Internal capital markets and the partial adjustment of leverage

Journal of Banking & Finance 37 (2013) 1029–1039

Contents lists available at SciVerse ScienceDirect

Journal of Banking & Finance journal homepage: www.elsevier.com/locate/jbf

Internal capital markets and the partial adjustment of leverage Stephen G. Fier a,⇑, Kathleen A. McCullough b, James M. Carson c a

University of Mississippi, 338 Holman Hall, University, MS 38677, USA Florida State University, 150 Rovetta Business Building, Tallahassee, FL 32306, USA c University of Georgia, 211 Brooks Hall, Athens, GA 30602, USA b

a r t i c l e

i n f o

Article history: Received 26 October 2011 Accepted 7 November 2012 Available online 21 November 2012 JEL classification: G22 G32

a b s t r a c t Prior literature provides support both for the existence of target capital structures and internal capital markets (ICM). The issue of whether firms use internal capital markets to reduce deviations from target capital structures, however, has yet to be examined. We provide the first empirical evidence of a link between deviations from target leverage and ICM activity. Based on data that allow us to trace intragroup capital market transactions for property–casualty insurers, our findings provide the first joint evidence that affiliated insurance companies have target leverage ratios and that ICM activity is used to manage deviations from target leverage. Ó 2012 Elsevier B.V. All rights reserved.

Keywords: Internal capital markets Leverage Reinsurance

1. Introduction The question of whether firms actively manage capital structure has been investigated since the proposition of the Modigliani and Miller irrelevance theorems in 1958. Subsequent research has examined whether firms actively manage the level of leverage given the costs and benefits of leverage, where support for active management is implied through evidence of a target capital structure (e.g., Flannery and Rangan, 2006; Huang and Ritter, 2009; De Haan and Kakes, 2010; Cheng and Weiss, 2012). Prior literature also provides evidence that firms deviating from target capital structure may make partial adjustments toward the target rather than immediate adjustments due to adjustment costs (e.g., Hovakimian et al., 2001; Flannery and Rangan, 2006; De Haan and Kakes, 2010). A second stream of literature1 explores the unique benefits of capital allocation within the group organizational structure. Specifically, this literature finds that conglomerates have the benefit of internal capital markets (ICMs), whereby the headquarters of the group has the ability to allocate capital across the various group members. The benefits of the internal allocation of capital include lower monitoring costs, reduced agency problems, greater efficiency of capital allocation and, ultimately, lower cost to obtain internal capital (compared to external capital). ⇑ Corresponding author. Tel.: +1 662 915 1353; fax: +1 662 915 5821. E-mail addresses: sfi[email protected] (S.G. Fier), [email protected] (K.A. McCullough), [email protected] (J.M. Carson). 1 E.g., Gertner et al. (1994), Stein (1997), and Powell et al. (2008). 0378-4266/$ - see front matter Ó 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.jbankfin.2012.11.003

Given the potential for target capital structures and deviations from the target, one may expect that ICMs are used to reduce deviations from target capital structure – particularly if deviations from the target are costly (e.g., Flannery and Rangan, 2006). However, to our knowledge, this relation has not been examined empirically. While limited reporting requirements in most industries restrict the ability to test the relation between these two streams of literature, we contend that the property–casualty insurance industry provides a natural setting to test this relation for a number of reasons, including: (1) property–casualty insurance companies may have target capital structures (e.g., Cummins and Doherty, 2002; Cummins and Nini, 2002; Harrington and Niehaus, 2002; Klein et al., 2002; De Haan and Kakes, 2010; Shim, 2010; Cheng and Weiss, 2012); (2) firms in the property–casualty insurance industry have the ability to operate in groups, which allows for an examination of ICM transactions (e.g., Powell and Sommer, 2007; Powell et al., 2008); and (3) property–casualty insurers are required to prepare statutory filings that detail financial transactions between insurance group members (i.e., ICM transactions). We test for the existence of target capital structures in the property–casualty insurance industry using a partial adjustment model. Evidence of a target leverage ratio would suggest that insurers actively manage their capital structure. We then analyze ICM activity among affiliated insurers to determine if the extent of ICM activity (in particular affiliated reinsurance activity) is related to deviations from target capital structure. Our results indicate that insurers have target capital structures and that there is a statistical relation between deviations from target leverage and ICM activity.

1030

S.G. Fier et al. / Journal of Banking & Finance 37 (2013) 1029–1039

We make a number of contributions. First, we build upon the capital structure literature and provide further evidence of management actively adjusting toward a target. Second, given the availability of intra-group capital transfer data in the property– casualty insurance market, our examination of ICMs with respect to capital structure adjustments provides a greater understanding of the mechanisms used to reduce deviations from target capital structures. Most importantly, this is the first study to provide an empirical link between the existence of a target capital structure and how deviations from the target are related to capital flows among group members. The remainder of this paper is organized as follows. Section 2 discusses the prior literature’s examination of target leverage, the costs and benefits of leverage, and how firms may adjust leverage. Section 3 describes internal capital markets and both the costs and benefits associated with ICMs compared to their external counterparts. The primary hypotheses of interest are provided in Section 4. Section 5 discusses the data, methodology, and the variables employed in the study. A discussion regarding the results and implications is provided in Section 6, and Section 7 concludes.

2. Adjustments toward target leverage Many studies examining capital structure maintain that firms have a target capital structure (e.g., Hovakimian et al., 2001; Harrington and Niehaus, 2002; Leary and Roberts, 2005; Flannery and Rangan, 2006; Kayhan and Titman, 2007; Antoniou et al., 2008; Huang and Ritter, 2009; De Haan and Kakes, 2010; Shim, 2010; Cheng and Weiss, 2012). This finding is important because it implies that firms actively manage capital structure, considering both the tax benefits of using debt and the increasing costs of bankruptcy associated with debt. It also suggests that firms make adjustments toward the target capital structure, although much of the evidence suggests that partial adjustments are made rather than immediate adjustments given the costs associated with making such adjustments. For example, Leary and Roberts (2005) report that firms in their sample make capital structure adjustments approximately once per year. Although the literature finds evidence of target leverage ratios in a variety of industries, some multi-industry studies remove insurers from the sample because insurance is a highly regulated industry (e.g., Leary and Roberts, 2005; Flannery and Rangan, 2006; Huang and Ritter, 2009). However, prior literature suggests that in the presence of regulation, insurer capital structure is not bound by regulated capital requirements (i.e., insurers hold more capital than what is required by regulation) and that insurers have target capital structures. For example, De Haan and Kakes (2010) show that when regulatory solvency requirements do not consider insurer risk characteristics, insurer solvency margins are still related to insurer risk characteristics. The authors show that most insurers hold significantly more capital than what regulatory authorities require, and that non-risk-based capital requirements are non-binding.2 Furthermore, Shim (2010) and Cheng and Weiss (2012) provide support of the existence of target capital structures in the insurance industry.3 2 Similarly, Cummins and Doherty (2002) state that insurers have a target capital structure because of a desire to meet consumer demand for ‘‘safe insurance.’’ 3 A number of studies examine the existence of target capital structures in the insurance industry (i.e., Cummins and Nini, 2002; Harrington and Niehaus, 2002; De Haan and Kakes, 2010; Shim, 2010; Cheng and Weiss, 2012), and some provide direct support in favor of the existence of target capital structures in the insurance industry. For instance, De Haan and Kakes (2010) find that insurers have target capital ratios and that those targets are higher than what is required by regulators. They also report that insurers reduce the deviation between actual and target capital ratios by approximately one-third each year. Cheng and Weiss (2012) examine the existence of target risk-based capital (RBC) ratios in the property–casualty insurance industry and report that insurers exhibit a tendency to adjust toward a target RBC ratio.

Similar to other industries, leverage in the insurance industry may be considered beneficial or costly depending on the level of leverage utilized. For an insurer, increased leverage reduces surplus, meaning an increased level of leverage can increase the likelihood of financial distress if the insurer faces higher than expected claims, higher than expected operating costs, or lower than expected investment returns (Staking and Babbel, 1995).4 While an increase in leverage may negatively impact policyholders, it can have either a positive or a negative effect on the owners of the firm. From the owner’s perspective, an increase in leverage can allow the firm to maximize the benefits of both the leverage tax shield and the insolvency put option. However, like most industries, too much leverage can increase the probability of insolvency and reduce the value of the firm. The previous discussion suggests that firms in the property– casualty insurance industry behave in a manner similar to other industries with respect to the existence of a target capital structure. However, unique data advantages exist within the industry that allows us to not only test for the existence of a target capital structure, but also to track internal capital market transactions among group members. The ability to track financial transactions across group members while accounting for important firm-specific factors such as risk-based capital requirements allows us to study issues surrounding target leverage and internal capital markets in a way that is not possible given the opacity of accounting reporting in other industries. Below we further discuss the role of internal capital markets both in an insurance context and beyond.

3. Internal capital markets Gertner et al. (1994) describe internal capital markets as a setting where ‘‘. . .corporate headquarters allocate capital to their business units.’’ The existence of internal capital markets is of particular importance because they can represent an available source of funding that is less costly and more efficient than external capital markets. The lower cost of capital is generally attributed to reduced information asymmetries and lower agency costs within the ICM, which allows for a more efficient deployment of capital among group members (Gertner et al., 1994). Much of the ICM literature evaluates the potential costs and benefits of ICMs and the efficiency of ICMs. Gertner et al. (1994) compare internal capital markets with external capital markets (i.e., bank lending) and argue that ICMs have a stronger ability to monitor how funds are used and a greater ability to reallocate assets from poorly performing projects to more successful projects. Stein (1997) also argues that reallocation is a benefit for the ICM, as the corporate headquarters has the ability to reallocate capital from projects or divisions that are ‘‘losers’’ to those that are ‘‘winners.’’5 Additionally, authors generally argue that internal capital should represent a lower cost option than external capital, given a reduction in agency costs and informational asymmetries. For instance, Desai et al. (2004) provide empirical evidence that firms use internal capital in place of external borrowing when firms are located in countries where the acquisition of external capital is costly and show that internal capital may act as a substitute for costly external capital. While there are benefits to ICMs, the potential exists for additional agency problems and inefficiencies resulting from the use 4 In an insurance context, ‘‘surplus’’ refers to any remaining value once liabilities have been deducted from assets (as with owners’ equity). Surplus is often viewed as the financial cushion that is available to the insurer in instances where losses are greater than anticipated, expenses are greater than expected, or investment income is less than expected. 5 For example, Houston and James (1998) find that bank holding companies create ICMs for the purpose of allocating capital across various subsidiaries.

S.G. Fier et al. / Journal of Banking & Finance 37 (2013) 1029–1039

of ICMs. Scharfstein and Stein (2000) make the argument that ICMs may be less efficient than external capital markets, with management allocating resources away from the stronger divisions and toward the weaker divisions. Similarly, Matsusaka and Nanda (2002) show that while ICMs do allow firms to avoid the use of external finance, ICMs may face an overinvestment problem because the ICM cannot constrain management’s allocation of resources (a benefit of external markets). Gertner et al. (1994) also argue that since managers do not own residual rights to the assets and because headquarters has the ability to reallocate assets, ICMs may reduce managerial incentive to act in an entrepreneurial manner. Empirical evidence supports the existence of active ICMs in the property–casualty insurance industry. For example, Powell and Sommer (2007) and Powell et al. (2008) analyze ICM activity in the property–casualty insurance industry and the efficiency of ICMs. The authors provide support for the use of ICMs (with a focus on affiliate reinsurance transactions) in the property–casualty insurance industry for the purpose of facilitating investment growth and show that a statistical difference between affiliate and non-affiliate uses of capital and reinsurance exists. These results indicate that capital from affiliates is as important (if not more important) than external capital. Reinsurance represents an important ICM transaction that may be used by firms to reduce the deviation between target and realized leverage. Reinsurance is a contractual arrangement between two insurance companies, whereby one insurer transfers insurance risk to another insurer or reinsurer. By purchasing reinsurance, the purchasing company is able to reduce the level of written premiums (and the potential liability associated with ceded risks) which ultimately allows the insurer to reduce leverage. Alternatively, an insurer may purchase less reinsurance (or may choose to sell reinsurance) in order to increase leverage. As discussed in prior literature, reinsurance may act as a substitute for equity capital when attempting to deal with diversifiable (unsystematic) risk (e.g., Berger et al., 1992). Such transactions may occur between affiliated or unaffiliated insurance companies. While insurers disclose a variety of ICM transactions, we focus only on affiliate reinsurance transactions. As noted by Powell et al. (2008), there are a number of reasons to focus on affiliate reinsurance transactions as a measure of ICM activity. First, affiliate reinsurance transactions represent the largest proportion of ICM transactions among affiliated insurers.6 Second, even though data are available on dividend transactions, dividends can only be used as a transfer mechanism when an affiliate owns part of another affiliate. Additionally, a firm does not have the ability to selectively pay dividends to an affiliate while choosing to not pay dividends to other owners, meaning that the payment of dividends to affiliates should not be a strategic ICM activity. Third, reinsurance transactions are not additive to the other methods of capital transfer within an ICM, so one cannot simply add the different types of affiliated transactions with one another.7 Because of these issues, we follow Powell et al. (2008) and focus exclusively on affiliate reinsurance transactions. 4. Hypotheses The primary goal of this study is to test for an empirical link between the target capital structure literature and the internal capital markets (ICM) literature by exploiting the transparency of the statutory filings in the property–casualty insurance marketplace. We present two primary hypotheses. 6 Powell and Sommer (2007) point out that approximately 80% of reinsurance activity occurs within groups rather than among insurers and unaffiliated reinsurers. 7 For example, ceding $1000 in premiums (i.e. buying reinsurance) is not the same as receiving $1000 in dividends or selling an asset to an affiliate for $1000.

1031

4.1. Hypothesis 1: Affiliated property–casualty insurance companies and target capital structure Prior literature argues that profit-maximizing firms should choose an optimal level of debt such that the marginal benefits of taking on the debt are equal to the marginal costs associated with the debt (Jensen, 1986). Staking and Babbel (1995) present evidence that insurers will maintain an optimal level of leverage to maximize the overall value of the firm.8 However, beyond that optimal point, the insurer faces an increased likelihood of financial distress that outweighs the potential benefits from obtaining the leverage.9 Given the findings and expectations of prior literature (e.g., Staking and Babbel, 1995; Cummins and Doherty, 2002; Harrington and Niehaus, 2002; De Haan and Kakes, 2010; Shim, 2010; Cheng and Weiss, 2012) our first testable hypothesis is given as: Hypothesis 1. Affiliated property–casualty insurance companies have target leverage ratios. 4.2. Hypothesis 2: Internal capital markets and deviations from target capital structure Previous literature hypothesizes that firms will make partial adjustments toward a given target capital structure rather than complete (immediate) adjustments because adjustments are costly (Strebulaev, 2007). The adjustments may be made in a variety of ways, including through the use of external and internal capital. The ability to make adjustments and the cost of the adjustment(s) will be determined in part by the availability of internal and external capital. Given the possible benefits of ICM financing (i.e., lower costs and greater access to capital), affiliated insurers should take advantage of ICMs when making adjustments to capital structure.10 Therefore, our second testable hypothesis is: Hypothesis 2. The use of internal capital markets (i.e., affiliate reinsurance) is dependent on the size of the insurer’s deviation from the target capital structure, where overleveraged insurers purchase (cede) more affiliate reinsurance and underleveraged insurers purchase less affiliate reinsurance. We test this hypothesis by examining the relation between deviations from target leverage and the use of ICMs. We anticipate that under-levered affiliates will utilize the ICMs in a manner that allows them to increase leverage by buying less reinsurance while over-levered insurers will utilize the ICMs in a manner that allows 8 While Jensen (1986) and Staking and Babbel (1995) discuss capital structure in terms of an ‘‘optimal’’ value, we are not explicitly examining optimal leverage. Rather, we simply argue that a firm may have a target (which may or may not represent the ‘‘optimal’’ leverage for the firm) to which the firm is converging to. 9 Studies such as De Haan and Kakes (2010) and Shim (2010) note that capital decisions could be determined in part by a holding company and may not entirely be the decision of the affiliated insurer. Although this is an important point, we perform testing of Hypothesis 1 at the firm level rather than at the group level for several key reasons. Of primary importance, this level of analysis is necessary in order to test the use of ICMs for the purpose of adjusting toward the target (i.e., Hypothesis 2, discussed below). For these tests, estimated target leverage values are required for each affiliated insurer (as opposed to the group). Furthermore, the prior literature typically has examined the potential existence of a target capital ratio for insurers at the firm level rather than at the consolidated (group) level (e.g., De Haan and Kakes, 2010; Cheng and Weiss, 2012). Finally, we perform our analysis at the affiliate level to account for affiliate-level regulation in the US Insurers in the US are regulated at the state level rather than at the federal level, and the state regulatory focus typically has been on the solvency of individual insurers rather than on the solvency of the group (e.g., Cummins et al., 1995; Cummins and Sommer, 1996). 10 Jean-Baptiste and Santomero (2000) theoretically examine the design of reinsurance contracts and find that in the presence of asymmetric information, the reinsurer will set a price higher than the ‘‘first best’’ price in order to account for the level of ‘‘noise’’ generated by the information asymmetry. Given reduced ‘‘noise’’ via an internal transaction, one should expect affiliate (internal) reinsurance transactions to be less costly than (external) reinsurance transactions with unaffiliated insurers.

1032

S.G. Fier et al. / Journal of Banking & Finance 37 (2013) 1029–1039

them to reduce leverage by purchasing greater levels of reinsurance (and thus ceding greater levels of premiums). The results of such an analysis would suggest whether deviations from target leverage are related to the use of ICMs and whether ICM transactions can benefit group members in the target capital structure context. 5. Data and methodology 5.1. Data

5.2. Methodology To determine if ICM activity is related to deviations from target leverage, we take the following steps: 1. Determine if a target capital structure exists in the property– casualty insurance industry employing the system GMM estimator developed by Arellano and Bover (1995) and Blundell and Bond (1998). 2. Estimate the target leverage ratio for each firm using the estimates obtained from the system GMM estimation. 3. Calculate the difference between realized leverage and the target (estimated) leverage ratio to determine deviations from the target. 4. Estimate the relation between ICM activity and the size of the deviation from target leverage. Below is a more detailed explanation of the methodology employed in this study. 5.3. Evidence of a target capital structure The first step in this analysis requires us to determine whether property–casualty insurance companies have target leverage ratios. Flannery and Rangan (2006) express target capital structure as:

ð1Þ

where Li;t represents the target leverage for firm i at time t, Xi,t1 is a vector of firm characteristics related to target leverage identified from prior literature for firm i at time t  1, and b is a coefficient vector. The standard partial adjustment model is given by Flannery and Rangan (2006) as: 11 We use the AM Best definition of ‘‘professional reinsurer,’’ where professional reinsurers are defined as any insurer whose reinsurance assumed from unaffiliated firms is greater than 75% of the direct premiums written less reinsurance assumed from affiliated insurers. 12 For robustness purposes, we also re-estimate each of the models after removing firms that are under regulatory scrutiny. The results of these variations of the models are consistent with the results presented in this study.

ð2Þ

Eq. (2) shows that the difference between the target leverage ratio and realized leverage for insurer i is reduced by some proportion, k. We anticipate that, with the existence of adjustment costs, k would fall somewhere within the range of 1 and 0. Because target capital structure is unobservable, we substitute Eq. (1) into Eq. (2), giving an estimable Eq. (3):

Li;t ¼ kbX i;t1 þ ð1  kÞLi;t1 þ di;t

Since ICMs imply that capital is transferred between group members, we examine target capital structure and ICMs for affiliated insurers only. Firm-level insurer-specific data are obtained from the National Association of Insurance Commissioners (NAIC) annual statements for the period from 1996 through 2009. A number of screens are applied prior to estimating the models. First, we remove all unaffiliated insurers from our sample (i.e., any insurer that does not have at least one other affiliated insurer) as these firms would not have the data required for the ICM analysis. Second, we remove all insurers with negative net total assets, negative liabilities, and negative surplus. Finally, we remove all professional reinsurers and insurers classified by the NAIC as a ‘‘US branch or alien insurer.’’11 We also remove observations with missing data for any of the dependent or independent variables used in the models discussed below.12 Dependent and independent variables are winsorized at the 1st and 99th percentiles in order to reduce the potential influence that extreme observations could have on the results.

Li;t ¼ bX i;t1

Li;t  Li;t1 ¼ kðLi;t  Li;t1 Þ þ di;t

ð3Þ

Because we use a dynamic panel data set, the potential exists for the lagged dependent variable to be correlated with the error term in the presence of firm-specific time invariant fixed effects, which can bias the results. In order to correct for this potential bias, we use the system generalized method of moments (GMM) approach of Arellano and Bover (1995) and Blundell and Bond (1998), which has been used in prior target leverage literature (e.g., Antoniou et al., 2008; Lemmon et al., 2008). The system GMM method is most appropriate for models with ‘‘small T, large N’’ panels, which is consistent with our dataset (Roodman, 2009). This methodology is similar to the Arellano and Bond (1991) methodology, except that rather than instrumenting first-differenced variables with level variables, the system GMM instruments level variables with the first differenced variables found on the right hand side of the model. As discussed by Bond (2002), estimating a dynamic panel dataset using ordinary least squares (OLS) results in biased coefficient estimates. Using an OLS model that only incorporates yearly fixed-effects results in an upward biased coefficient estimate, while using an OLS model that incorporates both year- and firm-specific fixed effects results in downward biased coefficient estimates. However, the coefficient estimates from these models may be used to identify a bound for the true estimates, where ‘‘Good estimates of the true parameter should therefore lie in the range between these two values – or at least near it’’ (Roodman, 2009). Given that the estimates from OLS models can provide some guidance regarding an appropriate bound, we also report the estimates using OLS fixed-effect models. 5.4. Target capital structure variables For purposes of this study, our primary measure of leverage is the ratio of total insurer liabilities to total surplus (LS). The principle advantage to using the ratio of liabilities to surplus is that it accounts for the insurer’s potential exposure to claims that it did not anticipate, as well as its potential exposure to other increases in liabilities.13 We anticipate a positive and significant relation between prior period leverage and current period leverage.14 13 Because premiums are paid to the insurer prior to the occurrence of a covered loss, the premiums paid to the insurer may be viewed as borrowed funds (i.e., debt), where the claims payments are analogous to the coupon and principal payments associated with debt (Cummins and Lamm-Tennant, 1994). Under the typical insurance arrangement, the insurer writes an event-contingent contract which states that, at the time of a covered loss, the insurer will indemnify the insured up to certain limits imposed by the contract. In return for this promise, the insured provides payment of a premium to the insurer (typically at the beginning of the policy period), which will amount to the value of the expected loss plus a loading to cover expenses and profit. The premiums then can be used by the insurer for investment purposes. Thus, leverage in the insurance industry (and for purposes of this study) may be viewed with respect to the liabilities of the insurer (primarily reserves to pay for current and future losses). 14 In addition to the ratio of liabilities to total surplus, we also test the ratio of premiums to surplus (PS) and liabilities to total assets (LA). While the ratio of premiums-to-surplus is a common measure of underwriting leverage in the insurance industry, prior literature suggests that the ratio is flawed for two primary reasons. First, it is based on the assumption that the surplus is intended to be used only for the purpose of supporting current business, when in actuality, surplus may be used to support previous, current, or future business (Klein et al., 2002). Second, using net premiums written requires the assumption that reinsurers will meet all of their reinsurance obligations to the insurer. While not reported, we anticipate a positive and significant relation between prior period leverage and current leverage for each leverage measure.

S.G. Fier et al. / Journal of Banking & Finance 37 (2013) 1029–1039

Prior literature has found that a firm’s leverage is impacted by factors including size, profitability, growth, diversification, business mix, and organizational form. While in some cases the expected relations between leverage and variables of interest are similar to what would be anticipated in other industries, insurance-specific issues often require us to refine the proxy and/or rationale for the hypothesized relations. Each of the independent variables and our expectations regarding the relation between these variables and the dependent variables is discussed below. 5.4.1. Size The bankruptcy cost for every dollar in assets is lower for larger firms than it is for smaller firms, which suggests that the cost of financial distress is lower for larger firms (Warner, 1977; Ang et al., 1982). If the cost of financial distress is lower for larger firms, larger firms should require relatively less capital than smaller firms (Klein et al., 2002; De Haan and Kakes, 2010). Thus, we anticipate a positive relation between size and leverage, and we proxy for firm size using the natural logarithm of total insurer assets. 5.4.2. Profitability During periods of greater profitability, insurers may be more likely to ‘‘hoard’’ capital for future use (e.g., Harrington and Niehaus, 2002; De Haan and Kakes, 2010). The effect of the increased holding of capital during periods of profitability should then reduce insurer leverage. This suggests a negative relation between an insurer’s profitability and the level of leverage maintained by the insurer. We account for profitability by including firm-specific return on assets (ROA), calculated as the ratio of net income to assets. 5.4.3. Premium growth Myers and Majluf (1984) note that firms with a greater opportunity for growth will prefer to use internal capital, suggesting that insurers with greater growth opportunities will tend to hold greater levels of capital than insurers with a different outlook. This suggests a negative relation between premium growth and leverage. However, given that premiums are a liability when received, premium growth could result in increased leverage as liabilities increase and surplus decreases to cover expenses (e.g., Browne et al., 1999). We include the change in net premiums written in the models to account for premium growth. 5.4.4. Diversification Greater diversification provides increased independence of individual exposures covered by the insurer and allows the insurer to reduce the impact of a loss shock that could occur within a single line of business or geographic location. Firms that are less diversified may maintain greater levels of capital, resulting in a lower leverage ratio (e.g., Cummins and Nini, 2002; Klein et al., 2002). In order to account for diversification, we include the firm’s geographic Herfindahl–Hirschman Index (HHI) and lineof-business HHI.15 5.4.5. Commercial lines of business Corporate purchasers of insurance generally have a greater knowledge of an insurer’s financial status than do individual purchasers of insurance (Cummins and Nini, 2002). Furthermore, switching costs are generally relatively lower for corporate purchasers of insurance than for individual purchasers. Because corporate insurance is generally less ‘‘sticky,’’ insurers writing a greater proportion of commercial insurance will have a tendency to 15 Geographic HHI is calculated using premiums written across each state in the United States for each insurer in the sample, while line-of-business HHI is calculated using premiums written within each line of business for each insurer in the sample.

1033

maintain a lower level of leverage than will insurers that write a greater amount of insurance in personal lines (Cummins and Nini, 2002). We include the proportion of total premiums written in commercial lines versus those written in personal lines to account for the impact of commercial business.16 We also include a variable to proxy for long-tail lines of business.17 Firms writing a greater proportion of long-tail lines of business may be more inclined to exhibit opportunistic behavior (due in part to their ability to make adjustments to highly subjective reserves). To discourage management from such behavior, firms with a greater proportion of long-tailed lines of business should maintain higher levels of leverage (Cummins and Nini, 2002). Accordingly, we anticipate a positive relation between long-tail lines of business and leverage. We account for long-tail lines of business using the ratio of insurance reserves to losses incurred. Cummins and Nini (2002) use this variable to proxy for the difference between the time when the policy is initially issued and the time at which claims payments are made. 5.4.6. Risk-based capital In addition to the aforementioned variables, we also include a regulatory measure of the firm’s financial strength. A higher RBC ratio suggests that the insurer maintains sufficient capital relative to the amount required by regulators.18 If the firm maintains sufficient levels of capital, the greater level of capital on hand should reduce the firm’s leverage ratio. We predict a negative relation between a firm’s RBC ratio and leverage. 5.4.7. Organizational form Prior literature finds substantial differences across the different organizational structures that exist within the property–casualty insurance industry (e.g., Mayers and Smith, 1994). Given these differences, we include an indicator variable denoting whether or not an individual firm is a mutual company. Either a positive or a negative relation may exist between leverage and the mutual binary variable. A positive relation is expected if mutual insurers maintain lower levels of capitalization because of the reduction of owner– policyholder conflicts (Lamm-Tennant and Starks, 1993; Cummins and Nini, 2002). Alternatively, a negative relation should exist if mutual insurers hold greater levels of capital due to their limited access to capital markets (e.g., Froot and Stein, 1998; Harrington and Niehaus, 2002). We also include an additional binary variable indicating whether a firm is of an organizational form other than stock or mutual.19 5.5. Internal capital markets and deviations from target leverage Once we have estimated the model using Eq. (3), we can then estimate the target using the coefficient estimates obtained from the partial adjustment model. Rearranging Eq. (2) yields the estimated target leverage, which is calculated as:

Li;t ¼

1 ½Li;t  Li;t1  di;t  þ Li;t1 k

ð4Þ

16 All lines of business are considered commercial lines except for homeowners multi-peril, auto physical, farmowners multi-peril, and private passenger auto liability. 17 Long-tail lines of business are those lines in which there is likely to be a considerable amount of time between the premium payment and the time claims are ultimately paid and closed, thus removing the liability from the insurer’s financial statements. 18 Risk-based capital (RBC) requirements were introduced by the NAIC in 1995 for property–casualty insurers in an effort to ensure a minimum level of capital adequacy for insurance companies. 19 To avoid singularity, we omit the ‘‘stock’’ group of insurers for each of the models. Thus, the coefficients on the binary variables representing mutual insurers and ‘‘other’’ organizational forms are relative to stock insurers.

1034

S.G. Fier et al. / Journal of Banking & Finance 37 (2013) 1029–1039

We estimate the target leverage for all affiliated property–casualty insurers in the sample. Once we determine the target leverage ratios, we then compare realized insurer leverage (Li,t) at time t to the estimated target leverage (Li;t ) at time t in order to determine whether firms are over- or under-levered. Firms are considered under-levered if Li;t is greater than Li,t and firms are considered over-levered if Li;t is less than Li,t. We are then able to examine the relation between the size of the deviation from the target and ICM activity. To examine this potential relation, we estimate the following model:

ICMi;t ¼ b0 þ b1 ðLi;t1  Li;t1 Þ þ bX i;t1 þ ei;t

ð5Þ

where ICM represents net internal capital market transactions occurring within a given year; Li;t1  Li;t1 represents the deviation between target and actual leverage; and Xi,t1 denotes a vector of firm characteristics. If ICM activity is related to deviations from target leverage, we anticipate an inverse relation between ICM activity and leverage. In other words, given that a negative deviation would suggest that an insurer was overleveraged, the insurer should purchase more (i.e., cede more) reinsurance. We account for firmspecific factors that are likely to relate to ICM activity in the model. These factors include size, diversification, and organizational form. Below we discuss the relation between ICM activity and target capital structure deviations for each of the variables used in our analysis. 5.6. Internal capital market variables We proxy for ICM activity using regulatory data reported to the NAIC. Following prior literature that examines ICM transactions in the property–casualty insurance marketplace (e.g., Powell and Sommer, 2007; Powell et al., 2008), we focus on reinsurance transactions as the ICM activity of interest. In particular, ICM transactions are calculated as affiliate reinsurance purchased (ceded) minus affiliate reinsurance sold (assumed), divided by total premiums earned. A higher value for this ratio suggests firms are purchasing greater amounts of reinsurance (and effectively reducing liabilities), while a lower value suggests that firms are purchasing less reinsurance (i.e., ceding lower levels of liabilities). To examine the link between deviations from target capital structure and ICM transactions, we include a number of independent variables hypothesized to be related to the level of ICM activity. 5.6.1. Target leverage deviations Hypothesis 2 posits a relation between deviations from target leverage and the use of ICMs, where over-levered insurers should purchase more reinsurance from affiliates or sell less reinsurance to reduce leverage and under-levered insurers should increase leverage either by selling more reinsurance to affiliates or purchasing less reinsurance from affiliates. As discussed previously, we anticipate an inverse relation between the ICM variable and the deviation between target and actual leverage. 5.6.2. Size Firm size, proxied by the natural logarithm of total insurer assets in year t  1, is included in the model. Houston and James (1998) note that larger, more diversified banks should be able to obtain external capital with greater ease than smaller banks. This suggests a possible negative relation between insurer size and the level of ICM activity. While the size of the firm may be related to the level of ICM activity, Powell et al. (2008) note that the size of the firm relative to the size of the group also may be an important determinant to the use of ICMs. Large firms (relative to their group) should not only face lower external capital costs, but they are also constrained by the fact that their ability to buy reinsurance from

affiliates is limited by the size of the affiliates. As such, we also include a variable to proxy for the relative size of the affiliated insurer, calculated as the ratio of the insurer’s total assets to the group’s total assets. We anticipate a negative relation between a firm’s relative size and ICM activity. 5.6.3. Diversification Line-of-business HHI and geographic HHI are also included in the analysis of ICM activity. Powell and Sommer (2007) argue that firms with less diversified lines of business may insure less-risky insureds, resulting in a reduced need for reinsurance. Because increased concentration may reduce the need for reinsurance, we posit a negative relation between the diversification measures and ICM activity. As the HHI approaches a value of 1, the firm becomes less diversified (i.e., more concentrated) and thus a negative relation indicates that more concentrated firms are less reliant on internal sources of capital. 5.6.4. Risk-based capital A higher risk-based capital ratio suggests that a firm is more financially stable and is carrying a greater level of capital. The greater level of capital implies that the insurer has greater capacity to take on risk, and is thus able to assume more risk from affiliates when needed. We predict a negative relation between an insurer’s RBC ratio and the use of ICMs. 5.6.5. Organizational form Binary variables denoting organizational structure are also included in the model. We include one binary variable denoting those insurers that are of the mutual organizational form and one binary variable denoting those insurers that are considered neither stock nor mutual firms. Lamm-Tennant and Starks (1993) argue that mutual insurers may take on less risk because policyholders are the owners. If mutuals have less risk than stock insurers, then mutuals may have less of a need to cede reinsurance than do their stock counterparts. We anticipate a negative relation between ICM activity and the mutual organizational form. Table 1 provides definitions, pooled summary statistics, and an overview of expected relationships for each of our variables. The summary statistics indicate insurers in the sample maintained a ratio of total liabilities to total surplus of 1.83 on average. The summary statistics also suggest that most insurers are fairly diversified across both lines of business and geographic location, although over one half of net premiums written are concentrated in the commercial lines of business. Finally, over 80% of the sample is of the stock organizational form and the average affiliate is approximately one fifth the size of the group. 6. Empirical results and discussion 6.1. Existence of target capital structure We first estimate Eq. (3) using OLS to determine an appropriate bound for the system GMM coefficient estimates and then use the system GMM methodology to determine if affiliated property– casualty insurers have target leverage ratios (Bond, 2002). The results are presented in Table 2.20 20 The lagged dependent variable (LS), return on assets (ROA), and the change in net premiums written (ChngNPW) are treated as endogenous variables. The following variables are used to create instruments for the three endogenous variables: the natural logarithm of total insurer assets; the proportion of business in long-tail lines; line of business HHI; geographic HHI; insurer-specific RBC ratios; the proportion of business written in commercial lines; binary variables for the mutual organizational form; binary variables for ‘‘other’’ organizational forms; and year binary variables. All variables used for the creation of instruments are lagged one period.

1035

S.G. Fier et al. / Journal of Banking & Finance 37 (2013) 1029–1039 Table 1 Variable definitions, summary statistics and expected relations. Variables

Definition

LS ICM

Ratio of liabilities to surplus Ratio of affiliate reinsurance purchased (ceded) minus affiliate reinsurance sold (assumed), divided by total premiums earned Natural logarithm of total assets Ratio of total affiliate assets to total group assets Return on assets, calculated as the ratio of net income to total assets Ratio of insurance reserves to losses incurred Line of business Herfindahl–Hirschman Index Geographic Herfindahl–Hirschman Index Risk-based capital ratio Change in net premiums written Proportion of net premiums written in commercial lines of business Binary variable denoting mutual insurers Binary variable denoting insurers not classified as stocks or mutuals

Size RelSize ROA Longtail LOB GEO RBC ChngNPW Comm Mutual Other

Mean

Median

Std. dev.

1.8268 0.5698

1.7002 0.0000

1.0970 2.2128

18.9758 0.2041 0.0238 3.2146 0.4450 0.4655 1284.39 0.1343 0.5548 0.1467 0.0385

18.8986 0.0685 0.0268 2.4314 0.3731 0.3501 743.81 0.0412 0.6028 0.0000 0.0000

1.7380 0.2640 0.0440 3.1564 0.2598 0.3687 2725.45 0.6548 0.3735 0.3538 0.1924

Expected Sign: Leverage

Expected Sign: ICM

+ 

 

+   +    +/  +/ +/

          +/

This table reports variable definitions and summary statistics for both independent and dependent variables used in Eqs. (3) and (5). Additionally, this table reports expected relationships between the independent variables and the various dependent variables. Dependent and independent variables (with the exception of RBC) are winsorized at the 1st and 99th percentiles. The RBC variable is not winsorized in an effort to maintain variation within the variable. The column titled ‘‘Expected Sign: Leverage’’ provides the posited relation between a given variable and current period leverage, where a ‘‘+’’ indicates an expected positive relationship, ‘‘’’ denotes an expected negative relationship, and ‘‘’’ denotes a variable that is not included in the given model. The column titled ‘‘Expected Sign: ICM’’ provides the posited relation between a given variable and current period ICM transactions. In order to ensure that firms in the sample have potential access to ICMs, all single entity (unaffiliated) insurers are removed.

Using liabilities to surplus as our proxy for leverage, we find a positive and statistically significant coefficient associated with the lagged leverage variables.21 Consistent with Hypothesis 1, evidence from Table 2 suggests that affiliated insurers have target leverage ratios.22 The results show the coefficient in the system GMM model on the lagged leverage variable to be greater than zero (0.8069), less than unity, and positive and significant.23 The existence of a target capital structure in the property–casualty insurance marketplace provides further evidence in support of Cummins and Doherty (2002), Harrington and Niehaus (2002), De Haan and Kakes (2010), and Cheng and Weiss (2012).24 To verify that the coefficient estimate is appropriate, we compare the estimate obtained using system GMM with the estimates obtained when using OLS. Using OLS with yearly fixed effects, we report a coefficient estimate of 0.8286, which is larger than the

21 Similar results (not reported in Table 2) are obtained using PS and LA as dependent variables. 22 The Arellano–Bond test for autocorrelation in first differences suggests the existence of negative first order serial correlation in the system GMM model, which is expected (Roodman, 2009). The results suggest a lack of second-order serial correlation in the system GMM model, indicating the serial correlation is not present in the above models. While not reported, we also check for serial correlation in higher orders (up to five orders) and find no evidence of serial correlation in the higher levels. 23 Some insurers in the sample only have a limited number of years in the dataset. If an insurer is only present in the dataset for a short period of time, it is difficult to obtain an unbiased estimate of the coefficient on the lagged leverage variables. We check the robustness of the results by re-estimating each of the models presented in Table 2 after excluding insurers that do not have at least 5 years of data during the sample period. This reduces the sample from 10,210 observations to 9878 observations. Using the reduced sample, the coefficient on the lagged dependent variable has a value of 0.8021, which is very similar to the 0.8069 estimate presented in Table 2. Results are stable across all other independent variables when constraining the model to only those firms with at least 5 years of data. 24 The results in Table 2 are relatively consistent with prior studies. Specifically, a coefficient of 0.8069 suggests a speed of adjustment (SOA) of roughly 19%. Using a sample of Dutch property–casualty insurers from 1995 to 2005, De Haan and Kakes (2010) find that insurers adjust toward target capital levels and that insurers reduce the gap between actual capital ratios and target capital ratios by slightly less than one-third. The authors also report insurers with capital ratios below the target increase the ratios more quickly than firms that are close to their targets. Using a sample of unregulated firms (i.e., not including financials or utilities), Flannery and Rangan (2006) report an SOA of 34.4%. Huang and Ritter (2009) present SOAs that are similar to those presented in this study, with an SOA of 17% annually for book leverage and 23% annually for market leverage.

estimate obtained using system GMM. This value should represent the upper bound for an appropriate coefficient estimate. The system GMM estimate reported above does fall below this value. Similarly, using OLS with both year- and firm-specific fixed effects, we report a coefficient estimate of 0.5206. Again, this value should represent the lower bound for an appropriate estimate. The system GMM estimate of 0.8069 is above that value. Given that the system GMM coefficient estimate falls within this range, the parameter estimate seems reasonable. With respect to the additional control variables in the model, the results for the firm-level variables are generally as anticipated. First, the results suggest a positive relation between firm size and leverage. This finding supports the notion that larger insurers may maintain higher leverage ratios because less capital is required to reach a given level of insolvency risk, relative to smaller insurers. Second, we report a negative relation between insurer RBC ratios and the level of leverage, supporting the expectation of an inverse relation as firms with a greater RBC ratio will have greater capital holdings. Third, we find that positive changes in net premiums written are associated with increases in firm leverage. While growth opportunities may motivate a firm to hold greater levels of capital which should effectively reduce leverage, this finding may be attributable to the negative impact that increased business can initially have on a firm’s surplus. Although many of the results are consistent with our expectations, there are some differences. First, we find a negative relation between long-tail lines of business and leverage. As discussed by Cummins and Nini (2002), agency theory predicts that firms with long-tail lines of business will maintain greater levels of leverage to reduce the potential for managerial opportunism. Our finding contradicts this expectation of a positive relation between the two variables. We posit that such a relation may be the result of management attempting to reduce the cost of external capital. As discussed by both Cummins and Nini (2002) and De Haan and Kakes (2010), insurers writing long-tail lines of business are associated with increased levels of information asymmetry and generally have a reduced capacity to raise internal capital compared to firms underwriting shorter-tailed lines of business. Given that these firms would thus rely on external capital to a greater degree than other firms, and given that the cost of external capital could be affected by asymmetric information as well as the financial

1036

S.G. Fier et al. / Journal of Banking & Finance 37 (2013) 1029–1039

Table 2 Partial adjustment model results. Variables

OLS

Std. error ***

OLS

Std. error ***

LS Size ROA Longtail LOB GEO RBC ChngNPW Comm Mutual Other Constant

0.8286 0.0126*** 0.5714*** 0.0094*** 0.0239 0.0024 0.0000*** 0.0262*** 0.0966*** 0.0858*** 0.0079 0.1401**

Yearly FE Firm FE F statistic R-squared AR(1) AR(2) Sargan Hansen Observations

Yes No 614.40*** 0.7173

Yes Yes 95.35*** 0.6694

10,210

10,210

0.0063 0.0042 0.1513 0.0021 0.0252 0.0190 0.0000 0.0069 0.0182 0.0171 0.0307 0.0600

0.5206 0.0335 1.0950*** 0.0115*** 0.1235 0.1693** 0.0000*** 0.0002 0.0360 0.2105 0.2158 0.5497

System GMM

Std. error

***

0.0179 0.0297 0.2375 0.0030 0.0831 0.0839 0.0000 0.0109 0.1049 0.2703 0.2566 0.3543

0.8069 0.0232** 0.2412 0.0082** 0.0760 0.0262 0.0000** 0.3896** 0.0874*** 0.0411 0.0122 0.0275

0.0530 0.0101 1.0663 0.0038 0.0512 0.0246 0.0000 0.1584 0.0298 0.0427 0.0456 0.1000

Yes No 174.40*** 0.642 6.01*** 1.40 7.74 3.42 10,210

This table reports the results obtained from estimating Eq. (3), Li;t ¼ kbX i;t1 þ di;t . Robust standard errors are presented with OLS coefficients. Dependent variables are taken at time t and all independent variables are taken at time t  1 (i.e., lagged one period). All independent variables are winsorized at the 1st and 99th percentiles (with the exception of RBC). The lagged dependent variable (LS), return on assets (ROA), and the change in net premiums written (ChngNPW) are treated as endogenous in the model. AR(1) and AR(2) report the Arellano–Bond test for AR(1) and AR(2) in first differences. The Arellano–Bond test for autocorrelation in first differences suggests the existence of negative first order serial correlation in the system GMM model, which is expected (Roodman, 2009). The results suggest a lack of second-order serial correlation in the system GMM model, indicating the serial correlation is not present in the above models. We use the Sargan test and the Hansen test to verify the identification of the system GMM model. Each test indicates that the instrumental variables employed in the model are identified and valid, as neither the Sargan nor the Hansen statistics are significant at the 10% levels of statistical significance.  Statistical significance at the 10% level. ** Statistical significance at the 5% level. *** Statistical significance at the 1% level.

Table 3 Realized leverage, target leverage, and deviations from the target. Year

Pooled sample – all firms (1) LS

1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

1.7344 1.7020 1.8046 2.0464 2.1599 2.0120 1.9301 1.8653 1.7073 1.6152 1.7352 1.6117

Target 1.7027 1.6866 1.8209 2.1000 2.2056 1.9872 1.9275 1.8693 1.6840 1.6121 1.7766 1.6032

Deviation 0.0317 0.0154 0.0163 0.0537* 0.0457 0.0247 0.0026 0.0040 0.0233 0.0031 0.0414* 0.0085

Decile 1 – underleveraged firms (2)

Decile 10 – overleveraged firms (3)

LS

LS

Target

Deviation

3.2835 3.1326 3.7405 4.0261 4.0741 3.5632 3.5572 3.6685 3.2025 2.9875 3.5721 3.3037

2.0071 1.9593 2.4067 2.5516 2.7291 2.4407 2.2825 2.3475 2.0446 1.9099 2.4030 1.9361

1.2764*** 1.1734*** 1.3338*** 1.4744*** 1.3450*** 1.1225*** 1.2747*** 1.3209*** 1.1579*** 1.0776*** 1.1691*** 1.3676***

1.2950 1.3659 1.3487 1.3239 1.5400 1.6457 1.6425 1.6040 1.4635 1.4234 1.4187 1.5277

Target 2.5777 2.7804 2.6691 2.3088 2.7895 3.0372 2.8300 2.9758 2.6544 2.7574 2.4060 2.8812

Deviation ***

1.2827 1.4145*** 1.3204*** 0.9849*** 1.2495*** 1.3916*** 1.1875*** 1.3717*** 1.1909*** 1.3340*** 0.9873*** 1.3535***

This table reports annual realized and target leverage ratios, as well as deviations from the target leverage ratio, for three groupings: (1) all (pooled) insurers in the sample, (2) the most underleveraged insurers in the sample, and (3) the most overleveraged insurers in the sample. Target leverage is calculated as Li;t ¼ 1k ½Li;t  Li;t1  di;t  þ Li;t1 , using estimates obtained from the system GMM model estimated in Table 2. The pooled sample in Column 1 includes all affiliated property–casualty insurers in the sample. Column 2 includes only those insurers that are the most under-levered (those assigned to the first decile), while Column 3 includes only those insurers that are the most overlevered (those assigned to the tenth decile). LS represents the mean liabilities to surplus ratio; Target represents the mean calculated target leverage using Eq. (4); Deviation represents the difference between the realized leverage ratio and the estimated target in a given year. LS is winsorized at the 1st and 99th percentiles. On average, approximately 396 firms are classified as ‘‘overleveraged’’ annually, while approximately 455 firms are classified as ‘‘underleveraged’’ annually. Standard t-tests are conducted to determine if reported differences are statistically different. * Statistical significance at the 10% level. ⁄⁄ Statistical significance at the 5% level. *** Statistical significance at the 1% level.

well-being of the firm (i.e., leverage), we contend that management may actively attempt to reduce the cost of external capital by maintaining lower leverage ratios than firms that write less long-tail business. Similarly, while we anticipated a negative relation between the proportion of commercial lines of business and leverage, the results indicate a positive relation exists.

6.2. Target capital structure and ICM activity The results from Table 2 indicate that affiliated property–casualty insurance companies do have target leverage ratios, given the relation between the current year and prior year leverage. Given the existence of target leverage ratios, we next examine whether

S.G. Fier et al. / Journal of Banking & Finance 37 (2013) 1029–1039

1037

Fig. 1. Liabilities to surplus at time t and time t + 1 sorted by quartiles.

Fig. 2. Internal capital market activity (reinsurance purchases via affiliates) at time t and time t + 1 sorted by quartiles.

a relation exists between ICM activity and an insurer’s deviation from the target leverage ratio. Using Eq. (4), we estimate target leverage for all affiliated insurers in our sample. Realized leverage, estimated target leverage, and the difference between target and realized leverage on a yearly basis are presented in Column 1 of Table 3. The average realized leverage ratio for affiliated insurers over the sample period is 1.826 while the average estimated target ratio is 1.831 (LS).25 The realized leverage ratios appear very close to the target leverage ratios using the pooled data, but the results presented in Column 1 suggest that yearly variation in leverage does exist. The realized leverage ratios and the (estimated) target leverage

25 The average realized leverage ratio for affiliated insurers over the sample using PS is 1.05, and 0.58 using LA. The average estimated target ratios are 1.07 (PS) and 0.59 (LA). Similar to the results using LS, the pooled results suggest that firms are typically close to their target leverage ratios.

ratios indicate that, in general, realized insurer leverage is typically slightly above target leverage rather than below the target. To examine the full effect of these deviations on ICM activity, we assign insurers into deciles based on the size of the insurer’s deviation from the estimated target leverage. Insurers are thus placed into one of ten categories, where the first decile contains those insurers that are the most under-levered (i.e., realized leverage is less than target leverage) and the tenth decile contains those insurers that are the most over-levered (i.e., realized leverage is greater than target leverage). Realized leverage ratios, estimated target leverage ratios, and differences between the realized and target ratios for the first and tenth deciles are reported in Columns 2 and 3 in Table 3. The results presented in Columns 2 and 3 of Table 3 show that while Column 1 suggests affiliated insurers are (on average) close to their estimated targets, there is significant dispersion across insurers with respect to deviations from target leverage. Column 2 and Column 3 consist of those insurers who are furthest from

1038

S.G. Fier et al. / Journal of Banking & Finance 37 (2013) 1029–1039

their respective target leverage ratios. In particular, Column 2 consists of those affiliated insurers that are underleveraged and Column 3 consists of affiliated insurers that are overleveraged. While Column 1 indicates relatively small deviations from the target (an average deviation of 0.004 for LS), Columns 2 and 3 suggests that some insurers have much greater deviations. Specifically, we find an average deviation of 1.239 for the most underleveraged insurers and an average deviation of 1.264 for the most overleveraged insurers.26 We also find that these deviations are statistically significant, both for those firms in the first decile and those in the tenth decile. By construction, those insurers in deciles two through nine are closer to the estimated targets. Thus far, the results suggest that (1) affiliated property–casualty insurers have target leverage ratios and (2) affiliated insurers differ with respect to the size of the deviation from target leverage, with some insurers having much larger deviations than other insurers. Given these deviations, a number of options are available for insurers to adjust toward a given target leverage ratio. One method available to affiliated insurers is the use of ICMs. This option should be particularly preferable if the cost of using internal capital is less than the cost of external capital. Because ICMs are available to affiliated insurers, we use the estimated targets yielded from Eq. (4) in order to examine whether ICM activity differs between those insurers that exhibit greater deviations from target leverage than those insurers with smaller deviations. Prior to empirically testing the relation between ICM activity and deviations from target capital structure, we provide a visual examination of changes in both the ratio of liabilities to surplus (Fig. 1) and ICM activity (Fig. 2) from period t to t + 1. Fig. 1 shows the average liabilities to surplus ratio for insurers assigned to quartiles based on the size of the insurer’s deviation from target leverage. Insurers in the first quartile are the most underleveraged at time t, and insurers in the fourth quartile are the most overleveraged at time t. The figure shows that as insurers move from time t to time t + 1, the most underleveraged firms increase leverage from approximately 1.34 to 1.44 and the most overleveraged firms reduce leverage from approximately 2.82 to 2.70.27 While movement exists in leverage ratios for those firms in quartiles 2 and 3, the movement is less dramatic. These adjustments are expected as firms move toward a given target leverage. Fig. 2 presents firm ICM activity at time t and t + 1. Specifically, firms are assigned into quartiles based on deviations from target leverage, where the most underleveraged firms are placed into the first quartile and the most overleveraged firms are placed into the fourth quartile. Similar to what was expected in Fig. 1, one would anticipate that those firms that are most underleveraged at time t (Quartile 1) would purchase less reinsurance at time t + 1 (resulting in a reduced ICM value) while overleveraged firms at time t would purchase more reinsurance at time t + 1 (resulting in an increased ICM value). Fig. 2 indicates that firms in Quartile 1 reduce ICM activity from time t to time t + 1, while the most overleveraged firms increase ICM activity via the ceding of reinsurance. The visual evidence presented in Figs. 1 and 2 provide further support in favor of both targeting behavior on behalf of affiliated

26 For PS, we find an average deviation of 0.609 for the most underleveraged insurers (i.e., those insurers in Decile 1) and an average deviation of 0.737 for the most overleveraged insurers (i.e., those insurers in Decile 10). For LA, we find an average deviation of 0.118 for underleveraged insurers and an average deviation of 0.123 for overleveraged insurers. Again, these results imply that while insurers are (on average) close to their targets (using pooled data), some insurers exhibit fairly large deviations from the target. 27 Standard t-tests were conducted to determine if the change in leverage from time t to t + 1 for Quartile 1 (and Quartile 4) was statistically and significantly different than the change in leverage for all other quartiles. Results support a statistically significant difference in the size of the change from one period to the next period for the most over-leveraged and the most under-leveraged firms in the sample.

Table 4 Internal capital markets and target leverage. Variable

Coefficient ***

Deviation Size RelSize RBC LOB GEO Mutual Other Constant

0.0241 0.3020*** 0.0645*** 0.0000*** 0.1562*** 0.5128*** 0.0330** 0.0510 4.3392***

Prob > Chi2 Pseudo-R2 Observations

0.0000 0.8217 8358

Std. error 0.0013 0.0048 0.0098 0.0000 0.0249 0.0254 0.0162 0.0545 0.0988

This table reports the results obtained from the regression of ICM activity on deviations from target leverage and other measures associated with the use of affiliate reinsurance. The following regression is estimated to obtain the results presented in this table: ICM i;t ¼ b0 þ b1 ðLi;t1  Li;t1 Þ þ bX i;t1 þ ei;t . The model includes year and firm fixed effects and is estimated using FGLS to account for both autocorrelation and heteroskedasticity. All independent variables are lagged 1 year. Dependent and independent variables (with the exception of RBC) are winsorized at the 1st and 99th percentiles. ‘‘Deviation’’ represents the calculated deviation between target leverage and realized leverage. The dependent variable (ICM) is internal capital market activity, which is proxied by the ratio of affiliate reinsurance purchased (ceded) minus affiliate reinsurance sold (assumed), divided by total premiums earned.  Statistical significance at the 10% level. ** Statistical significance at the 5% level. *** Statistical significance at the 1% level.

insurers and the relation between ICM activity and underwriting leverage. With evidence suggesting a target leverage ratio and differences in ICM activity across insurers based on deviations from the target, we use Eq. (5) to examine whether ICM activity among affiliated property–casualty insurers varies based on deviations from target leverage. We use feasible generalized least squares (FGLS) to estimate Eq. (5) in order to account for both heteroskedasticity and autocorrelation in the model and we include year and firm fixed effects to account for year- and firm-specific variation.28 The results from this estimation are reported in Table 4. The results presented in Table 4 provide evidence in support of a relation between deviations from target leverage and ICM activity. Focusing first on the independent variable of primary interest (Deviation), we report a negative relation between ICM activity and deviations from target leverage. The negative relation suggests that insurers that are over-leveraged relative to their target leverage purchase (cede) more reinsurance. This finding is consistent with Hypothesis 2 and supports the notion that property–casualty insurers utilize reinsurance for the purpose of adjusting toward target leverage.29 The model also provides support for other previously identified relationships between insurer-specific characteristics and the use of ICMs. Specifically, the negative relation between ICM activity and the Size variable is as expected and consistent with prior literature (e.g., Powell et al., 2008). We also report an inverse relation between ICM activity and insurer concentration (LOB and GEO). This finding is consistent with the argument that firms writing more concentrated business will also write less-risky business and ultimately reduce the firm’s need to cede reinsurance. The coefficient on the Mutual binary variable is negative and significant, suggesting that the mutual organizational form uses ICMs 28 The Hausman test indicates that fixed effects are preferred over the use of random effects. We use the Breusch–Pagan/Cook–Weisberg test for heteroskedasticity and Wooldridge’s (2002) test for serial correlation. Results suggest the existence of heteroskedasticity and autocorrelation. 29 The results are consistent when estimating the models with alternative measures of leverage (i.e., PS and LA).

S.G. Fier et al. / Journal of Banking & Finance 37 (2013) 1029–1039

less than other organizational forms. Agency theory contends that mutual insurers should write less risky business, which would lead to a reduced reliance on the ceding of reinsurance. The results on other variables differ from our a priori expectations. Specifically, we find a positive relationship between insurer RBC ratios, relative insurer size, and ICM activity, which is inconsistent with our initial expectations. 7. Conclusions Prior literature has not, to our knowledge, jointly examined the relation between deviations from target capital structure and internal capital market activity. Data from the property–casualty insurance industry allow us to conduct such an examination and we draw on recent empirical studies to examine the link between deviations from target capital structure and ICM activity. The results suggest that affiliated property–casualty insurers have target leverage ratios and that insurer use of internal capital markets varies depending on deviations from target leverage. We provide evidence that, in their purchases of reinsurance from affiliates via internal capital markets, over-leveraged insurers tend to purchase greater levels of reinsurance, effectively reducing leverage, while under-leveraged insurers tend to purchase less reinsurance. These findings indicate that insurers use ICMs in an effort to make partial adjustments toward target capital structure. The results of this study are important for two primary reasons. First, the results provide additional support for the existence of target capital structures in the property–casualty insurance industry. Prior literature has suggested that insurers may have target capital structures, even while they operate within a regulated industry, and these results support this hypothesis. Secondly, and most importantly, the results are the first to show a relation between deviations from target capital structure and ICM activity. Given that internal capital is less costly (and more easily obtainable) than external capital, one would anticipate that firms should use ICMs (as one source of capital) when making adjustments to capital structure. The findings provide a greater understanding of capital movements within groups and how group financial transactions potentially benefit the firm and various stakeholders. Acknowledgements The authors would like to gratefully acknowledge the helpful comments of the editor, an anonymous referee, Patricia Born, Pamela Coats, Cassandra Cole, Randy Dumm, and Andre Liebenberg. Additionally, the authors also benefited from the feedback provided by the participants of the 2010 Southern Risk and Insurance Association annual meeting. References Ang, J.S., Chua, J.H., McConnell, J.J., 1982. The administrative costs of corporate bankruptcy: a note. Journal of Finance 37, 219–226. Antoniou, A., Guney, Y., Paudyal, K., 2008. The determinants of capital structure: capital market-oriented versus bank-oriented institutions. Journal of Financial and Quantitative Analysis 43, 59–92. Arellano, M., Bond, S., 1991. Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies 58, 277–297. Arellano, M., Bover, O., 1995. Another look at the instrumental variable estimation of error-components models. Journal of Econometrics 68, 29–51. Berger, L.A., Cummins, J.D., Tennyson, S., 1992. Reinsurance and the liability insurance crisis. Journal of Risk and Uncertainty 5, 253–272. Blundell, R., Bond, S., 1998. Initial conditions and moment restrictions in dynamic panel models. Journal of Econometrics 87, 115–143. Bond, S.R., 2002. Dynamic panel data models: a guide to micro data methods and practice. Portuguese Economic Journal 1, 141–162. Browne, M.J., Carson, J.M., Hoyt, R.E., 1999. Economic and market predictors of insolvencies in the life-health insurance industry. Journal of Risk and Insurance 66, 643–659.

1039

Cheng, J., Weiss, M.A., 2012. Capital structure in the property–liability insurance industry: tests of the tradeoff and pecking order theories. Journal of Insurance Issues 35, 1–43. Cummins, J.D., Doherty, N.A., 2002. Capitalization of the property–liability insurance industry: overview. Journal of Financial Services Research 21, 5–14. Cummins, J.D., Harrington, S.E., Klein, R., 1995. Insolvency experience, risk-based capital, and prompt corrective action in property–liability insurance. Journal of Banking and Finance 19, 511–527. Cummins, J.D., Lamm-Tennant, J., 1994. Capital structure and the cost of equity capital in the property–liability insurance industry. Insurance: Mathematics and Economics 15, 187–201. Cummins, J.D., Nini, G.P., 2002. Optimal capital utilization by financial firms: evidence from the property–liability insurance industry. Journal of Financial Services Research 21, 15–53. Cummins, J.D., Sommer, D.W., 1996. Capital and risk in property–liability insurance markets. Journal of Banking and Finance 20, 1069–1092. De Haan, L., Kakes, J., 2010. Are non-risk based capital requirements for insurance companies binding? Journal of Banking and Finance 34, 1618–1627. Desai, M.A., Foley, C.F., Hines Jr., J.R., 2004. A multinational perspective on capital structure choice and internal capital markets. Journal of Finance 59, 2451–2487. Flannery, M.J., Rangan, K.P., 2006. Partial adjustment toward target capital structures. Journal of Financial Economics 79, 469–506. Froot, K.A., Stein, J.C., 1998. Risk management, capital budgeting, and capital structure policy for financial institutions: an integrated approach. Journal of Financial Economics 47, 55–82. Gertner, R.H., Scharfstein, D.S., Stein, J.C., 1994. Internal versus external capital markets. The Quarterly Journal of Economics 109, 1211–1230. Harrington, S.E., Niehaus, G., 2002. Capital structure decisions in the insurance industry: stocks versus mutuals. Journal of Financial Services Research 21, 145– 163. Houston, J.F., James, C., 1998. Do bank internal capital markets promote lending? Journal of Banking and Finance 22, 899–918. Hovakimian, A., Opler, T., Titman, S., 2001. The debt-equity choice. Journal of Financial and Quantitative Analysis 36, 1–24. Huang, R., Ritter, J.R., 2009. Testing theories of capital structure and estimating the speed of adjustment. Journal of Financial and Quantitative Analysis 44, 237– 271. Jean-Baptiste, E.L., Santomero, A.M., 2000. The design of private reinsurance contracts. Journal of Financial Intermediation 9, 274–297. Jensen, M.C., 1986. Agency costs of free cash flow, corporate finance, and takeovers. The American Economic Review 76, 323–329. Kayhan, A., Titman, S., 2007. Firms’ histories and their capital structures. Journal of Financial Economics 83, 1–32. Klein, R.W., Phillips, R.D., Shiu, W., 2002. The capital structure of firms subject to price regulation: evidence from the insurance industry. Journal of Financial Services Research 21, 79–100. Lamm-Tennant, J., Starks, L.T., 1993. Stock versus mutual ownership structures: the risk implications. Journal of Business 66, 29–46. Leary, M.T., Roberts, M.R., 2005. Do firms rebalance their capital structure. The Journal of Finance 60, 2575–2619. Lemmon, M.L., Roberts, M.R., Zender, J.F., 2008. Back to the beginning: persistence and the cross-section of corporate capital structure. The Journal of Finance 63, 1575–1608. Matsusaka, J.G., Nanda, V., 2002. Internal capital markets and corporate refocusing. Journal of Financial Intermediation 11, 176–211. Mayers, D., Smith Jr., C.W., 1994. Managerial discretion, regulation, and stock insurer ownership structure. Journal of Risk and Insurance 61, 638–655. Modigliani, F., Miller, M.H., 1958. The cost of capital, corporation finance, and the theory of investment. The American Economic Review 48, 261–297. Myers, S.C., Majluf, N.S., 1984. Corporate financing and investment decisions when firms have information that investors do not have. Journal of Financial Economics 13, 187–221. Powell, L.S., Sommer, D.W., 2007. Internal versus external capital markets in the insurance industry: the role of reinsurance. Journal of Financial Services Research 31, 173–188. Powell, L.S., Sommer, D.W., Eckles, D.L., 2008. The role of internal capital markets in financial intermediaries: evidence from insurer groups. The Journal of Risk and Insurance 75, 439–461. Roodman, D., 2009. How to do xtabond2: an introduction to difference and system GMM in Stata. Stata Journal 9, 86–136. Scharfstein, D.S., Stein, J.C., 2000. The dark side of internal capital markets: divisional rent-seeking and inefficient investment. The Journal of Finance 55, 2537–2564. Shim, J., 2010. Capital-based regulation, portfolio risk and capital determination: empirical evidence from the U.S. property–liability insurers. Journal of Banking and Finance 34, 2450–2461. Staking, K.B., Babbel, D.F., 1995. The relation between capital structure, interest rate sensitivity, and market value in the property–liability insurance industry. Journal of Risk and Insurance 62, 690–718. Strebulaev, I.A., 2007. Do tests of capital structure theory mean what they say? The Journal of Finance 62, 1747–1787. Stein, J.C., 1997. Internal capital markets and the competition for corporate resources. The Journal of Finance 52, 111–133. Warner, J.B., 1977. Bankruptcy costs: some evidence. Journal of Finance 32, 337–347. Wooldridge, J.M., 2002. Econometric Analysis of Cross Section and Panel Data. MIT Press, Cambridge, MA.