Are Internal Capital Markets Ex Post Efficient?

Abstract

Internal capital markets enable conglomerates to allocate capital to segments throughout the enterprise. Prior literature provides evidence that internal capital markets efficiently allocate capital based predominantly on group member prior performance, consistent with the “winner picking” hypothesis. However, existing research has not examined the critical question of how these “winners” perform subsequent to receiving internal capital—that is, do winners keep winning? We extend the literature by providing empirical evidence on whether or not internal capital markets are ex post efficient. We find, in contrast to mean reversion, that winners continue their relatively high performance. Our study contributes to the literature examining the efficiency of internal capital markets and the conglomerate discount, as well as the literature specifically examining capital allocation in financial firms.

ACKNOWLEDGMENTS

The authors acknowledge helpful comments from Pat Brockett (editor), Ken Seng Tan (coeditor), and two anonymous referees. We also thank Patty Born, Cameron Ellis, Stephen Fier, Martin Halek, Rob Hoyt, Brad Karl, Ambrus Kecskés, Andre Liebenberg, David McCarthy, Chuck Nyce, Marc Ragin, participants at the 2017 Western Risk and Insurance Association Conference, participants at the 2019 American Risk and Insurance Association Conference, participants at the 2019 Vietnam Symposium in Banking and Finance, and participants at the 2020 Munich Risk and Insurance Center Winter Workshop for helpful comments.

Notes

1 Warren Buffett wrote, “Berkshire’s value is maximized by our having assembled the (group members) into a single entity. This arrangement allows us to seamlessly and objectively allocate major amounts of capital, eliminate enterprise risks, avoid insularity, fund assets at exceptionally low cost, occasionally take advantage of tax efficiencies, and minimize overhead” (Buffett 2019, 6).

2 Our focus is on property–casualty insurers and their use of internal reinsurance as the primary form of ICM transaction. Though we focus on internal reinsurance, we find evidence that firms utilize different types of ICM transactions to fulfill different corporate goals. Our results for property–casualty insurers based on the two other types of ICM transactions (dividends and internal capital) are consistent with the findings of Niehaus (2018) for life insurers.

3 We thank an anonymous reviewer for suggesting that we consider this “twisted” efficiency in our research.

4 The authors thank an anonymous referee for noting that conglomerates also have disadvantages that lead some firms to divest themselves of some group members, which may help to increase corporate transparency as well as improve access to capital markets. From a modeling standpoint, firms organized as conglomerates face a decision each period whether to continue as a conglomerate or to break up the conglomerate into stand-alone units. Owners must weigh a multitude of costs and benefits when making this decision. Prior studies on the conglomerate ownership structure (e.g., Stulz 1990; Jensen 1986; P. G. Berger and Ofek 1995) suggest that owners must consider agency costs (e.g., it may be more difficult to coordinate across many business units), moral hazard (e.g., individual unit mangers may act in their own self-interest), and information asymmetry (e.g., firms can access less costly internal capital). In cases where firms decide to break up, presumably managers/owners decide that the costs of continuing to operate as a conglomerate outweigh the benefits of breaking up the firm. One of the benefits these firms have chosen to forego is the ability to distribute available internal capital across various group members at the discretion of headquarters. Thus, even if we find evidence that internal capital markets can be value-enhancing, the idea that some groups decide to move away from the conglomerate structure does not by itself imply a different result (i.e., that internal capital markets are not value enhancing). Rather, it simply suggests that in some situations, the value enhancements are not sufficiently high to offset the costs of the conglomerate structure.

5 Hovakimian (2011) observed that firms are able to increase ICM efficiency during recessions. Almeida, Kim, and Kim (2015) noted that firms with favorable growth prospects received internal capital following the 1997 Asian Financial Crisis, and Santioni, Schiantarelli, and Strahan (2020) and Buchuk et al. (2020) observed similar outcomes for group-affiliated firms during the 2008 crisis.

6 Extant studies on the extent of winner picking are all ex ante analyses of internal capital allocation following group member performance (Stein 1997; Powell, Sommer, and Eckles 2008; Niehaus 2018).

7 For example, though Ben-David, Palvia, and Spatt (2017) have firm-level (branch-level) observations for banks, their data do not allow the same detailed view into intra-segment transactions.

8 Capital transfers include stock offerings, bonds, short-term borrowing, reserve account transfers, and surplus notes.

9 Several recent studies have examined the use of “shadow insurance,” which is a type of internal reinsurance transaction (e.g., Koijen and Yogo 2016; Hepfer, Wilde, and Wilson 2020). Though this practice is relevant to our study in that it is a type of internal capital reallocation, it is outside the scope of our article in that it only applies to life insurers, whereas we focus on P-C firms.

10 To put the numbers in Figure 1 in perspective, the U.S. P-C insurance industry produced a total of $582 billion direct premiums written in 2016 across the total 2623 operating insurers and affiliates, when including U.S. territories. Values for total premiums written are before reinsurance transactions.

11 Because reinsurance pooling arrangements are common, we separated pooling agreements from our data to create Figures 1, 2, and 3. We note that even when separating pooling arrangements, reinsurance activity still comprises the bulk of ICM transactions in our data.

12 In our sample, regressions using internal reinsurance are consistent when using total ICM activity.

13 For external reinsurance, other explanations of utilization exist—namely, reinsurer expertise, underwriting stabilization, and catastrophic loss protection. Internal reinsurance is considered a part of the firm’s internal underwriting and investment strategy. We provide a more detailed overview of internal and external reinsurance transactions in Appendix B.

14 In fact, we found that capital and dividend contributions across affiliates are consistent with the models and empirics presented by Niehaus (2018), in that the flow of internal funds is negatively associated with performance.

15 We offer two potential explanations for the significant increase in reinsurance during our sample period. First, Cummins and Trainar (2009) and Marović, Njegomir, and Maksimović (2010) suggested that following a crisis (e.g., the 2008 market crash) the demand for insuranc-linked securities will stagnate, leading to increased reinsurance use. Second, von Dahlen and von Peter (2012) found evidence suggesting that reinsurance demand is correlated with natural disasters, and many costly disasters occurred during our sample period (e.g., Hurricane Sandy). Therefore, it is reasonable to suggest that the combination of the 2008 crisis and costly catastrophic risk exposure is driving the increase in reinsurance in our sample.

16 Though state law and tax variation may alter firm behavior (e.g., Gramlich, Limpaphayom, and Rhee 2004; Markle 2016), state taxes on premiums are retaliatory, limiting the tax advantages of ICMs across state lines. For a more detailed look into the world of insurer taxation and issues therein, see Grace, Sjoquist, and Wheeler (2007) and Neubig and Vlaisavljevich (1992).

17 GEICO, Allstate, and Travelers Insurance saw combined ratios (i.e., the ratio of incurred losses plus loss adjustment expenses to premiums earned) above 100 in personal auto policies for 2016.

18 Specifically we cite the findings of Powell, Sommer, and Eckles (2008) in regard to internal reinsurance allocation, where they found that internal reinsurance is positively associated with prior year performance.

19 The results of our specifications are consistent when transactions are scaled by net premiums written, surplus, and group ICM activity (e.g., firm internal reinsurance as a proportion of total group internal reinsurance).

20 Underwriting income is net of internal reinsurance from losses and loss adjustment expenses.

21 To capture risk-adjusted returns for insurers in our sample, we perform robustness tests replacing net income scaled by assets (ROA) with return on equity (ROE) for capital and dividend models (e.g., Niehaus 2018). Results for all models using ROE as a performance metric are consistent.

22 Considering that variance in state regulations may drive internal capital, dividends, and reinsurance transactions, we include state fixed effects in all models unless otherwise specified. Our results are also consistent when controlling for an insurer’s state of domicile.

23 Although we focus on underwriting returns (which are under the control of individual group members), our results that also include investment returns are consistent with those presented in Powell, Sommer, and Eckles (2008).

24 To address potential endogeneity in using performance (underwriting ROA) as our dependent variable in Equation (2)(2) $$\text{Underwriting}{\text{ROA}}_{i,t}={\alpha }_0+{\alpha }_1\text{Capital}{\text{Transactions}}_{i,t-j}+{\alpha }_2{X}_{i,t-1}+{\upsilon }_{i,t},$$(2) , we perform two tests. First, we include lagged underwriting ROA by using first-differenced generalized method of moments estimators following Arellano and Bond (1991). Second, we use two-stage least squares tests, instrumenting for internal reinsurance by using the number of members in each insurance group. The instrument loads positively and significantly in the first stage (p = .036), and the second stage estimated coefficient on the predicted value for internal reinsurance remains positive and statistically significant. Our results are robust using generalized method of moments first-differenced and two-stage least squares methods.

25 Net internal reinsurance recoverable is recorded in column 13 of National Association of Insurance Commissioners Schedule Y and captures the “net effect on surplus of reinsurance transactions with affiliates.”

26 Prior studies (e.g., Powell, Sommer, and Eckles 2008; Fier, McCullough, and Carson 2013) calculated internal reinsurance as internal reinsurance ceded less assumed. We instead use net reinsurance recoverable from Schedule Y because it is intended to capture the net effect on surplus. Carson et al. (2021) further detailed the advantages of this methodology and found evidence consistent with ICM proxies misstating net capital changes for the firm. Our results are consistent when using traditional internal reinsurance measures.

27 Results are consistent when scaled by group internal reinsurance, net premiums written, and firm surplus, as well as when using gross (ceded) reinsurance.

28 Results are consistent when scaled by group internal reinsurance, net premiums written, and firm surplus, as well as when using gross (ceded) reinsurance.

29 For example, Gertner, Scharfstein, and Stein (1994) argued that segment managers own no residual rights to capital and headquarters is more actively involved in asset management. This is not necessarily the case with internal reinsurance, because contracts are formulated between specific group members based on investments and underwriting gains/losses.

30 Niehaus (2018) used panel data analyses and observed internal capital and dividends flow to life insurers (as opposed to P-C insurers) with lower RBC ratios.

31 For our Performance${}_{i,t}$ variable, we compare firms to the 5th, 10th, 25th, 50th, 75th, 90th, and 95th percentiles within their group and calculate a separate indicator equal to one if the firm had the absolute highest underwriting returns in the group (i.e., the winner).

32 For estimations of Model 3, we focus on underwriting income and internal reinsurance owing to limited evidence of winner picking in capital and dividend transfers.

33 Total assets for our sample period average $1.47 trillion annually, whereas total industry assets average $1.81 trillion.

34 An important consideration regarding winner picking is whether or not managers also raise additional capital for winners in the group outside of the ICM transactions we examine. We test this using methodology similar to Berry-Stölzle, Nini, and Wende (2014). We find that the amount of capital issued is weakly related to winner picking and firms that are winners do raise additional capital separate from internal reinsurance, capital, and dividends.

35 Though our results are consistent when using firm fixed effects, Petersen (2009) noted that firm clustered standard errors are unbiased and result in correctly sized confidence intervals in the face of either permanent or temporary firm effects. Therefore, we present results using firm clustered standard errors.

36 We note that certain risks (e.g., catastrophic) are often transferred and that reinsurance plays a significant role in managing these risks for insurers and insurer groups. We therefore perform robustness checks controlling for catastrophic risks. We use percentage of premiums written in coastal states plus earthquake lines by insurers as a proxy for catastrophic exposure, following Powell and Sommer (2007). Our results are consistent when controlling for these exposures.

37 One important consideration regarding capital allocation of insurers is RBC. RBC ratios provide an early warning signal regarding a firm’s financial strength to state insurance regulators. Accordingly, we perform robustness tests controlling for RBC following Niehaus’s (2018) methodology. Though we find that firms above the 50th percentile for RBC utilize internal reinsurance more than those below the 50th percentile, overall RBC ratio does not significantly impact ICM utilization. Our results are robust to these RBC specifications.

38 Results in Tables 4 though 7 are consistent when including all transactions (internal reinsurance, internal capital, internal dividends, and external reinsurance) in a single regression. We separate each transaction in our outcome tests for clarity.

39 Results in Tables 11 and 12 are consistent when using group fixed effects. However, given the findings of Thompson (2011), we use fixed effects for the lowest N dimensions (year and state) and cluster at the highest N (group) to reduce bias in our estimations.