Consolidation and Systemic Risk in the International Insurance Industry

Transcription

1 Consolidation and Systemic Risk in the International Insurance Industry Gregor N.F. Weiß Juniorprofessur Finance, Technische Universität Dortmund Janina Mühlnickel Juniorprofessur Finance, Technische Universität Dortmund 5th May 2013 Abstract This paper is the first to examine the effects of consolidation in the international insurance industry on the acquirers contribution to systemic risk. We analyze a sample of 409 international domestic and cross-border mergers and find a strong positive relation between consolidation in the insurance industry and moderate systemic risk in the insurance and banking sector. Furthermore, we find strong empirical evidence in support of hypotheses that firm size, leverage and diversification across insurance lines all add to the destabilizing effect of insurance consolidation while geographic diversification is found to contribute to financial stability. Keywords: Financial Crises, Insurance Industry, Systemic Risk, Consolidation, Mergers. JEL Classification Numbers: G22, G01, G34. The authors are grateful for comments from Thomas Berry-Stölzle, Benno Keller, Ray Rees, Carlo Savino, Art Snow, Achim Wambach and participants at the 39th Annual EGRIE seminar and the 47th Annual Meeting of the Western Risk & Insurance Association. Janet Gabrysch provided outstanding research assistance. Support by the Collaborative Research Centers Statistical Modeling of Nonlinear Dynamic Processes (SFB 823) and Economic Risk (SFB 649) of the German Research Foundation (DFG) is gratefully acknowledged. Corresponding author: Otto-Hahn-Str. 6a, D Dortmund, Germany, telephone: , gregor.weiss@tu-dortmund.de. Otto-Hahn-Str. 6a, D Dortmund, Germany, telephone: , janina.muehlnickel@tudortmund.de.

2 Consolidation and Systemic Risk in the International Insurance Industry Abstract This paper is the first to examine the effects of consolidation in the international insurance industry on the acquirers contribution to systemic risk. We analyze a sample of 409 international domestic and cross-border mergers and find a strong positive relation between consolidation in the insurance industry and moderate systemic risk in the insurance and banking sector. Furthermore, we find strong empirical evidence in support of hypotheses that firm size, leverage and diversification across insurance lines all add to the destabilizing effect of insurance consolidation while geographic diversification is found to contribute to financial stability. Keywords: Financial Crises, Insurance Industry, Systemic Risk, Consolidation, Mergers.

3 1 Introduction The recent financial crisis of with the (near-)collapse of American International Group (AIG) and Lehman Brothers has renewed the interest of financial economists and policymakers in the analysis of systemic risks in the global financial sector. 1 The financial economics literature has long been concerned with the consequences of bank runs and defaults of isolated banks. In the seminal work by Diamond and Dybvig (1983), banks are shown to be inherently fragile and subject to runs and there exists a wide consensus among economists nowadays that bank runs could have severe macroeconomic implications. 2 While most of the discussion of systemic risk has concentrated on systemic risks in banking, the fact that the high tide of the recent financial crisis was brought on by the near-collapse of AIG has led economists and insurance regulators to reassess the possibility of systemic risk in insurance. The existence of systemic risks in the global insurance industry has been discussed controversely in the literature. As insurers do not accept customer deposits, they do not face the risk of a sudden shortage in liquidity due to a bank run. In addition, insurers in contrast to banks often rely more strongly on long-term liabilities thus further decreasing their exposure to liquidity risk. Furthermore, insurers are said to be less interconnected than banks resulting in a lower probability of contagion among insurers (see Bell and Keller, 2009). Since the near-collapse of AIG, however, it has been argued that the more similar an insurance company becomes to a bank, the more will it contribute to the systemic risk of the financial system. In other words, non-core and non-insurance activities of an insurer could increase the systemic importance of an insurance company. 3 The same 1 Throughout this paper, we follow the definition of systemic risk by the Group of Ten (2001) which defines systemic financial risk as the probability of a loss of economic value in a substantial portion of the financial system as a result of an exogenous shock to individual financial institutions or the system as a whole. As a consequence of this shock, the disruption of the financial system can also lead to adverse effects on the real economy thus (in part) justifying the need for regulating the financial sector (see also De Bandt and Hartmann (2000) for a rigorous discussion of systemic risk). Note that this definition by the Group of Ten, though originally casted for banks, has also been adopted in the insurance economics literature by, e.g., Cummins and Weiss (2010) and insurance regulators (see International Association of Insurance Supervisors (IAIS), 2009). 2 For some recent studies on systemic risks in banking during the subprime crisis see, e.g., Acharya et al. (2010); IMF (2010); Beltratti and Stulz (2012); Brunnermeier et al. (2012); Fahlenbrach et al. (2012). 3 For banks, Brunnermeier et al. (2012) empirically confirm for a sample of U.S. banks the hypothesis that noncore activities of banks like, e.g., investment banking (and not banking services per se) contributed to systemic risk. 1

4 argument is brought forward by the Geneva Association (2010) which hypothesizes that the insurance industry could contribute to the systemic risk of the financial sector if insurers engage in too extensive derivatives trading on non-insurance balance sheet or in case their short-term financing is mismanaged. This view has been adopted by insurance regulators as well as, for example, the International Association of Insurance Supervisors (IAIS) has based their recently proposed measure of systemically important global insurers on the size of an insurer and the extent in which the company engages in non-insurance activities (see International Association of Insurance Supervisors (IAIS), 2012). While there exist only few studies on the systemic risk of insurers, the possible effects of consolidation in the insurance industry on both the default risk of an insurer as well as on its contribution to the systemic risk of the financial sector have not been analyzed in the literature. A merger between two insurers can have both beneficial as well as adverse effect on the default and subsequently the systemic risk of the merging firms. On the one hand, a merger can reduce competition among the remaining insurers and thus allow insurers to attain monopoly rents. Moreover, the increase in the insured pool should lead to risk reductions and possibly increase profitability. As a consequence, acquirers in insurance mergers should decrease their default as well as their systemic risk following a merger. On the other hand, post-merger integration problems could outweigh the efficiency gains from a merger and lead to a higher default risk of the acquiring insurer. Similar to banks, the increase in size due to a merger could make an insurer too-big-to-fail therefore providing managers with incentives to take on excessive risks. Furthermore, Furfine and Rosen (2011) empirically show that mergers in general increase the default risk of the acquiring firm despite the potential for asset diversification. Even if the default risk of an acquirer could be reduced due to a merger, systemic risk in the insurance sector or even the financial sector as a whole might still increase. While testing the so-called concentration-fragility hypothesis against the competing concentration-stability hypothesis has been a frequent exercise in the empirical banking litera- 2

5 ture, 4 the nexus between M&A activity and systemic risk in the global insurance industry has not been analyzed before. As such, this paper is the first to examine the effects of consolidation in the international insurance industry on the systemic risk of insurers. We analyze a sample of 409 international domestic and cross-border mergers which took place between 1984 and First, we measure the merger-induced changes in the insurers contribution to moderate systemic risk using the Marginal Expected Shortfall (MES) methodology by Acharya et al. (2010) and a novel measure of extreme systemic risk based on the lower tail dependence between a bidder s stock returns and the returns on a market index introduced in Weiß et al. (2012). 5 The key result from our analyses is that mergers in the insurance industry can have a destabilizing effect on both the insurance as well as the banking sector. While our results indicate a strong positive relation between consolidation in the insurance industry and moderate systemic risk in the insurance and banking sector, this effect does not carry over to extreme systemic risk. Whereas insurance mergers thus (expectedly) on average do not lead to immediate crashes of the financial system, they nevertheless coincide with a significant increase in the potential of a system-wide crash. We also investigate the factors driving these merger-related changes in the moderate systemic risk of both the insurance as well as the banking sector. In our cross-sectional analyses on the changes in the bidding insurers MES, we find strong empirical evidence in support of hypotheses that size, leverage and diversification across insurance lines all add to the destabilizing effect of insurance consolidation. At the same time, cross-border mergers are revealed to have a limiting influence on the merger-related changes in moderate systemic risk. In addition, our results confirm 4 The possibly destabilizing effect of concentration in the banking sector has, e.g., been studied by De Nicoló and Kwast (2002), Beck et al. (2006a,b), Schaeck et al. (2009), Uhde and Heimeshoff (2009) and Vallascas and Hagendorff (2011). 5 The recent financial crisis has spawned the development of several competing measures of systemic risk. Alternative measures of systemic risk include the Systemic Risk Indicator by Huang et al. (2011), which is based on credit default swap (CDS) prices, measures of systemic connectedness proposed by Billio et al. (in press), which are based on principal-components analysis and Granger causality and which have also been used in an insurance context by Chen et al. (2012) as well as the CoVaR measure of Adrian and Brunnermeier (2010), which is closely related to the MES methodology used in this study. 3

6 the argument that large life insurers contribute more strongly to the systemic risk of the insurance sector by means of a merger than non-life insurers. The remainder of this article is structured as follows. Section 2 presents the related theoretical and empirical literature on possible systemic risk effects of consolidation in the global insurance industry. Section 3 and 4 discuss the data and the methodology employed in the empirical study, respectively. Section 5 presents the results of our empirical study. The concluding remarks are given in Section 6. 2 Related literature While the effects of bank mergers have been addressed both theoretically and empirically in numerous studies, the impact of mergers in the insurance industry (especially risk effects) has not been analyzed in detail. Both Cummins et al. (1999) and Cummins and Xie (2008), e.g., examine the efficiency effects of mergers in the U.S. insurance industry. They show for the U.S. life as well as property-liability insurance industry that mergers lead to higher efficiency for the acquiring company. These mergerrelated gains in efficiency could, however, also induce a change in the individual default risk. For example, Cummins and Xie (2008) conclude that higher efficiency gains can be attained if the firm is able to obtain economies of scale or scope. After the merger, the acquiring insurance company may benefit from allocating its fixed costs on an increased number of insurance policies. If the merger results in an extended product range, the acquirer may benefit from reducing overall production costs by utilizing additional recources such as customer lists, information technology and customer service capabilities. Additionally, a more diversified product range may lower the default risk, as diversification will facilitate a higher level of adaptability and flexibility in the dynamic insurance industry. Attracting new costumers who prefer to buy several insurance policies from the same company may also lead to economies of scope (see, e.g., Cummins et al., 1999; Akhigbe and Madura, 2001). Due to the merger-related reduction in operating costs and the increase in prof- 4

7 itability, the acquirer may also be able to increase its investments and diversify its asset portfolio. 6 Thus, a merger and possible economies of scale or scope may reduce a firm s individual default risk. Another motive for insurers to merge that could lead to higher efficiency is to gain market power. Being able to set prices over marginal costs increases cash flows and thus leads to higher profit efficiency. Berger et al. (1999) discuss market power as a motive for firms to merge and its consequences. Related studies on banks provide evidence that the wish to increase market power is indeed a reason for consolidation and that firms with market power tend to act more risk-averse (see Berger, 1995). This motive could again imply that the bidder s default risk decreases as a consequence of the merger. Similarly, a reduction of the bidder s default risk is also implied by the law of large numbers as diversification in the (now enlarged) insured pool is ameliorated and losses become more predictable (see Cummins and Xie, 2008). There are also merger motives that may lead to an increase in the insurance company s default risk. Empirical evidence suggests that managers, in response to distorted incentives, sometimes act in their own interests rather than in the firm s interest. In order to safeguard their jobs, they purposely acquire a company that requires their personal skills. Another motive for the merger may be the managers intention to maximize compensation (see Cummins and Weiss, 2004; Cummins and Xie, 2008). In both cases, the merger may lead to an increase in default risk as potential risks may not be managed adequately or could even be neglected. But even in case managers act in the company s interest, they might still overestimate the target s fair value (see Roll, 1986) and underestimate post-merger integration problems (see Cummins and Xie, 2008) due to managerial hubris thus increasing the firm s default risk. The hypotheses mentioned above suggest that default risk effects should differ for small and large acquiring insurers. The empirical evidence on this question, however, is inconclusive. On the one hand, Amel et al. (2004) point out that economies of scale can rather be achieved by 6 In their paper on the consequences of the financial crisis on risk management and supervision in the insurance industry, Eling and Schmeiser (2010) stress the importance of a sufficiently diversified asset portfolio to counter adverse effects during a financial crisis. 5

8 smaller insurers. On the other hand, the merger-related gains in market power are more likely to be achieved by large acquirers (see Berger, 1995) as small acquirers may still be too small after the merger to significantly increase their market power. Additionally, difficulties in post-merger integration are more likely to occur in smaller firms. Consequently, Berger et al. (2000) conclude that large insurers should benefit disproportionately from economies of scope. In contrast, Amel et al. (2004) state that it is unclear whether economies of scope exist at all in the insurance industry. Insurers could also be motivated to merge to increase their geographical, product as well as insurance line diversification in turn again decreasing their default risk. Sheremet and Lucas (2009) show that insurance companies, especially reinsurers, use geographical diversification to absorb large losses. The default risk of an insurer engaging in a cross-border merger could also decrease as the combined firm can shift its priorities between multiple markets (see Elango, 2006). The study by Cummins and Xie (2008) reveals that geographical diversification, on average, is accompanied by higher efficiency, which may also add to a lower default risk. Cross-border mergers, however, are also hypothesized to have a negative impact on the merging firms default risk. Boubakri et al. (2008) conclude in their study that cross-border mergers are value-destroying due to post-merger integration problems, a result which could also imply a higher default risk for the acquirer. Furthermore, Focarelli and Pozzolo (2008) state that the same difficulties arising from geographic distance and economic or cultural integration problems affect both cross-border mergers in the banking sector as well as in the insurance industry. Turning to the empirical evidence in the banking literature, Amihud et al. (2002) show that the risk-reducing effects offset the risk-enhancing effects and thus cross-border mergers are risk-neutral. In a comparative study, Estrella (2001) conjectures that diversification benefits of mergers between banks and insurers are driven by different common factors. It is therefore questionable whether geographical diversification in bank mergers has the same implications for default risk as in the case of insurance mergers. Cross-sectional diversification could also lead to risk reductions for the acquiring company as it enables the firm to operate with a broader range of products. As a result, an insurer should be able to adjust more quickly to shifts of the market. Boubakri et al. (2008), however, 6

9 conclude that property-liability insurers are in the long run more successful if they buy insurers operating in the same sector indicating that the default risk is lower for focusing mergers. Finally, cross-border diversification will usually coincide with an enlargement of the insured pool implying further risk diversification and consequently, a decrease in default risk. In summary, default risk could again increase or decrease as a result of a cross-border merger. In addition to the unknown side-effects of consolidation on the default probabilities of the merging insurers, consolidation could also have a positive or adverse impact on the stability of the financial sector. 7 While the existence of systemic risks in the global insurance industry has been discussed (controversely) to limited extent in the literature, 8 the effects of mergers on the contribution of the merging insurers to systemic risk have not been empirically studied. As systemic risk is in part driven by the default risk of individual insurers, the arguments given above for or against a stabilizing effect of consolidation on default risk should automatically be applicable to systemic risk as well. There are, however, additional hypotheses on the potential relation between consolidation in the insurance sector and systemic risk. On the one hand, authors that support the view that insurance companies do not pose a threat to the financial system in case of default present the argument that insurance companies have to fulfill more rigorous capital requirements than other financial institutions (see, e.g., Harrington, 2009). In this context, De Haan and Kakes (2007) confirm in their empirical study that most insurers even exceed the requirements by far. Moreover, an insurance company run is less likely than a bank run thus eliminating a major cause of systemic relevance of insurers. First, the cancellation of long-term insurance policies usually implies losses because the contract s surrender value is lower than the amount of the premiums paid, whereas closing a bank account only implies losing 7 Note that the empirical literature on consolidation in banking only finds ambiguous evidence on the effects of bank mergers on financial stability. Using the Z-score technique, Uhde and Heimeshoff (2009) present empirical evidence that market concentration and thus consolidation has a negative effect on financial stability. Thus they support the concentration-fragility hypothesis which states that a more concentrated financial market is more prone to a crash. In contrast, Schaeck et al. (2009) and Beck et al. (2006b) find evidence supporting the concentration-stability view which predicates that a market with few large banks and high concentration level is less fragile. 8 The few examples of studies on systemic risk in the insurance industry include the works of Acharya et al. (2009), Cummins and Weiss (2010), Eling and Schmeiser (2010), Baluch et al. (2011) and Bernoth and Pick (2011). 7

10 future interest. Second, obtaining the cancellation repayment usually takes considerably longer than closing a bank account (see Das et al., 2003). Another argument stated against insurance companies posing a risk is the structure of the insurance market. Unlike the highly connected interbank market, the insurance sector is structured hierarchical. Primary insurance companies are linked to reinsurance companies, but interconnections among primary insurance companies or reinsurance companies are seen as relatively limited (see International Association of Insurance Supervisors (IAIS), 2011). Furthermore, in their analysis of systemic risk in the U.S. insurance industry, Cummins and Weiss (2010) state that insurers carrying out core activities are not of systemic importance and that there is less interconnection between insurers and banks than within the insurance industry. In summary, it could be argued that consolidation in insurance should not have any significant impact on the merging insurers contribution to systemic risk as insurers are not of systemic relevance to the rest of the financial sector in the first place. A further argument against a destabilizing effect of insurance mergers can be deduced from the finding by Uhde and Heimeshoff (2009) who point out that the supervision of few larger banks may be easier for the regulator thus leading to a decrease in systemic risk due to consolidation. In an insurance market with few large companies, problems leading to the instability of the financial sector could be recognized earlier by regulators and provisions to prevent system-wide contagion could be more effective. On the other hand, several authors point out that during the last few decades, insurance companies have become more and more involved with banks. 9 In recent years, both trading with credit default swaps (CDS) to underwrite credit risk as well as the usage of alternative risk transfer (ART) arrangements such as catastrophe bonds (CAT-bonds) increased. Baluch et al. (2011) point out that this increased involvement in the financial market made both insurers and reinsurers more predisposed to the recent subprime crisis. Moreover, they state that a failure of a dominating insurance company could weaken the costumers confidence in a related bank which could eventually go 9 See, e.g., Schinasi (2006), Rule (2001) and Baluch et al. (2011). 8

11 bankrupt. Thus an insurance company s failure could not only destabilize the insurance sector but systemic risk could also spill over to the banking system. Also, the bankruptcy of a large reinsurance company could potentially endanger financial stability by leading to the failure of several primary insurance companies possibly destabilizing the financial system as a whole (see Das et al., 2003). Acharya et al. (2009) show that insurance companies can indeed pose a systemic risk. They argue that large insurance companies may be too-interconnected-to-fail, which causes moral hazard problems. Larger insurance companies could be inclined to invest in more risky assets because they feel safeguarded by the government s safety net. 10 Finally, both the Geneva Association (2010) and the International Association of Insurance Supervisors (IAIS) (2011) find that insurers could pose a systemic risk to the financial system in case of excessive engagement in non-core and non-insurance activities. Consequently, an insurance merger could also lead to increases in systemic risk as consolidated institutions could become more interconnected with other insurers and banks. Also, the increase in size of an insurer due to a merger could improve the insurer s access to global financial markets thus increasing the probability of the insurer engaging in non-core and ART activities. Summarizing the hypotheses given above, consolidation in insurance could have unknown sideeffects on financial stability. 3 Data The following section outlines the data used in the empirical study. We first present the data on insurance mergers and then present details on the data used in our cross-sectional regressions. 3.1 Mergers We analyze a global sample of insurance company mergers which we obtain from the Thomson One Banker database. Our study focuses on merger transactions where the acquirers and targets 10 Acharya et al. (2009) cite AIG as the prime example of an insurance company that was too-interconnected-to-fail. 9

12 are insurance companies with SIC codes 6311, 6321, 6331, 6351, 6361 or Thus, we include life insurers, accident and health insurers, fire, marine and casualty insurers, surety insurers, title insurers and other insurance carriers. We explicitly exclude mergers involving any type of banks or insurance agents. The mergers included in our sample were announced and completed between 1984 and All bidding insurance companies are listed at a stock exchange with share price data being available from Thomson Reuters Financial Datastream. We obtain financial accounting data for the merging firms from Thomson Worldscope. For reasons of economic relevance, we follow Weiß et al. (2012) and exclude transactions with a deal value of less than $ 10 million or an acquired stake below 50%. From an initial sample of 1247 completed mergers we therefore lose 100 deals because of the requirement on the deal value and 455 deals because of the requirement on the acquired stake. Moreover, we have to exclude deals due to lacks in share price or accounting data by which we lose another 273 mergers. 11 Additionally, we lose 10 deals due to lacks in data on price indexes. Our final sample consists of 409 transactions. We relate each acquirer in our final sample to one of three regions according to the location of its headquarters: North America, Europe and Asia/Pacific. The region where most acquirers are located is North America, containing the United States, Canada and the Bermudas. The region Europe includes the member countries of the European Union, Switzerland, the United Kingdom and Norway. Transactions completed in Australia, Japan, Malaysia, Singapore and New Zealand are summarized in our region Asia/Pacific. Table I summarizes the mergers regional distribution. insert Table I here The majority of 218 (53.3%) transactions took place in North America, 166 (40.6%) were completed in Europe and 25 (6.1%) mergers occurred in the region Asia/Pacific. For the majority of acquisitions in our sample (82.9%), target and acquirer belong to the same region. In The exclusion of mergers due to missing or incomplete accounting data could create a selection bias in our sample as potentially opaque insurers are systematically omitted from our analysis. As firm opacity is known to be positively related to risk-taking and consequently firm risk, our sample could be biased in favor of less opaque and less risky insurers. To control for such a potential selection bias, we manually check for each merger excluded from our sample due to missing accounting data whether we can find at least one annual report from the respective insurer from a publicly accessible data source. The results show that our sample does not suffer from a selection bias. 10

13 of 409 cases, target companies and bidder companies originate in the same country. Thus, our sample contains 64.8% domestic and 35.2% cross-border mergers. Moreover, the mergers can be categorized by their bidder s main class of insurance. The distribution of the mergers in our sample across the different classes of insurance is presented in Table II. insert Table II here While 267 (65.3%) deals were completed in the line of life insurance and 7 (1.7%) in accident and health insurance, 90 (22.0%) mergers occurred in fire, marine and casualty insurance, 26 (6.4%) in surety insurance, 13 (3.2%) transactions were completed in title insurance and 6 (1.5%) mergers were performed by other insurance carriers. In 284 cases (69.4%), the target company was a life insurer. Moreover, 265 of 409 mergers occurred within the same line of business, thus our sample includes 64.8% focusing and 35.2% cross-sectional transactions. Table III shows the temporal distribution of our sample. insert Table III here The majority of the mergers in our sample was announced and completed between 1993 and During the recent financial crisis between 2007 and 2009, 65 transactions were announced and 70 were completed. Descriptive statistics for the acquiring insurers are presented in Table IV. insert Table IV here Considering all mergers in the sample, the total assets of the acquiring insurance companies are on average about $ 93 billion and the average deal value is $ 0.9 billion. Acquirers operating in North America report on average total assets about $ 39 billion and the average deal value in North America amounts to $ 0.7 billion. In Europe, the acquirers total assets average around $ 173 billion and the average deal value is about $ 1.1 billion. The average total assets of bidders in Europe are thus about four times larger than in North America, whereas the average deal value of European mergers is about 57% higher than the average deal value of mergers in the U.S. 11

14 To answer the question which factors drive the merger-related changes in the systemic risk of insurers, we conduct a cross-sectional analysis of the found risk effects. To this end, we collect a set of idiosyncratic and macroeconomic control variables which we use later in the cross-sectional regressions. 3.2 Deal characteristics We investigate the hypothesis that deal characteristics like the size of the deal play an important role in explaining the risk effects of merging insurers. To control for the deal size, we include the natural logarithm of the deal value in US dollars (LDEALSIZE) in our regressions. The size of the deal may have either a negative or positive impact on systemic risk. 12 On the one hand, larger deals may enable acquirers to diversify their asset portfolio and to compensate risk by a larger insurance portfolio. On the other hand, post-merger integration is more difficult for larger deals and for relatively small bidders. Furthermore, we include a dummy variable in our regressions taking on the value 1 if the merger is a cross-border and 0 if it is a domestic merger (CROSS). We do not have any expectations on the variable s sign because possible diversification benefits could coincide with possible post-merger integration problems (see Cummins and Weiss, 2004). Also, bidders in cross-border transactions usually have to pay higher premiums thus generally leading to a destruction of firm value caused by a cross-border merger (see Rossi and Volpin, 2004). Turning to the lines of business of both the acquirer and the target, we expect diversifying mergers to perform differently than focusing mergers (see Boubakri et al., 2008). Not only should diversifying mergers lead to economies of scope and scale (e.g., through the joint use of customer databases, brand names, etc.), Boubakri et al. (2008) also hypothesize that mergers across insurance lines should decrease the default risk of the acquirer. Empirical evidence by Estrella (2001), however, contradicts this argument as mergers within the property-liability insurance are found to increase default risk less than diversifying mergers. Their finding is in line with the notion that insurers should concentrate on their core lines of business rather than venturing into unknown 12 See also Vallascas and Hagendorff (2011). 12

15 territories through a diversifying merger. 13 To capture the influence of diversifying and focusing mergers on the acquirer s risk effects, we use a dummy variable which equals 1 if the acquirer and the target have different and 0 if they have the same SIC codes (DIVERS). 3.3 Insurer characteristics The second group of variables we use contains the acquiring insurers characteristics. We make use of several key data items from the insurers balance sheets and income statements. The accounting data are taken from the Thomson Worldscope database. In order to capture the impact of profitability on changes in systemic risk, we include the insurers return on assets (ROA) in our regressions. In a study on the capital structure of Dutch insurers, De Haan and Kakes (2007) show that companies with higher profits have a higher solvency margin and therefore a lower risk of insolvency. Uhde and Heimeshoff (2009) present evidence that higher profitability has a positive influence on financial stability in the banking sector. We therefore expect ROA to be negatively associated with systemic risk. We also include the insurers operating efficiency (OPEF) given by the ratio of operating expenses to total assets in our regressions to control for the quality of management. We assume this variable to have a decreasing effect on the insurers contribution to systemic risk. The market to book ratio (MTBR) given by the ratio of market capitalization to total assets is included to control for the hubris of managers. Both Vallascas and Hagendorff (2011) as well as Milidonis and Stathopoulos (2011) predict a negative influence of MTBR on the distance to default of banks and insurance companies, respectively. While Vallascas and Hagendorff (2011) argue that a high MTBR is indicative of distorted managerial incentives driving managers to engage in a risky merger, Milidonis and Stathopoulos (2011) hypothesize that higher values of MTBR could indicate high growth expectations on the part of the investors leading to excessive risk-taking. As a measure for the bidder s size, we include the logarithm of the bidding insurer s total assets (LTOTAL). We expect the coefficient on LTOTAL to influence the acquirers default risk with a negative sign, 13 Berger et al. (2000) also find empirical evidence for both arguments. 13

16 because larger insurance companies have a wider range of different risks insured thus being less prone to suffer from cumulative losses (see Hagendorff et al., 2011). At the same time, larger insurance companies could become too-interconnected-to-fail and thus systemically relevant (see Acharya et al., 2009). For these two reasons, we expect the coefficient on LTOTAL to enter our regressions of the merger-induced changes in systemic risk with an unrestricted sign. Furthermore, we employ the bidder s loss ratio (LOSSRAT) in our regression to capture the risk of the insurance portfolio (see Hagendorff et al., 2011). The loss ratio is defined as the claim and loss expense plus long term insurance reserves divided by premiums earned times 100. We expect LOSSRAT to have an increasing effect on the acquirer s contribution to systemic risk because a riskier portfolio implies a higher default risk. We also include the leverage ratio given by the insurer s common equity divided by total assets (LEVRAT). Vallascas and Hagendorff (2011) argue that managers of highly levered companies have less incentives to invest in risky projects to increase their wages because the company is under the pressure to provide enough liquid assets to cover the payments of interest. Conversely, managers could be inclined to excessively take on risks as leverage forces managers to increase profitability. Thus, LEVRAT is expected to have an unrestricted effect on systemic risk. Finally, Baluch et al. (2011) argue that life insurers are exposed to a higher level of default and systemic risk than other insurers. They argue that life insurers invest in bonds and funds carrying a higher degree of market risk than non-life investments thus making life insurers more susceptible to systemic risk during periods of market turmoil. To test this hypothesis, we include in our regressions the dummy variables ALNL and TLNL taking the value 1 if the acquirer or target is a life insurance company, respectively, and 0 otherwise. 3.4 Macroeconomic control variables As a third group of variables we include macroeconomic and country-level governance variables in our cross-sectional analyses to control for differences in the economic development and political institutions across the countries in our sample. As our first macroeconomic control variable, we employ the real interest rate INTR. In a study 14

17 on the determinants of the performance of insurance companies, Shiu (2004) presents evidence that insurers usually perform better when the interest rate level is higher as it enables insurers to obtain a higher investment income which may offset underwriting losses. We therefore expect INTR to be negatively related to default risk. Similarly, as an increase in the real interest rate affects all companies in a country s insurance sector, we expect the systemic risk of insurers to decrease. 14 Next, we compute the Herfindahl-Hirschman index (HHI) for each country to control for the concentration of the insurance sector. The HHI is defined as the sum of the squared market shares of all insurance companies based in a country and measures the amount of competition in a country s insurance sector. 15 Empirical evidence from the banking literature is ambiguous w.r.t. the expected sign of the coefficient on HHI. While Uhde and Heimeshoff (2009) find evidence that concentration has a negative effect on systemic risk, Beck et al. (2006a) find the opposite result and conclude that concentration is indeed beneficial to financial stability. We therefore have no expectations concerning the sign of the coefficient on HHI in the regressions of systemic risk. We also use the Chinn-Ito index (KAOPEN) to control for the financial openness of each country. A large index value implies low restrictions of cross-border capital transactions (see Chinn and Ito, 2008). Moreover, we include the log of the annual change in inflation rate (INFL) in our regressions. Bernoth and Pick (2011) find evidence that both financial openness as well as the inflation rate have a positive influence on the insurers contribution to systemic risk. Further evidence is found by Shiu (2004) who points out that inflation is negatively connected with insurance and bank performance. To control for adverse macroeconomic shocks that could influence the merger-induced risk effects of insurers, we also include the GDP growth rate (GDPGR) and the unemployment rate (UNEMPL) in our regressions. In order to capture the quality of country-level governance, we employ two indexes describing the political stability (POLSTAB) and the rule of law (ROL) in each acquirer s home country. Additionally, we also include the anti-directors rights index (ADRI) of La Porta et al. (1998) revised 14 In contrast, Demirgüç-Kunt and Detragiache (1998) demonstrate that a crisis is more likely to occur in the banking sector when the real interest rate level is higher. 15 See Uhde and Heimeshoff (2009) for an extensive use of the HHI in a related study on consolidation in banking. 15

18 in Spamann (2010) in our regressions to proxy for the protection of investors. Everything else equal, we would expect external governance to be negatively related with risk effects around insurance mergers. Data for the variables real interest rates, GDP growth rate, unemployment rate and inflation rate are retrieved from the World Bank s World Development Indicator (WDI) Database, while the data on the Chinn-Ito index are provided by the authors on their website. 16 The data on the revised ADRI is taken from the Appendix in Spamann (2010). Ideally, we would like to additionally control for the regulatory environment of the insurers. Unfortunately, such data on global insurance regulation systems are not available. 17 To correct for this possible source of omitted variable bias, we reestimate the multiple regressions controlling for country fixed effects. 18 Due to multicollinearity problems with the country dummy variables, we omit the macroeconomic control variables from our regressions with country fixed effects. Furthermore, the use of a large number of country dummies leads to an incidental parameter problem necessitating the estimation of one-dimensional robust clustered standard errors. 4 Methodology The purpose of this section is to shortly present the methodology we use for capturing the systemic risk effects of insurance mergers. Here, we mostly follow the methodologies outlined in Brunnermeier et al. (2012) and Weiß et al. (2012) who all use comparable methods for estimating systemic risk. 4.1 Marginal Expected Shortfall The first part of our analysis focuses on the effects of insurance company mergers on the moderate systemic risk of both the banking and the insurance sector. As our first measure of an individual insurer s contribution to systemic risk, we use the Marginal Expected Shortfall as proposed 16 See ito/chinn-ito website.htm. 17 This is in contrast to empirical banking, where several studies provide comprehensive data sets on deposit insurance schemes and banking supervision (see Demirgüç-Kunt et al., 2008; Barth et al., 2008). 18 See also Beltratti and Stulz (2012). 16

19 by Acharya et al. (2010). We start by recalling that the Expected Shortfall (ES) of the stocks of insurer i is defined by ES i α E [R i R i q α ] (1) with R i being the insurer s daily return and q α being the α quantile of the return. In other words, the ES measures the average return on those days when losses exceed the value-at-risk. Acharya et al. (2010) show theoretically that the ES methodology can be used to measure the amount that an insurer s equity capital drops below a certain critical level in case of a market crash, i.e., when the aggregate insurers equity is less than a fraction of the aggregate insurers assets. More formally, we define the Systemic Expected Shortfall (SES) of insurer i as SES i E [za i w i 1 W 1 < za], (2) where a i is the insurer s total assets, z is the critical fraction of a i, w i 1 is the insurer s common equity, W 1 n i=1 wi 1 is the aggregate equity and A n i=1 ai is the aggregate assets. 19 The Marginal Expected Shortfall is thereafter defined as the insurer s average net equity return in case of a systemic crisis, more formally MES 5% i E [ w i 1 w i 0 I 5% ] (3) with wi 1 being the ith insurer s net equity return and I w i 5% being the days of a given year on which 0 the market experienced its worst 5% outcomes. 20 The MES can be interpreted as the contribution of a company to the aggregate losses of the respective market. Following the procedure laid out in Acharya et al. (2010), we use the log returns on the insurer s stocks as a proxy for net equity return and estimate the MES for three region-specific indexes. We estimate the MES with respect to an insurance sector index, a banking sector index and the MSCI world index to measure the differen- 19 See also Brunnermeier et al. (2012). 20 Note that the MES results from the SES by scaling the latter by the insurer s initial equity w i 0. Per definition, the MES is thus scaled by the respective insurer s size. 17

20 tial effect of insurance consolidation on different parts of the financial sector and the economy as a whole. 21 For all three indexes, we compute the MES calculating the 5% of days on which the respective index had its worst outcomes in the pre- and post-merger period. The pre-merger period is again the timeframe [-180; -11] days before the merger announcement and the post-merger period covers [+11; +180] days after the deal completion. Finally, after computing each acquirer s pre- and post-merger MES, we test whether the differences ΔMES i 5% MES 5% 5% i;[+11;+180] MES i;[ 180; 11] (4) are on average different from zero. As noted by Acharya et al. (2010), the MES can be seen as a measure of moderate tail risk which regulators can use to predict systemic events in the extreme tail of the insurers and the market s joint distribution. As such, the MES can only be regarded as a measure of an insurer s contribution to moderately bad tail events. In contrast, the next section presents a measure of extreme systemic risk. 4.2 Lower tail dependence In addition to MES, we use a second measure of extreme systemic risk which is based on lower tail dependence (LTD) coefficients. 22 The concept of using the lower tail dependence between financial returns has previously been used, e.g., by Rodriguez (2007) in the measurement of financial contagion. A similar approach based on tail betas is due to De Jonghe (2010) who analyzes the stability of the banking system. Here, we follow the methodology proposed in Weiß et al. (2012) for using LTD coefficients to measure extreme systemic risks between individual firms 21 All indexes are retrieved from the Thomson Reuters Datastream database. The insurance sector index related to each acquirer is either the North America-, Europe-, Australasia-, or Asia Datastream Life Insurance index or the North America-, Europe- or Asia Full Line Insurance index, depending on its home region and its class of insurance. The acquirer s banking sector index is either the North America-, Europe-, Australasia- or Asia Datastream Banks index. 22 See Weiß et al. (2012) for a first use of lower tail dependences coefficients for measuring a bank s contribution to systemic risk. 18

21 and a relevant market index. For two given variables, the coefficient of lower tail dependence is the asymptotic probability that an observation of their joint distribution lies in the distribution s extreme lower tail. In the context of modeling the returns on an individual insurer s stock and a relevant market index, the coefficient of lower tail dependence captures the (asymptotic) probability of both variables experiencing a joint crash. In contrast to the MES, which only captures the contribution of a firm to moderate tail events, the LTD captures the joint probability of an extreme crash. Again, we estimate the LTD with respect to the insurance-, banking- and world market. Then, the coefficient of LTD is defined as LT D 1,2 LT D(X 1, X 2 ) = lim P(X 2 F 1 2 (u) X 1 F 1 1 (u)) (5) u 0 where u (0, 1) is a quantile and F i are the marginal distribution functions. Note that the concept of lower tail dependence is embedded in the theory of copula functions. Consequently, many papers like, e.g., Rodriguez (2007), employ parametric copulas to estimate the LTD. Instead of a parametric modeling, however, we use a nonparametric estimator for the coefficient of LTD. We follow Schmidt and Stadtmüller (2006) and first define the lower tail copula by Λ L (x, y) lim t tc(x/t, y/t). (6) With this definition, the lower tail dependence coefficient can be written as LT D =Λ L (1, 1). (7) To estimate the lower tail dependence coefficient directly, we use a nonparametric estimator proposed by Schmidt and Stadtmüller (2006). 23 Suppose we are given an i.i.d. sample (X (1), Y (1) ),...,(X (m), Y (m) ) of the bivariate random vector (X, Y) with distribution function F which 23 Using a nonparametric approach greatly decreases the model risk inherent in parametric approaches relying on copula goodness-of-fit tests or mixture copulas. 19

22 has marginal distribution functions G, H and a copula C. With the assumption that G and H are continuous, let C m be the unique empirical copula defined by C m (u, v) = F m (G 1 m (u), H 1 m (v)), (u, v) [0; 1]2 (8) where F m, G m, H m are the empirical distributions corresponding to F, G, H. Furthermore, let R ( j) j) m1 and R( m2 with j = 1,...,m denote the rank of the observations X( j) and Y ( j). A nonparametric estimator for the lower tail copula is then given by ˆΛ L;m (x, y) = m k C m ( kx m, ky ) 1 m k m j=1 1 { R ( j) j) m1 kx and R( m2 ky}, (9) where k {1,...,m} is chosen by means of a plateau-finding algorithm. We then compute ˆΛ L;m (1, 1) to estimate the lower tail dependence coefficients. As the bivariate sample is assumed to be i.i.d., the original data are first filtered by use of GARCH(1,1) models with t-distributed innovations. We then estimate the coefficients of lower tail dependence from the standardized residuals from the fitted GARCH(1,1) models. In order to estimate the merger-related change in the bidder s contribution to systemic risk, we compare the LT D between the acquiring company s stock returns and the returns on the respective region-specific index in the pre-and post-merger period. Thus the change in the LT D is defined by ΔLT D i LT D i;[+11;+180] LT D i;[ 180; 11], (10) with LT D i;[ 180; 11] and LT D i;[+11;+180] being the lower tail dependence between insurer i and the relevant insurance sector index, banking sector index or MSCI world index in the pre-and post-merger period, respectively. 20

23 5 Results In this section we discuss the results of our analysis of merger-related changes in moderate and extreme systemic risk. We aim to answer the question whether insurance mergers in general lead to a higher contribution of the acquirer to systemic risk. Furthermore, we identify the factors driving these changes in moderate and extreme systemic risk. 5.1 Moderate and extreme systemic risk effects In order to measure changes in systemic risk, we compute the MES as well as the LT D for the acquiring insurance companies with respect to the three different indexes. The results for the MES are presented in Table V. insert Table V here The table reports the values in the pre- and post-merger periods as well as the change in MES and its statistical significance for the full sample and the regional subsamples with respect to the insurance sector index, the banking sector index and the MSCI world index. The results in Table V present strong evidence supportive of a destabilizing effect of insurance mergers. For all three indexes we analyze, the MES of the acquiring insurer increases significantly at the 1% level. While the MES w.r.t. the insurance sector index only increases significantly for European insurers, the changes in the MES w.r.t. the banking and the global stock index increase significantly for mergers in North America and Europe. While the changes are all statistically significant, they are also economically significant. For example, the average acquiring insurer lost 1.8% on its stock on those days the 5% worst days of the insurance sector in the (relatively short period of) 180 days preceding the merger. After the merger, the respective MES increases to 2.1%. Comparing the changes in MES with regard to the different indexes, we can see that the results for the insurance sector index are similar to those obtained for the banking sector index. This might indicate that the interlinkages among insurance companies are as strong as the interlinkages between insurers and banks. It is interesting to note that consolidation in the insurance industry 21