State University of New York, 1988). Metz, C.E., P.L. Wang, and H.B. Kronman, "A New Approach for Testing the Significance of Difference Between ROC Curves Measured from Correlated Data," in F. Deconick, ed., Information Processing in Medical Imaging: Proceedings of the 8th Conference (Boston, MA: Martinus Nijhoff, 1984), pp. 432-445. NAIC, Insurance Regulatory System, (Kansas City, MO: National Association of Insurance Commissioners Publication Department, 2002). National Conference on Insurance Guaranty Funds, Assessment and Financial Information (Indianapolis, IN: National Conference on Insurance Guaranty Funds, 2003). Pérez-González, F., "Inherited Control and Firm Performance," American Economic Review, 96 (2006), 1559-1588. Phillips, R.D., J.D. Cummins, and F. Allen, "Financial Pricing of Insurance in a Multiline Insurance Company," Journal of Risk and Insurance, 65 (1998), 597-636. Powell, J., "Least Absolute Deviations Estimation for the Censored Regression Model," Journal of Econometrics, 25 (1984), 303-325. Powell, J., "Symmetrically Trimmed Least Squares Estimation for Tobit Models," Econometrica, 54 (1986), 1435-1460. Pregibon, D., "Logistic Regression Diagnostics," Annals of Statistics, 9 (1981), 705-724. Schoar, A., "Effects of Diversification on Productivity," Journal of Finance, 57 (2002), 23792403. Shumway, T., "Forecasting Bankruptcy More Accurately: A Simple Hazard Model," Journal of Business, 74 (2001), 101-124. Simar, L., and P. Wilson, "Sensitivity Analysis of Efficiency Scores: How to Bootstrap in Nonparametric Frontier Models," Management Science, 44 (1998), 49-61. Simar, L., and P. Wilson, "A general methodology for bootstrapping in non-parametric frontier models," Journal of Applied Statistics, 27 (2000), 779-802. State of New York, "Annual Report of the Superintendent of the Insurance to the New York Legislature," (2000-2003). Sullivan, M., "Testimony of Martin Sullivan Before The Committee on Oversight and Government Reform," United States House of Representatives (Washington, DC: 2008). Waldman, D., G.G. Ramirez, R.J. House, and P. Puranam, "CEO Leadership and Organizational Performance: The Moderating Effect of Environmental Uncertainty," Academy of Management Journal, XLIV (2001), 134­143. Warner, J.B., R.L. Watts, and K. Wruck, "Stock Prices and Top Management Changes," Journal of Financial Economics, 20 (1989), 461-492. Weisbach, M., "CEO Turnover and the Firm's Investment Decisions," Journal of Financial Economics, 37 (1995), 159-188.

28

DATA APPENDIX

Variable Inputs and Outputs Input Quantities Administrative Labor Agent Labor Definition Source(s)

Materials and Business Services Financial Equity Capital The average of the beginning and end-of-year real equity capital (policyholders' surplus). Debt Capital The sum of real loss reserves and real unearned premiums reserves. Input Prices Administrative Labor The average weekly wage rate, in the state in which the insurer's home office is located, for the property-liability insurer Standard Industrial Classification (SIC 6331). Agent Labor The average weekly wage rate for insurance agents (SIC 6411). The average weekly wage rate for business services (SIC 7300). Materials and Business Services Financial Equity Capital The average 90-day Treasury bill rate in year t , plus the long-term (1926 to the end of year t ) average market risk premium on large company stocks. Debt Capital Total investment income minus expected investment income attributed to equity capital divided by average policyholder-supplied debt capital. Expected investment income attributable to equity capital is the rate of investment return multiplied by average equity capital for the year. Output Quantities Personal Short-Tail Present Value of Real Losses Incurred. The tail length refers to the Personal Long-Tail length of the loss cash flow stream. Estimates of the length of the Commercial Short-Tail loss cash flow stream are obtained by applying Taylor separation Commercial Long-Tail methods (see Cummins [1990]). Discounting is performed using U.S. Treasury yields obtained from the Federal Reserve Economic Database. Intermediation The average real invested assets for each year. Output Prices Personal Short-Tail Output prices are defined as the difference of real premiums Personal Long-Tail earned extracted from the balance sheet of the NAIC annual Commercial Short-Tail statement and the real present value of losses incurred for the Commercial Long-Tail output divided by the real present value of losses incurred. Intermediation The expected rate of return on the insurer's assets. The ratio of actual investment income (minus dividends on stocks) to debt holdings is used to represent the rate of return on the debt component of the portfolio (Cummins and Weiss [2001]). The expected return on stocks is calculated as the return on the 90-day Treasury bill rate at the end of the preceding year plus the longterm (1926 to the end of the preceding year) average market risk premium on large company stocks. We create a weighted average of the debt and equity returns with the weights equal to the proportion of the firm's total portfolio invested in debt instruments and equity.

The sum of salaries, payroll taxes, and employee relations and NAIC Annual Statement welfare divided by input price of administrative labor. The sum of net commissions, brokerage fees and allowance for NAIC Annual Statement agents divided by the input price of agent labor. The sum of all non-labor expenses divided by its input price. NAIC Annual Statement NAIC Annual Statement NAIC Annual Statement

U.S. Dept. of Labor

U.S. Dept. of Labor U.S. Dept. of Labor Ibbotson Associates

NAIC Annual Statement

NAIC Annual Statement & Federal Reserve Economic Database

NAIC Annual Statement NAIC Annual Statement

NAIC Annual Statement & Ibbotson Associates

(Continued on the next page)

29

DATA APPENDIX

Variable (Continued) Definition Sources

Managerial Ability Pure Technical Efficiency The proportion by which the firm could reduce its input usage by Authors' Calculation adopting the best technology. Scale Efficiency Scale efficiency results from firms operating with constant returns to Authors' Calculation scale; therefore, firms operating with increasing and decreasing returns exhibit scale inefficiency. Allocative efficiency is the proportion by which the firm could reduce Authors' Calculation its input usage by adopting the cost-minimizing input combination. Cost efficiency is the ratio of minumum cost to observed cost. A Authors' Calculation firm is cost efficient if, and only if, it is pure technically, scale, and allocatively efficient. Revenue efficiency is a measure of the firm's ability to maximize Authors' Calculation output with a given input level. Indicator variable set equal to one if the firm has a mutual A.M. Best Company organizational structure, and zero otherwise. Indicator variable set equal to one if the firm has a stock organizational structure, and zero otherwise. Natural logarithm of total assets. Indicator variable set equal to one if the firm is a direct writer of insurance, and zero otherwise. Indicator variable set equal to one if the individual insurer is a member of a group of affiliated insurers, and zero otherwise. A.M. Best Company NAIC Annual Statement A.M. Best Company A.M. Best Company

Allocative Efficiency

Cost Efficiency

Revenue Efficiency Firm Characteristics Mutual Stock Size Direct Group State of Domicile Firm Quality Line of business diversification

The state in which the insurer is incorporated, or, if it is not A.M. Best Company incorporated, the state under whose laws it was formed. Herfindahl index of the firm's written business in each of the 26 NAIC Annual Statement insurance product lines. The line of business Herfindahl Index is computed as the sum of squares of direct premiums written in state i divided by its total direct premium written. A larger value indicates greater concentration of the firm's production across the various lines of insurance. Herfindahl index of the firm's written business in each of the 50 NAIC Annual Statement states and the District of Columbia. It is calculated as the sum of the squares of premium written in product line p divided by its total premium written. A larger value indicates a greater concentration. Ratio of net premiums written to policyholders' surplus. Percent of total premiums written in commercial long-tail lines of business. Percent of total premiums written in commercial short-tail lines of business. Percent of total premiums written in personal long-tail lines of business. Percent of total premiums written in personal short-tail lines of business. NAIC Annual Statement NAIC Annual Statement NAIC Annual Statement NAIC Annual Statement NAIC Annual Statement

Geographical diversification

Leverage Commercial long-tail Commercial short-tail Personal long-tail Personal short-tail

(Continued on the next page)

30

DATA APPENDIX

Variable Firm Quality (Continued) Fail Indicator variable set equal to one if the firm goes insolvent within NAIC's Report on Receiverships and the our sample period. A.M. Best Company Percent of total assets in stocks, investment grade bonds, and NAIC Annual Statement cash. Percent of total premiums written in property lines in Gulf Atlantic Coast states and in earthquake insurance. Ratio of total liabilities to total assets. Natural logarithm of the number of product lines that the firm conducts business. NAIC Annual Statement NAIC Annual Statement NAIC Annual Statement (Continued) Definition Sources

Liquid assets Catastrophe Liabilities / assets Ln lines of business Ln states Regulatory Variables Cost of insolvency

Natural logarithm of the number of states in which the firm conducts NAIC Annual Statement business. Net guarantee fund assessments in 2003 divided by the firm's pre- Assessment and Financial Information insolvency assets. Report published by the National Conference of Insurance Guaranty Funds (NCIGF, 2003) Regulatory solvency screening systems designed to screen and NAIC Annual Statement prioritize insurance companies for more in-depth financial analysis. The Insurance Regulatory Information System (IRIS) is a set of twelve ratios. The Financial Analysis and Surveillance Tracking (FAST) system is a set of twenty-five financial ratios. For additional details we refer readers to Klein [1995]. A firm is classified as under scrutiny if the firm's probability of Authors' Calculation insolvency is greater than the probability cut-off point. The probability cutoff point is the discrete probability level for which the 40:1 relative cost ratio between type I and type II error is minimized. Type I error is the probability that a firm which subsequently fails is predicted to remain solvent and type II error is the probability that a firm which remains solvent is predicted to fail. The year of formal regulatory action against the insurer. Formal NAIC's Report on regulatory action is classified as proceedings for conservation of Receiverships and the A.M. Best Company assets, rehabilitation, receivership, or liquidation. An indicator variable set equal to one if the first event year is greater than 1994 and zero otherwise. NAIC's Report on Receiverships and the A.M. Best Company

IRIS and FAST

Year of regulatory scrutiny

First event year (FEY)

FEY > 1994

Single State Insurer Liquidation year - FEY FEY - year of regulatory scrutiny

An indicator variable set equal to one if the insurer writes business NAIC Annual Statement in only one state, and zero otherwise. Year that the assets of the firm were liquidated minus year of formal NAIC's Report on Receiverships and the regulatory action against the insurer. A.M. Best Company The first event year minus the the year of regulatory scrutiny. NAIC's Report on Receiverships and the A.M. Best Company

31

TABLE I Descriptive Statistics of Insurer Inputs and Outputs Mean (2) 1965 2653 5275 94158 160476 5.314 4.543 3.118 0.139 0.066 9910 25971 18053 13699 236455 0.327 0.289 0.901 0.062 0.080 St. dev. (3) 10455 11830 30710 578562 725031 0.818 0.608 0.765 0.015 0.592 117880 233304 762412 301976 1212156 0.611 0.655 1.339 0.283 0.029 16914

(1) Input Quantities (000s) Administrative (Home Office) Labor Agent Labor Materials and Business Services Financial Equity Capital Debt Capital Input Prices Administrative (Home Office) Labor Agent Labor Materials and Business Services Financial Equity Capital Debt Capital Output Quantities (000s) Personal Short-Tail Personal Long-Tail Commercial Short-Tail Commercial Long-Tail Intermediation Output Prices Personal Short-Tail Personal Long-Tail Commercial Short-Tail Commercial Long-Tail Intermediation Firm-Year Observations

a. The table presents the input and output quantities and prices used in the frontier efficiency analysis. Quantities and prices are unweighted sample means and medians over the period, 1989 to 2000. All variables are defined in the Data Appendix.

32

Table II Descriptive Statistics of the Manager-Firm Matched Sample and the Full Sample Full sample: Manager in more than 1992 to 1999 one firm: 1992 to 1999 Mean St. dev. Mean St. dev. (1) (2) (3) (4) (5) Pure technical efficiency 0.669 0.222 0.667 0.222 Scale efficiency 0.886 0.143 0.898 0.135 Allocative efficiency 0.583 0.175 0.591 0.179 Cost efficiency 0.344 0.172 0.352 0.175 Revenue efficiency 0.221 0.226 0.233 0.242 Ln assets 18.322 1.953 17.824 1.982 Mutual 0.137 0.344 0.230 0.421 2545 12256 Sample size a. "Manager in more than one firm sample" refers to the set of firm-year observations that have at least one manager observed in multiple firms from 1992 to 1999. b. "Full sample" refers to the complete set of firm-year observations. c. All differences between the mean values of the manager-firm matched sample and the full sample are significant at the 10 percent level. d. All variables are defined in the Data Appendix.

33

TABLE III CEO Effects on Firm Performance Proportion of Fixed Effects F-tests on Fixed Effects for Statistically Significant (p< 0.10, two-tailed) Firm CEO Group Firm CEO Group Adjusted R2 (1) (2) (3) (4) (5) (6) (7) (8) Pure Technical Efficiency 28% 4.45 (<.0001, 633) 0.8534 Pure Technical Efficiency 47% 4.42 (<.0001, 338) 0.8097 Pure Technical Efficiency 46% 65% 3.36 (<.0001, 520) 4.02 (<.0001, 214) 0.8632 Pure Technical Efficiency 27% 27% 22% 3.20 (<.0001, 475) 3.16 (<.0001, 194) 2.03 (<.0001, 150) 0.8665 Scale Efficiency 11% 4.00 (<.0001, 633) 0.7394 Scale Efficiency 15% 2.88 (<.0001, 338) 0.6267 Scale Efficiency 22% 7% 3.31 (<.0001, 520) 3.33 (<.0001, 214) 0.7461 Scale Efficiency 38% 34% 26% 3.26 (<.0001, 475) 2.71 (<.0001, 194) 1.49 (0.0002, 150) 0.7483 Allocative Efficiency 14% 4.66 (<.0001, 633) 0.8958 Allocative Efficiency 20% 5.94 (<.0001, 338) 0.8788 Allocative Efficiency 20% 8% 2.81 (<.0001, 520) 3.72 (<.0001, 214) 0.9037 Allocative Efficiency 18% 17% 17% 2.65 (<.0001, 475) 2.87 (<.0001, 194) 1.87 (<.0001, 150) 0.9066 Cost Efficiency 20% 3.80 (<.0001, 633) 0.9327 Cost Efficiency 35% 5.32 (<.0001, 338) 0.9271 Cost Efficiency 21% 13% 2.31 (<.0001, 520) 3.30 (<.0001, 214) 0.9362 Cost Efficiency 34% 41% 36% 2.06 (<.0001, 475) 2.42 (<.0001, 194) 1.81 (<.0001, 150) 0.9376 Revenue Efficiency 78% 3.65 (<.0001, 633) 0.9386 Revenue Efficiency 54% 4.35 (<.0001, 338) 0.9289 Revenue Efficiency 29% 38% 2.60 (<.0001, 520) 3.53 (<.0001, 214) 0.9427 Revenue Efficiency 20% 26% 21% 2.59 (<.0001, 475) 2.85 (<.0001, 194) 1.76 (<.0001, 150) 0.9445 a. The sample is the manager-firm matched sample described in subsection II.C and Table II. The number of firm-years is 2545. Details on the definition and construction of the efficiency variables reported in the table are available in section III.A. b. Reported in the table are the results from fixed effects panel regressions estimated with a standard Heckman two-stage regression methodology to control for potential sample selection bias. Standard errors are clustered at the firm level. The dependent variable is the natural logarithm of the variable reported in column one. For each dependent variable the fixed effects included are row 1: firm and year fixed effects; row 2: CEO and year fixed effects; row 3: Firm, CEO and year fixed effects; row 4: firm, CEO, group and year fixed effects. In addition to the fixed effects, all regressions include logarithm of total assets and a mutual organizational form indicator variable. c. Reported are the proportion of the firm fixed effects (column 2), CEO fixed effects (column 3), and group fixed effects (Column 4) that are statistically significant (p < 0.10, two-tailed). We also report the F-tests for the joint significance of the firm fixed effects (column 5), CEO fixed effects (column 6), and group fixed effects (Column 7). For each F-test we report the value of the F-statistic, the p-value, and the number of constraints. Column 8 reports the adjusted R2 for each regression.

34

TABLE IV Distribution of Manager Fixed Effects Median (2) -0.063 0.949 0.745 0.778 0.569 Standard deviation (3) 1.094 1.013 1.334 1.475 1.050 25th 75th Percentile Percentile (4) (5) -0.743 0.508 -0.118 1.444 -0.442 1.322 -0.612 1.456 0.008 0.898

(1) Pure technical efficiency Scale efficiency Allocative efficiency Cost efficiency Revenue efficiency

a. The fixed effects used in this table are retrieved from the regressions reported in Table III (row 4). There are 194 manager fixed effects. b. Column 2 reports the median fixed effect for each efficiency variable. Column 3 reports the standard deviation of the fixed effects. Columns 4 and 5 report the fixed effects at the twenty-fifth percentile and seventy-fifth percentile of the distribution, respectively. c. Each fixed effect is weighted by the inverse of its standard error to account for estimation error (Bertrand and Schoar, 2003).

35

TABLE V Relationship Between the Manager Fixed Effects Pure technical Scale Allocative Cost efficiency efficiency efficiency efficiency (2) (3) (4) (5) (1) Scale efficiency 0.6635 (.1186) [0.1402] Allocative efficiency 0.4501 0.2431 (.0553) (.0316) [0.2567] [0.2351] Cost efficiency 0.3709 0.1756 0.4496 (.0192) (.0136) (.0179) [0.6600] [0.4643] [0.7657] Revenue efficiency 0.1741 0.0650 0.1780 0.4171 (.0224) (.0145) (.0280) (.0511) [0.2527] [0.1018] [0.1852] [0.2724] a. Each entry in this table corresponds to a different regression. b. Each entry reports the coefficient from a weighted regression of the fixed effects from the row variable on the fixed effects from the column variable. The fixed effects used in this table are retrieved from the regressions reported in Table III (row 4). The regressions are weighted by the inverse of the standard error on the independent variable (Bertrand and Schoar, 2003). c. All coefficients are signficant at the 1 percent level. Standard errors are reported in parentheses below the coefficient. The R2 is reported in square brackets below the coefficient. There are 194 observations in each regression.

36

TABLE VI Characteristics of Firms Under Regulatory Scrutiny All Firms (1) Probability cut-off point Number of firm-years Firm Characteristics Mutual Stock Size (Ln assets) Firm Quality Probability of insolvency Line of business diversification Geographical diversification Leverage Commercial long-tail Commercial short-tail Personal long-tail Personal short-tail Fail (2) 0.026 16899 0.250 0.715 17.758 0.016 0.409 0.582 1.326 0.295 0.330 0.333 0.101 0.017 Firms Under Scrutiny (3) 1913 0.106 0.829 16.356 0.071 0.513 0.730 1.749 0.282 0.271 0.346 0.111 0.087 Firms Not Under Scrutiny (4) 14986 0.268 0.701 17.937 0.009 0.396 0.564 1.272 0.296 0.337 0.331 0.099 0.008 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.336 0.370 0.743 0.002 0.000 p-value (5)

a. This table provides mean values of firm characteristics for the years 1989 to 2000. b. Firms are classified into two groups: under scrutiny and not under scrutiny. A firm is classified as under scrutiny if the firm's probability of insolvency is greater than the probability cutoff point. It is classified as not under scrutiny if the firm's probability of insolvency is less than or equal to the probability cut-off point. The probability cutoff point is the discrete probability level for which the 40:1 relative cost ratio between type I and type II error is minimized. Type I error is the probability that a firm which subsequently fails is predicted to remain solvent and type II error is the probability that a firm which remains solvent is predicted to fail. d. Column 5 reports the p-value for the test of the null hypothesis that the difference between firms under scrutiny and those not under scrutiny is equal to zero. e. All variables are defined in the Data Appendix.

37

Table VII Managerial Ability and the Duration of Regulatory Scrutiny (1) Pure technical efficiency Scale efficiency Allocative efficiency Revenue efficiency (2) -0.110 (0.04) -0.169 (0.079) 0.062 (0.051) (3) -0.082 (0.033) -0.113 (0.062) 0.042 (0.04) (4) -0.084 (0.034) -0.099 (0.066) 0.043 (0.042) (5) (6) (7)

-0.086 (0.03) Yes Yes No No 1755 -814.294 Yes Yes Yes No 1755 -520.383 Yes Yes Yes Yes 1755 -500.261 Yes Yes No No 1755 -818.954

-0.075 (0.027) Yes Yes Yes No 1755 -522.701

-0.075 (0.028) Yes Yes Yes Yes 1755 -502.186

Firm Characteristic Variables Firm Quality Variables Year Controls State of Domicile Controls Number of Observations Log Likelihood

a. This table reports the results from a maximum-likelihood log-logistic hazard regression model of the time under regulatory scrutiny. The duration of regulatory scrutiny is the elapsed time from an insurers' first year subject to scrutiny to the year the firm is no longer under watch or the year at which formal regulatory action is taken against the firm. The standard errors adjusted for clustering at the firm level are reported in parentheses. b. The intercept is omitted from the table. c. The "Firm Characteristic Variables" are size, mutual, direct, and group (see Data Appendix for descriptions of these variables). d. The "Firm Quality Variables" are leverage, liquid assets, catastrophe, line of business diversification, geographical diversification, commercial long-tail, commercial short-tail, personal long-tail, and fail (see Data Appendix for descriptions of these variables). The regressions also include the interaction of fail with catastrophe, leverage, and group, respectively.

38

TABLE VIII Managerial Ability at Entry Into and Exit From Regulatory Scrutiny Mean (1) Managerial Ability Pure technical efficiency Scale efficiency Allocative efficiency Cost efficiency Revenue efficiency Firm Characteristics Size (Ln assets) Firm Quality Leverage Liquid assets Catastrophe Geographical diversification Line of business diversification Commercial long-tail Commercial short-tail Personal short-tail Personal long-tail Number of observations Entry (2) 0.603 0.902 0.517 0.273 0.183 16.392 1.871 0.804 0.073 0.710 0.485 0.265 1.277 0.316 0.109 1030 Exit (3) 0.650 0.908 0.505 0.297 0.198 16.536 1.733 0.819 0.079 0.718 0.479 0.257 0.311 0.311 0.109 1030 (3) - (2) (4) 0.047 0.006 -0.012 0.024 0.014 0.144 -0.138 0.014 0.007 0.008 -0.006 -0.008 -0.966 -0.005 0.001 Entry (5) 0.569 0.958 0.505 0.235 0.051 16.279 1.441 0.842 0.000 1.000 0.446 0.079 0.199 0.237 0.001 1030 Median Exit (6) 0.617 0.964 0.511 0.268 0.082 16.450 1.369 0.856 0.000 1.000 0.432 0.079 0.194 0.223 0.006 1030 (6) - (5) (7) 0.048 0.005 0.005 0.033 0.031 0.171 -0.072 0.014 0.000 0.000 -0.015 -0.001 -0.005 -0.014 0.004

a. This table provides summary statistics of firm characteristics at entrance into and exit from regulatory scrutiny. Differences between the values at entrance into and exit from regulatory scrutiny that are significant at the 5 percent level are highlighted in bold. Significance is determined using a mean two sample t-test or Wilcoxon matched-pair signed-ranks test. b. All variables are defined in the Data Appendix.

39

TABLE IX Managerial Ability and Likelihood of Insolvency (1) Firm Characterstics Size (Ln assets) Mutual Managerial Ability Pure technical efficiency Scale efficiency Allocative efficiency Revenue efficiency IRIS ratios FAST ratios Year controls State of domicile controls Number of observations Number of insolvencies Log likelihood Yes Yes Yes No 16705 301 -1229.3 435.0 0.150 0.8186 Yes Yes Yes No 16705 301 -1214.4 464.9 0.161 0.8297 (2) -0.339 (.051) -1.456 (0.331) (3) -0.298 (0.058) -1.373 (0.343) -0.938 (0.382) -0.184 (0.638) -1.653 (0.538) -0.273 (0.282) Yes Yes Yes No 16705 301 -1228.7 436.3 0.151 0.8191 Yes Yes Yes Yes 14252 301 -1130.5 541.0 0.193 0.8362 Yes Yes Yes Yes 14252 301 -1118.8 564.5 0.201 0.8482 (4) -0.349 (0.052) -1.494 (0.342) (5) -0.34309 (0.064) -1.19062 (0.375) (6) -0.299 (0.070) -1.100 (0.379) -0.762 (0.373) -0.041 (0.680) -1.545 (0.524) -0.273 (0.285) Yes Yes Yes Yes 14252 301 -1129.9 542.2 0.194 0.8365 (7) -0.354 (0.065) -1.229 (0.383)

2

Pseudo R2 Area index of ROC

a. This table presents the results from a discrete-time hazard model (Shumway, 2001) for the years 1989 to 2000. The dependent variable is equal to one if the insurer is classified as insolvent and zero otherwise. Insurers are classified as insolvent if they were subject to formal regulatory proceedings for conservation of assets, rehabilitation, receivership, or liquidation. Robust standard errors adjusted for clustering at the firm level are in parentheses below the coefficient. b. Intercepts are omitted from the table. c. All models contain year fixed effects and include IRIS and FAST ratios. d. All variables are defined in the Data Appendix.

40

TABLE X Characteristics of U.S. Property & Liability Insolvencies 1989-2000 (1) Cost of insolvency All observations Values truncated at the 95th percentile Values above the 95th percentile are eliminated Only insurers that access the guaranty funds Managerial Ability Pure technical efficiency Scale efficiency Allocative efficiency Cost efficiency Revenue efficiency Firm Characteristics Size (Ln assets) Mutual Group Firm Quality Liabilities / assets Liquid assets Catastrophe Ln lines of business Ln states Regulatory Forbearance Year of regulatory scrutiny First event year (FEY) FEY > 1994 Single State Insurer Liquidation year - FEY FEY - year of regulatory scrutiny Obs. (2) 148 148 140 148 148 148 148 148 148 148 148 148 148 148 148 148 148 148 148 148 148 148 148 Mean (3) 0.765 0.528 0.404 1.020 0.597 0.903 0.459 0.251 0.230 16.508 0.068 0.412 0.688 0.678 0.096 1.592 1.288 Median (4) 0.223 0.223 0.175 0.440 0.568 0.970 0.449 0.205 0.079 16.443 0.000 0.000 0.732 0.694 0.025 1.701 0.693 Std. dev. (5) 2.463 0.749 0.552 2.801 0.269 0.147 0.189 0.188 0.304 1.503 0.252 0.494 0.190 0.189 0.142 0.770 1.486 3.628 3.955 0.501 0.501 1.500 1.808 Min. (6) 0.000 0.000 0.000 0.000 0.109 0.074 0.080 0.034 0.001 11.505 0.000 0.000 0.114 0.185 0.000 0.000 0.000 Max. (7) 27.916 2.706 2.661 27.916 1.000 1.000 1.000 1.000 1.000 22.629 1.000 1.000 0.974 1.000 0.811 3.258 4.025

1992.453 1991.000 1995.216 1994.000 0.466 0.000 0.466 0.000 0.378 0.000 2.764 3.000

1989.000 2000.000 1990.000 2002.000 0.000 1.000 0.000 1.000 0.000 12.000 1.000 12.000

a. This table provides summary statistics for U.S. property-liability insurers that fail between 1989 and 2002. b. Cost of Insolvency: the cost of insolvency is calculated as the net guarantee fund assessments in 2003 divided by the firm's pre-insolvency assets. c. All variables are defined in the Data Appendix.

41

TABLE XI Managerial Ability and the Cost of Insolvency (1) Outliers present Estimation technique Distribution of error Number of observations Pure technical efficiency (2) Yes Tobit Normal 148 -2.226 [-0.867] (0.993) -5.203 [-2.028] (1.656) -1.890 [-0.736] (1.436) (3) No Tobit Normal 140 -0.781 [-0.483] (0.208) -0.446 [-0.276] (0.439) -0.498 [-0.308] (0.303) (4) Yes Tobit Logistic 148 -1.590 [-0.95] (0.601) 0.013 [0.012] (0.822) 0.480 [0.22] (1.071) -0.909 [-0.358] (0.901) Yes Yes Yes Yes Yes -248.14 Yes Yes Yes Yes Yes -69.44 Yes Yes Yes Yes Yes -149.05 Yes Yes Yes Yes Yes -255.40 -0.359 [-0.216] (0.184) Yes Yes Yes Yes Yes -75.77 -1.704 [-0.392] (0.429) Yes Yes Yes Yes Yes -148.04 (5) Yes Tobit Normal 148 (6) No Tobit Normal 140 (7) Yes Tobit Logistic 148

Scale efficiency

Allocative efficiency

Revenue efficiency

Firm Characteristic Variables Firm Quality Variables Regulatory Forbearance Variables State of domicile controls Year controls Log likelihood

a. This table presents the relationship between firm and forbearance characteristics and the cost of insolvency. The dependent variable is the cost of insolvency, which is calculated as the net guarantee fund assessments in 2003 divided by the firm's pre-insolvency assets. Columns (2) and (5) report Tobit maximum likelihood estimates on the full sample of firms. Columns (3) and (6) report Tobit maximum likelihood estimates all observations below the 95th percentile of the cost of insolvency. Columns (4) and (7) report Tobit maximum likelihood estimates on the full sample of firms, under the assumption that the disturbances are drawn from a logistic distribution. b. Unconditional marginal effects are reported in square brackets. For continuous variables marginal effects are recorded at the mean value and for indicator variables the discrete changes are reported. Standard errors are reported in parentheses. c. All models contain state of domicile and year fixed effects and include firm characteristic, firm quality, and regulatory forbearance variables (see TABLE X for a list of these control variables). d. Intercepts are omitted from the table. e. All variables are defined in the Data Appendix.

42

Information

44 pages

Find more like this

Report File (DMCA)

Our content is added by our users. We aim to remove reported files within 1 working day. Please use this link to notify us:

Report this file as copyright or inappropriate

1210383


You might also be interested in

BETA
article4.p65
Microsoft Word - fallaciesfinal57
tirole
Derivatives and Corporate Risk Management
Nicolaus Loos - Dissertation - EDIS - Final.pdf