Reinsurance and systemic risk: the impact of reinsurer downgrading on property-casualty insurers.
This article analyzes the interconnectedness between reinsurers and U.S. property-casualty (P/C) insurers and presents the first detailed examination on the likely impact of major global reinsurer insolvency on the U.S. P/C insurance industry, in order to illustrate the potential systemic risk caused by the interconnectedness of the insurance sector through reinsurance. We find that the likelihood of a primary insurer's downgrade increases with its reinsurance default risk exposure from downgraded reinsurers. Counterparty primary insurers' stocks also react negatively to their reinsurers' downgrades. The negative effects also spill over to insurers that are not directly exposed to the credit risk of downgraded reinsurers. Despite the close interconnectedness, worst-case scenario analyses show that the likelihood of systemic risk caused by reinsurance transactions is relatively small for the U.S. P/C insurance industry.
The danger of systemic risk to the financial services industry and the world economy as a whole, as triggered by the potential failure of the reinsurance industry, has drawn much attention from industry practitioners, regulators, and academic scholars since the early 2000s (Swiss Re, 2003; Rossi and Lowe, 2002; The Group of Thirty, 2006). Earlier research in general finds the risk is small and inconsequential. Such a view was challenged in the wake of the recent financial crisis, as the meltdown of insurance giant American International Group (AIG) severely deepened the crisis. As a result, a new round of research has emerged to examine the financial stability of the insurance industry and its potential to pose systemic risks to the whole financial system and to national/international economies (Geneva Association, 2010; Cummins and Weiss, 2014; Grace, 2010; Bell and Keller, 2009; Acharya et al., 2009; Harrington, 2009; Billio et al., 2012; Kessler, 2014; Chen et al., 2014). The discussion has continued and has become increasingly important, to the degree that various regulation parties have gotten involved in order to try to identify systemically important financial institutions. In 2011, the Financial Stability Board identified a set of Globally Systemically Important Financial Institutions (Weil Financial Regulatory Reform Center, 2012). Meanwhile, the International Association of Insurance Supervisors (LAIS) was tasked to develop an assessment methodology to identify Global Systemically Important Insurers, and an initial list is expected after April 2013 (IAIS, 2012). In those documents, reinsurance was proposed as an important factor to assess the systemic importance of an insurer. This increasing focus of regulation on the insurance-reinsurance industry definitely calls for in-depth academic research in the area.
The literature generally uses three primary indicators to assess the degree of systemic risk posed by an institution/industry: size, interconnectedness, and substitutability. It is argued that the property-casualty (P/C) insurance industry may be subject to systemic risk because of its heavy dependence on reinsurance and the complexity of the reinsurance market (Cummins and Weiss, 2014). As argued in Acharya et al. (2009), "The reinsurance market increases the interconnectedness of the system exponentially and therefore might increase the systemic risk in the overall market" because of the "bilateral [relationship] in nature and [the lack of] adequate risk controls due to the opacity of bilateral markets." Despite the broad discussion on reinsurance and systemic risk in existing literature, little empirical work has been done to examine the actual interconnectedness of the insurance and reinsurance systems and test how significant the risk could be. Our research intends to fill this gap to some extent by investigating this interconnectedness through examining the reaction of P/C insurers to reinsurer downgrading and conducting scenario analyses to show the hypothetical impacts of major reinsurance groups' insolvency on the U.S. P/C insurance industry. As an addition to existing literature, this paper adopts a more sophisticated methodology and provides more comprehensive empirical analyses to test if reinsurers could be a significant source of systemic risk.
Reinsurance companies are at the top of the insurance sector network. The failure of reinsurance companies may create financial instability within the broader insurance sector, which could cause a spillover effect into the whole economy. In addition, this risk could be aggravated if the increased default risk of primary insurers due to the failure of reinsurers cannot be conceived of transparently in the market, as we have seen in the recent financial crisis. In fact, to outside investors, reinsurance arrangements between primary insurers and reinsurers often seem to be quite complicated, given the complexity of the contract terms and the number of parties involved in the cession and retrocession arrangements. Therefore, it is important to understand the connectedness of the insurance and reinsurance industries and the ability of the market to evaluate the reinsurance risk exposure of primary insurers. In this research, we analyze the impact of reinsurance company credit rating downgrades on counterparty primary insurance companies' credit ratings and stock returns, in order to illustrate the interconnectedness of the insurance sector and to investigate whether the reinsurance credit risk information is transparently delivered to the capital market.
Understanding the interconnectedness is an important step in the context of evaluating the potential systemic risk caused by reinsurance companies. However, this does not provide us information on how serious the potential problem could be. We cannot assess systemic risk brought by reinsurers using historical data because, to date, there has never been a major reinsurance company collapse (Swiss Re, 2003). To get some sense of the magnitude of systemic risk, we conduct multiple scenario analyses in which major global reinsurer(s) collapse(s).
By providing empirical evidence of interconnectedness, the market's ability to evaluate the risk, and the potential impact on the U.S. P/C industry caused by major reinsurance insolvency, we hope that this article can shed light on the systemic risk that the reinsurance sector may pose to the entire financial system and overall economy. The remainder of the article proceeds as follows. After a discussion of the relevant literature on insurance industry interconnectedness, we move on to discuss the data, sample, and methodology, then present empirical results and discussion.
REINSURANCE AND INSURANCE INDUSTRY INTERCONNECTEDNESS
Reinsurance companies are essential to the global insurance industry and have functioned smoothly in the past. However, some concerns in relation to the possibility of systemic risk posed by reinsurance companies have been raised recently, and these concerns can be summarized as follows. First, the top five reinsurance groups (1) provided approximately 60 percent of reinsurance worldwide in 2009 (A.M. Best Company, 2010). The U.S. P/C insurance market also depends heavily on the top reinsurance groups. Based on data reported to the NAIC, the top five global reinsurance groups provide about 30 percent of unaffiliated reinsurance to U.S. P/C insurers. Therefore, these reinsurance companies are at the top of the insurance sector's interconnectedness (Swiss Re, 2003; Cummins, 2007; Cummins and Weiss, 2014). Reinsurance company failure would have a significant impact on primary insurers because those reinsurers may no longer be able to pay the primary insurers' losses. Unfortunately, little is known about the pattern and degree of damages caused by reinsurer failure on primary insurers throughout the world and, consequently, the systemic risk to the real economy (Swiss Re, 2003).
Second, it is hard to isolate the impact of major reinsurer failure from primary insurers and the economy due to the complexity and opacity of reinsurance. There is a serious lack of transparency associated with the risk of reinsurance transactions due to the international nature of reinsurance companies and a lack of standardized prudential supervision (Cole and McCullough, 2006; Rossi and Lowe, 2002; Acharya et al., 2009). To some extent, rating agencies may help reduce some information asymmetry and perhaps may serve as the de facto regulators in the insurance industry (International Monetary Fund, 2004); however, they still cannot completely eradicate the lack of transparency and supervision problems.
Third, there are risks of retrocession/reinsurance spirals that once spread out during the 1988-1992 period of the London Market Excess (Schwartzman, 2008; Cummins and Weiss, 2014). The retrocession spirals may trigger failures of multiple reinsurers all at once, and this shock may cause a ripple effect in a broad range of primary insurers.
There have been a few studies examining the systemic risk posed by the reinsurance industry; many were done by research institutions sponsored by insurance companies. Representative work in earlier years includes Swiss Re (2003) and The Group of Thirty (2006). Both papers conclude that the risk is insignificant, primarily because of the sticky nature of reinsurance liability (withdrawal is only allowed when loss is actually realized) and because of the reinsurance sector's limited connection to the banking sector and capital market. Despite these findings, the interest and concern shown by financial institution supervisors and academic researchers have not abated since the financial crisis of 2007-2008. Bell and Keller (2009) and the Geneva Association (2010) revisit this issue, and both conclude that the insurance sector is fundamentally different from the banking sector, and thus the systemic risk posed by reinsurers seems to be insignificant. The only possible source of systemic risk posed by the insurance/reinsurance industries is through their noncore activities, such as derivative transactions, including Credit Default Swaps, financial derivative trading, short-term funding, and security lending, all of which were major factors behind the AIG crisis. Cummins and Weiss (2014) examine various dimensions of systemic risk posed by the insurance sector. They also conclude that the possibility of systemic risk caused by core insurance activities is limited. However, there could be a significant systemic vulnerability within the insurance sector through reinsurance spirals and the interconnectedness of the insurance sector, which calls for further empirical study.
Previous studies argue that reinsurers pose a low systemic risk because of the very low default probability of major reinsurance companies. For example, Swiss Re (2003) identifies 24 reinsurer bankruptcies during the 1980-2002 period, and none of them involved major reinsurance companies. Due to the limited number of bankruptcies and the relatively small size of bankrupt reinsurers, counterparty credit risks regarding reinsurance companies were considered to be insignificant. Several empirical studies on primary insurer failures also find that, historically, reinsurer bankruptcy accounts for only about 2-5 percent of primary insurers' failure cases (McDonnell, 2002; The London Working Group, 2002; Cummins and Weiss, 2014). However, as we learned from the 2007-2008 financial crisis, there is no such thing as "too big to fail" in the financial world. In recent years, the financial strength rating of reinsurers has deteriorated. As shown by S&P and Moody's, the percentage of reinsurers with AAA and AA ratings decreased significantly between 2002 and 2010, with more firms now falling into the A and BBB rating categories, and there are also more downgrades than upgrades of professional reinsurers during the period. Meanwhile, the global reinsurance industry has become more concentrated than ever. Cummins and Weiss (2000) report that the top 10 reinsurers accounted for 35 percent of the world reinsurance market in 1991, but that percentage increased to 52 percent in 1998 following the merger and acquisition waves during the 1990s. The number further increased to 79 percent by net premiums earned in 2009 in the P/C market (A.M. Best Company, 2010). The increased concentration of the reinsurance market, combined with the seemingly deteriorating quality of reinsurers and the possibility of failure of the major reinsurance companies, has magnified concerns over potential reinsurer failure and the possible spillover effect into the whole insurance industry and beyond.
Credit risk from reinsurance counterparties has been a concern of ceding companies, as reflected in regulation and contracting terms. For example, in the United States, the NAIC specifies that the risk-based capital (RBC) of P/C firms will include a risk charge equal to 10 percent of reinsurance recoverable to guard against the risk of collectability of reinsurance recoverable. Additionally, ceding companies have been increasingly using the special termination clause with rating triggers in their reinsurance contracts (Reynolds Porter Chamberlain LLP, 2007), in order to reduce their credit risk. However, such practices may actually exacerbate the problem of retrocession spirals because the clause would allow the primary company to cancel the reinsurance policy if the reinsurer's rating was downgraded below a certain threshold, making already weak reinsurers even weaker (Cummins and Weiss, 2014), thus leading to a greater potential for systemic risks.
Given the increasing concern over interconnectedness in the insurance market, this article will empirically investigate the dependency of U.S. P/C insurers on reinsurance and the ability of rating agencies and the capital market in assessing the reinsurance risk by examining how the downgrading of reinsurers affects the credit risk of primary insurers (and, therefore, their ratings) and the stock price of publicly traded insurance groups. We also provide scenario analyses of the impact of reinsurer failure(s) by examining how many rating downgrades and insolvencies could be triggered if leading world reinsurers were to collapse.
We study the interconnectedness of insurers and reinsurers in the U.S. P/C insurance industry by using a sample from 2003 to 2009. (2) Financial data for ceding insurers are obtained from NAIC annual statements, and the rating information for these firms is extracted from A.M. Best's Key Rating Guide. Our analysis is at the firm level. The sample originally includes 19,374 observations reported to the NAIC, of which 10,510 observations have rating information available. Some firms in this sample are professional reinsurers, (3) and we exclude them from the analyses. (4) The final regression sample for rating downgrading analysis includes 7,289 firm-year observations, as some observations have missing variables and some are professional reinsurers.
The reinsurance premiums ceded and reinsurance recoverable data are extracted from the NAIC Schedule F--Part 3. Rating information for domestic and global reinsurers is obtained from A.M. Best, S&P, and Moody's. Since the NAIC Schedule F--Part 3 data may involve some reporting errors, especially regarding the names of reinsurers (Cummins, 2007), we use our best discretion to clean the raw data in order to correct errors. We manually merge the ratings and NAIC data by matching the reinsurer's name, domicile, and NAIC code if available. During 2003-2009, our sample of ceding insurers had 353,738 reinsurance transaction data recorded in Schedule F--Part 3. After merging the reinsurance transaction data with rating data, we were able to find ratings for 74.67 percent of reinsurance transactions, representing 84.12 percent in reinsurance recoverable. We record those reinsurers without any ratings as no change in rating over time rather than discarding them from our sample.
More than 80 percent of unrated reinsurance transactions are unauthorized reinsurance that includes various types of alternative risk transfers. As our analyses focus on authorized reinsurance, we expect that the impact of unrated reinsurance should be minimal. (5) However, those insurers with more unrated reinsurance are generally smaller in size and have slightly lower rating. In case the differences cause biased results, we run a subsample analysis where we exclude firms with more than 50 percent of reinsurance recoverable unrated. (6) The results are robust. We also conduct a few more analyses where we exclude firms with more than 5 percent, 10 percent, and 30 percent unrated reinsurance recoverable, and the results remain qualitatively the same in all cases.
THE DEPENDENCY OF PRIMARY INSURERS ON REINSURERS
Summary Statistics: Reinsurance Usage by the U.S. P/C Industry
This section discusses the dependency of primary insurers on reinsurers. Insurance companies can choose to cede their business to affiliated companies within the same insurance group for the benefits of intragroup portfolio diversification. They can also cede business to unaffiliated insurers for the benefits of intercompany diversification. These affiliates and nonaffiliates could domicile in the United States and be subject to U.S. regulation, but they could also be alien companies that are not subject to U.S. regulations. Both types of reinsurance can pose an insolvency threat to insurers, as pointed out by Cummins and Weiss (2014), though nonaffiliated reinsurance is generally considered to pose more counterparty risk than affiliated reinsurance.
Reinsurance could pose significant credit risks to ceding insurers. Reinsurance recoverable may have a significant impact on balance sheets of ceding insurers. In Schedule F--Part 3, the reinsurance recoverable on paid losses and loss adjustment expenses (LLAEs) represents the reinsurance receivable item on an insurer's balance sheet, and the reinsurance recoverable on unpaid LLAEs represent the contraliability item to loss reserves of primary insurers. As a result, the net reinsurance recoverable item in Schedule F--Part 3 is the total effect that reinsurance could have on ceding insurers' surplus (Feldblum, 2002). The recoverable item helps reduce the ceding company's leverage ratio and expand its capacity to write insurance.
Table 1 shows the dependence of U.S. P/C insurers on reinsurance at the industry level. Panel A of Table 1 shows the percentage of total ceded premiums to total direct premiums written and the percentage of total net reinsurance recoverable to policyholder surplus for U.S. P/C industry (excluding professional reinsurers). There is an obvious upward trend in the premiums ceded percentage and a downward trend in recoverable percentage, suggesting that U.S. P/C insurers use more reinsurance services over time but become less "dependent" on reinsurance because of their strong capital position. However, percentage-wise, reinsurance could still pose significant risks to primary insurers. For example, the percentage of net reinsurance recoverable over surplus ratio was 131.3 in 2009.
When breaking down reinsurance activities by contracted reinsurers' type, we find that the industry cedes more premiums to and has more recoverable from affiliated reinsurers (Panel B and Panel C) than from unaffiliated reinsurers (Panel D and Panel E). This trend may be the result of mergers and acquisitions in the U.S. domestic insurance market and global insurance markets, where more unaffiliated firms have become affiliated (Cummins and Xie, 2008; Cummins and Weiss, 2004). The percentages show that U.S. primary insurers depend the most on affiliated reinsurers domiciled in the United States (Panel B). The second largest category is composed of U.S. unaffiliated reinsurers (Panel D), followed by alien affiliated reinsurers (Panel C) and alien unaffiliated reinsurers (Panel E). (7)
One concern regarding the significance of the research is the collateralization of reinsurance recoverable. To reduce reinsurer counterparty credit risk, U.S. insurance regulators applied the Credit for Reinsurance Model Law to unauthorized reinsurance (Cole, McCullough, and Powell, 2010). (8) The rule states that any reinsurance ceded to unauthorized foreign or alien reinsurers cannot take a credit against the ceding company's loss reserve or unearned premium reserve if uncollateralized. To receive surplus relief from such reinsurance transaction, some ceding companies require reinsurers to deposit collaterals. This collateral requirement helps reduce the insolvency risk of ceding insurers, which may weaken the risk interconnectedness between the ceding company and the reinsurer. To address the concern, we provide summary statistics on unauthorized reinsurance and provide analyses that distinguish authorized reinsurance and unauthorized reinsurance from the downgraded reinsurers. (9)
Diversification of Reinsurance Portfolios (at Firm Level)
Table 2 shows the diversification of reinsurance portfolios for U.S. P/C insurers (professional reinsurers excluded). Five types of Herfindahl indices are calculated based on ceded premiums and net reinsurance recoverable, respectively: (1) Herfindahl index for all reinsurers, regardless of their affiliation and domicile; (2) Herfindahl index for U.S. affiliated reinsurers only; (3) Herfindahl index for alien affiliated reinsurers only; (4) Herfindahl index for U.S. unaffiliated reinsurers only; and (5) Herfindahl index for alien unaffiliated reinsurers only. Both mean and median values of Herfindahl indices are reported, along with the number of ceding firms that use those types of reinsurers. Since the premiums-based Herfindahl indices and recoverable-based Herfindahl indices are highly correlated (10) and analyses in our article focus on reinsurance recoverable, we proceed with the recoverable-based Herfindahl.
The results indicate that U.S. insurers overall are not diversified in their reinsurance portfolios (with a mean Herfindahl index higher than 0.6). (11) The situation does not improve much over time (from 0.657 in 2002 to 0.639 in 2009). The concentration in reinsurance portfolios is mainly attributable to firms that cede premiums to their U.S. affiliates. About 60 percent of ceding firms have reinsurance recoverable from their U.S. affiliates. This affiliated reinsurance portfolio is extremely concentrated (with a mean Herfindahl index higher than 0.9) and shows no sign of changing over time. This result supports the view of Cummins and Weiss (2014) that affiliates could be a significant source of credit risk to U.S. insurers. There is also a growing number of U.S. insurers using alien affiliated reinsurer services (from 10 percent in 2002 to 13 percent in 2009, calculated from the number of ceding firms, recoverable based). This reinsurance portfolio is concentrated as well (with a mean Herfindahl index higher than 0.8). The higher concentration in affiliated reinsurance may reflect ceding insurers' limited choices of reinsurance partners within a group, as compared to choices from a larger number of nonaffiliated firms outside the group.
More than 70 percent of U.S. insurers have reinsurance transactions with U.S. nonaffiliates, and this set of reinsurance portfolios is more diversified (with a mean Herfindahl index of 0.543 in 2002 and 0.532 in 2009) than the affiliated reinsurance portfolio. A significant percentage of U.S. insurers (40 percent in 2002 and 52 percent in 2009) use alien unaffiliated reinsurance services; this set of reinsurance portfolios is the most diversified, and the diversification level increases slightly over time (mean Herfindahl index 0.484 in 2002 and 0.449 in 2009). (12)
THE IMPACT OF REINSURER DOWNGRADES ON PRIMARY INSURERS' RISK
This section analyzes the impact of reinsurer downgrades on primary insurers' risk. A downgrade of a reinsurer may lead to an increase in the risk of its counterparty primary insurers because of the increased default risk of reinsurance recoverable. Therefore, through reinsurance transactions, the risk of reinsurers is connected to the primary insurers' risk. If rating agencies and capital market can assess the extent of the connection, rating downgrades of counterparty reinsurers should negatively affect primary insurers' financial strength ratings and stock prices. The objective of this section is to conduct joint tests examining the extent of risk transition from reinsurers to primary insurers and the ability of the market to conceive of these risks.
The Impact of Reinsurer Rating Downgrades on Counterparty Primary Insurers' Rating
To examine the interconnectedness in the insurance-reinsurance market, we first analyze the rating changes of primary insurers following the downgrades of their reinsurer counterparties. Specifically, we run the following logit regression model:
[Pdown.sub.it] = [alpha] + [beta] x [RDownRec_AU.sub.i,t-1] + [gamma] x [RDownRec_UN.sub.i,t-1] + [delta] x [X.sub.it] + [theta] x [Year.sub.t] + [[epsilon].sub.it], (1)
[Pdown.sub.it] = dummy variable with the value of 1 when a primary insurer i is downgraded;
[RDownRec.sub.A][ U.sub.i,t-1] = the sum of authorized unaffiliated reinsurance recoverable from all downgraded counterparty reinsurer(s)/surplus of the insurer i;
[RDownRec.sub.U] [N.sub.i,t-1] = the sum of unauthorized unaffiliated reinsurance recoverable from all downgraded counterparty reinsurer(s)/surplus of the insurer i;
[X.sub.it] = a set of control variables that may affect the rating change of the insurer i; and
[Year.sub.t] = year-fixed effects dummy variables that allow us to control both the macroeconomic conditions and insurance industry-specific conditions that affect the ratings of primary insurers.
The timing of the downgrading of primary insurer i and its reinsurer(s)' downgrading is as follows: assume primary insurer i has ratings at date D1 (prior rating date) and date D2 (current rating date); if this firm is downgraded at date D2 from D1, then [Pdown.sub.it] is set equal to 1. Next, if any of this firm's reinsurers are downgraded between dates D1 and D2, we record it as a reinsurer downgrading and calculate [RDownRec.sub.A] [U.sub.it,-1] and [RDownRec.sub.U] [N.sub.i,t-1] accordingly.
[RdownRec.sub.A] [U.sub.i,t-1] ([RDownRec.sub.U] [N.sub.i,t-1]) is constructed as follows. For insurer i at year t, we calculate its sum of authorized (unauthorized) reinsurance recoverable at t - 1 from all downgraded unaffiliated reinsurers, then scale this by insurer i's surplus at t - 1 to measure the relative size of the default risk from downgraded reinsurers. We focus on the unaffiliated reinsurance transactions out of concern that subsidiary insurers within the same insurance group often receive the same rating, and affiliated insurers' risks are interconnected in much more complicated ways than the explicit reinsurance transactions. However, we include in the regression the proportion of downgraded affiliated reinsurance recoverable to surplus (RDownRec_Aff) to control the possible impact of affiliated reinsurance downgrade. (13,14)
Because authorized reinsurance usually calls for no collateral, the credit risk of primary insurer will become higher when it has more recoverable on authorized reinsurance from downgraded reinsurers. Therefore, we expect [beta] > 0. In contrast, as recoverable from unauthorized reinsurers is usually collateralized, credit risk associated with it is smaller than authorized reinsurance. Hence, we expect [gamma] to be insignificantly different from 0.
In addition to reinsurance arrangements, a primary insurer's rating may be downgraded due to a host of other factors that affect the firm's financial strength. We follow the insurance rating literature to select control variables [X.sub.it] (Cummins, Harrington, and Klein, 1995; Doherty and Phillips, 2002; Kartasheva and Park, 2011; Doherty, Kartasheva, and Phillips, 2012). To control asset-side risks, we include investment yield and percentage of junk bond holdings. Higher achieved investment yield can improve rating, but larger junk bond holdings can increase default risk. On the liability side, we control for exposures of catastrophe risks, net premiums written to surplus ratio, reserve to surplus ratio, and combined ratio. Higher values of these variables are expected to weaken a firm's financial strength and thus have a negative impact on the insurer's rating. Next, reinsurance recoverable to surplus ratio is added to capture the overall size of reinsurance-related credit risk. We also control for asset size and Best's Capital Adequacy Ratio (BCAR), which is calculated by A.M. Best as a capital adequacy measure and is claimed by A.M. Best as the most important factor in issuing a rating (A.M. Best Company, 2009). To examine the validity of these control variables, we run an ordered probit regression with the dependent variable being the numerical conversion of the A.M. Best rating. (15) All variables are significant with expected signs (see Table 3). Because the dependent variable of the regression model (1) is not the rating itself but the change in rating (downgrade), we include the difference between year t-1 and year t of these selected variables as control variables. (16,17)
In addition to the above-mentioned control variables, we also control the previous A.M. Best Rating in the regression. The previous A.M. Best rating controls possible heterogeneity in rating changes for different rating categories. It is also argued that both ceding insurers and reinsurers may be more likely to get downgraded during the hard market, and this may bias up the coefficients of RDownRec_UN and RDownRec_AU. To address this issue, we include the industry combined ratio to control for the impact of underwriting cycle on primary insurers' downgrading (Harrington, Niehaus, and Yu, 2012).
A concern regarding collateralization is that insurers that are more likely to be downgraded may choose to use more unauthorized reinsurance with collaterals to lower credit risks. An endogeneity issue may exist in that the same factors driving the downgrade may drive the type of reinsurance selected as well. To address this issue, we treat RRUN_TTREC (the proportion of unauthorized reinsurance recoverable to total reinsurance recoverable) as endogenous and adopt the following 2SLS regression method in the regression analyses (Table 5) and robustness checks (see the Appendix).
[RRUN_TTREC.sub.it] = [alpha] + [beta] x [Z.sub.i,t] + [rho] x [Year.sub.t] + [[omega].sub.it]
[PDown.sub.it] = [alpha] + [beta] x [RDownRec_AU.sub.i,t-1] + [gamma] x [RDownRec_UN.sub.i,t-1] + [delta] x [X.sub.it] + [phi] x [RRUN_TTREC.sub.it] + [theta] x [Year.sub.t] + [[epsilon].sub.it], (2)
where Z is a set of factors affecting the usage of unauthorized reinsurance or ratings, which includes investment yield, the ratio of junk bonds to surplus, the percentage of business in catastrophic risk exposed lines, premium to surplus ratio, reserve to surplus ratio, combined ratio, BCAR, log asset, lagged financial rating, industry combined ratio, single unaffiliated firm dummy, and publicly traded firm dummy. For the purpose of identification, we need instrumental variables that affect a firm's usage of unauthorized reinsurance, but do not affect its rating change directly. As a result, we compute and include in the first-stage regression the industry's "preference" to use unauthorized reinsurance (the industry's unauthorized reinsurance recoverable ratio) as an instrument. The identified endogeneity issue does not materially impact our key variables and other control variables.
Table 4 shows the summary statistics of variables used in model (1). For our regression sample, 3.2 percent of primary insurers have been downgraded. The average ratio of authorized reinsurance recoverable from downgraded unaffiliated reinsurers to policyholders' surplus is 2.75 percent, and the ratio for unauthorized reinsurance recoverable is 0.56 percent. (18)
The results of equation (2) (second stage) are shown in Table 5. (19) As expected, the coefficient of RDownRec_AU is positive and significant, but the coefficient of RDownRec_UN is insignificant. This implies that a primary insurer is more likely to be downgraded when it has greater credit risk exposure from the contracted downgraded reinsurer(s). However, when the reinsurance transaction is collateralized, the reinsurer's default risk does not impact the primary insurer's risk.
In addition, we also conjecture that unaffiliated single companies that only have access to unaffiliated reinsurance will be more adversely impacted by the downgrading of unaffiliated reinsurers as these firms can only cede to unaffiliates. To test this, we add the interaction terms of a single company dummy, RDownRec_AU, and RDownRec_UN. The result is presented in column 2 of Table 5. As conjectured, the interaction term of single and RDownRec_AU is significantly positive, which suggests that the increased uncollateralized reinsurance risk has more impact on unaffiliated single insurers than on the subsidiaries of a group.
We run a few more regressions as robustness checks. First, we create alternative RDownRec AU and RDownRec_UN as the time gap between reinsurer downgrade and primary insurer rating assignment could be too long. (20) To address this concern, we create a new RDownRec_AU and RDownRec UN, where we restrict reinsurer downgrades to the ones having occurred between the primary insurer's current rating date and 30 days prior to the primary insurer's current rating date. The result is presented in column 3. The interaction term of RDownRec_AU*Single is positive and significant. (21,22)
Second, we conduct a regression using only the leading insurer of each group to correct the potential problem caused by the relatedness of ratings of firms within a group. We select leading insurance companies based on asset size. The result (regression 4 of Table 5) is robust. (23) Lastly, we conduct more robustness checks regarding econometric specifications. To address possible heteroskedasticity and autocorrelation issues, we run alternative models and report the results in columns 1-4 of the table in the Appendix. The signs of our key variables remain the same in all specifications. (24)
The results in this section provide evidence on the interconnectedness between primary insurers and reinsurers and rating agencies' ability to properly incorporate the increased risk from reinsurance recoverable. Since the result holds after controlling for the changes in primary firm's combined ratio and industry-wide combined ratio, it reasonably excludes contamination of the reverse causality effect, in which a loss suffered by a primary insurer may affect the reinsurers' rating at the same time and then spuriously cause a positive relationship between reinsurers' downgrade and the primary insurer's downgrade.
The Impact of Reinsurer Rating Downgrades on Primary Insurers' Stock Price
Since systemic risks and adverse shocks to an industry are usually first captured by the stock market, in this section, we examine the link between the reinsurer ratings downgrades and the primary insurers' stock price in the event study framework. We have shown that counterparty reinsurers' risk adversely affects primary insurers' ratings. Similar adverse effects should also appear in the stock market, should market participants perceive the interconnectedness between reinsurers and primary insurers. Here, stock market analyses provide an advantage over the ratings downgrade analyses, since impacts of adverse events are usually more directly reflected in short-term stock price movements. As reinsurance downgrades can be triggered by unexpected losses from primary insurers, one might argue that rating downgrades of primary insurers may not be the result of reinsurance downgrades but the cause. The event study method can actually address the problem rather nicely. Given that there is lag time between primary insurers' loss events and reinsurers' downgrades, the effect of a large loss on the part of primary insurers that may trigger a reinsurer's downgrade should have already been absorbed in the primary insurer's stock price by the time of the reinsurer downgrade. Therefore, changes in primary insurers' stock price following a reinsurer's downgrade are most likely attributable to the reinsurer's downgrade.
We assess the market reaction of publicly traded primary insurers to the news of reinsurers' downgrading. The first analysis (Table 6, Panel A) presents the stock return reaction of counterparty primary insurers of the downgraded reinsurer(s), which measures the direct impact of reinsurer downgrades. The second analysis (Table 6, Panel B) presents the negative spillover effects of reinsurer downgrades, that is, the reaction of noncounterparty primary insurers (insurers that do not have reinsurance arrangements with the downgraded reinsurers).
Event study is used to measure the stock price reaction to reinsurer downgrading announcements. Given the fact of firm clustering around any downgrading event, we adopt the seemingly unrelated regression (Zellner, 1962) method to estimate abnormal returns for the set of firms affected in each event to allow for dependence among residuals. The equation form is specified as below, and we use a 250-day estimation period ending 30 days prior to each event. (25)
[R.sub.it] - [r.sub.ft] - [[alpha].sub.i] + [[beta].sub.i.sup.*] ([R.sub.mt] - [r.sub.ft]) + [K.summation over (k=1)] [[gamma].sub.ik][D.sub.k] + [[epsilon].sub.it], (3)
where [R.sub.it] is the return of insurer i at day t; [R.sub.mt] is the equally weighted return on all NYSE, AMEX, and NASDAQ stocks of day t; and [r.sub.ft] is the 1-month Treasury bill rate at day t and is collected from Fama/French database. [D.sub.k] is a set of dummy variables that is equal to 1 on days in the event period and 0 otherwise. (26) The coefficient [[gamma].sub.ik] is the event-induced abnormal return for the ith insurer on event day k. The cumulative abnormal return for firm i in the ([[tau].sub.1], [[tau].sub.2]) event window is calculated as [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII].
The average cumulative abnormal returns across all firms for all events are:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII],
where N is the total number of firms affected by the downgrading events in our sample.
To test whether the market reaction is actually attributable to the proportion of reinsurance recoverable from the downgraded reinsurers, we conduct a regression analysis for counterparty primary insurers. The model is specified as below:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII], (4)
where [RDownRec.sub.A] [U.sub.i,t-1] and [RDownRec.sub.U] [N.sub.i,t-1] are defined similarly to those in "The Impact of Reinsurer Rating Downgrades on Counterparty Primary Insurers' Rating" section, and [X.sub.it] is a set of control variables of primary insurer characteristics that include: size of the insurer i (measured by log value of firm i's market equity), Q value (market value of assets/book value of assets), leverage (equity-to-assets ratio), and percentage of business with catastrophic risk exposure, combined ratio, percentage of business in long-tail lines, and industry combined ratio. We expect that firms with greater exposure to authorized reinsurance from downgraded reinsurers will experience more negative returns. The results are shown in Table 6, Panel C.
Table 6 shows that reinsurer downgrade events have a strong, statistically significant negative impact on the stock prices of counterparty primary insurers, with an average CAR -1.74 percent for the (-15, +15) days window (Panel A). (27) This suggests that an increase in reinsurance risk brings additional risk to the primary insurers and therefore reduces stock value. The regression further shows that the negative stock response is directly related to the reinsurer downgrading. After controlling for other firm characteristics, risk exposures, and industry situation, primary insurers with a higher percentage of authorized reinsurance recoverable from the downgraded reinsurers do experience more negative stock responses.
In addition to the direct effects on counterparty primary insurers, the lack of transparency in the reinsurance market may create a contagion effect in the primary insurer market. That is, a reinsurer's failure could have negative effects on insurers that have no direct business with the troubled reinsurers. To test this, we examine the stock reactions of primary insurers with no direct credit risk exposure to the downgraded reinsurers. The result shows that reinsurer downgrade announcements also have significant externalities on the stocks of noncounterparty primary insurers, with a negative CAR of -0.59 percent for the (-15, +15)-day window (Panel B), the magnitude of which is smaller than these events' impact on counterparty primary insurers.
The negative reaction of noncounterparty insurers can be interpreted in two ways. One is that it could represent pure contagion effects caused by opacity: the market irrationally reprices all insurers, regardless of their relationship with the downgraded reinsurers. Alternatively, it could be information based: the market worries about the indirect impact of downgraded reinsurers through retrocession spirals. As we do not have access to the retrocession transactions, testing these hypotheses is outside the scope of this study. However, the fact that the contagion effect is only about 30 percent of direct effect suggests that the reinsurance transaction is reasonably transparent. (28)
SCENARIO ANALYSIS OF LARGE REINSURANCE COMPANIES' INSOLVENCY
In this section, we examine the hypothetical impact of large reinsurer insolvency. Despite the fact that no major reinsurer insolvencies occurred historically, in the wake of the collapse of AIG and other giant financial institutions, it is imperative that we improve our understanding of the dynamics and anticipate the scenarios of large reinsurer insolvency in the future.
We conduct scenario analyses by allowing one of the top three reinsurer groups (Swiss Re, Munich Re, and Berkshire Hathaway) to become insolvent. Because part of the recoverable can be paid off even with complete liquidation of reinsurers, we run multiple cases where the authorized reinsurance recoverable is defaulted by 100 percent, 50 percent, 30 percent, or 10 percent. (29) We examine the effects of this recoverable default on the primary insurers' ratings and insolvencies using year 2009 data. We use Swiss Re as an example to describe our analyses.
To examine the impact of the insolvency of Swiss Re group (the parent and all subsidiaries) on primary insurers' ratings, we use the probit regression model in Table 3. First, we run the rating regression with the original surplus and get the ratings estimate for each insurer. Second, we calculate the hypothetical surplus of insurers by assuming that 100 percent, 50 percent, 30 percent, or 10 percent of their reinsurance recoverable from Swiss Re will default. Next, we calculate all explanatory variables using the hypothetical surplus. (30) For BCAR, we cannot simply compute the hypothetical number as the function is not known, so we provide two cases where (1) BCAR stays constant, and (2) the new BCAR is defined as old BCAR multiplied by the ratio of new hypothetical surplus divided by original surplus. This way, we allow the new BCAR to vary between zero and the original BCAR. We then estimate the hypothetical rating of the primary insurer by plugging new hypothetical explanatory variables into the fitted model shown in Table 3, and compare the original estimated ratings with the estimated hypothetical ratings to draw a conclusion on ratings downgrades. (31)
Table 7 shows the scenario analysis of the fall of major reinsurers and the likely impacts on primary insurers' ratings. The number of downgraded insurers as a result of the reinsurance recoverable default is presented. (32) The result shows that the impact of major reinsurers' insolvency on U.S. P/C insurers is not serious. Even under the extreme and unlikely assumption of 100 percent recoverable default, fewer than 36 insurers would be downgraded. The impact of Swiss Re's insolvency on U.S. insurers is the strongest; 31 (or 36 with BCAR_new) out of 1,367 insurers would be downgraded if assuming 100 percent default from Swiss Re. Under the more realistic assumption of a 30 percent default, less than 1 percent of insurers would be downgraded when one of the top three reinsurers is insolvent. If the default rate is set to 10 percent, then at most one insurer would be downgraded if any of the three reinsurance groups become insolvent.
We also conduct scenario analysis to assess how many primary insurers would become insolvent as a result of reinsurer insolvency. Using Swiss Re as an example again (see Figure 1), we first calculate the hypothetical surplus of primary insurers (insurers A-1, A-2, C, and D) if their unaffiliated reinsurer--Swiss Re--becomes insolvent. If the new surplus of any insurer is negative, we treat this firm as insolvent. However, as insurance regulators in the United States start to monitor an insurer closely if its surplus drops below 200 percent of RBC, we also use the 200 percent RBC level as a conservative criterion of insolvency. That is, once a firm's hypothetical surplus goes below 200 percent of its RBC, we record it as an insolvent case.
Tracking the direct impact of Swiss Re's insolvency on its counterparty primary insurers is not sufficient to assess its overall impact on the insurance industry, because the insolvent primary insurers may also have assumed reinsurance; that is, a chain effect may exist. Therefore, a primary insurer's insolvency as a result of Swiss Re group's insolvency may make more insurers become insolvent through affiliated and unaffiliated reinsurance transactions. To examine this chain effect, we estimate the total reinsurance recoverable that may be subject to default for a primary insurer by adding (1) its unaffiliated reinsurance recoverable from any insurers and reinsurers in Swiss Re group and (2) its reinsurance recoverable from other contracted affiliated and unaffiliated reinsurers that are hypothetically insolvent as a result of Swiss Re's insolvency. For example, as shown in Figure 1, if insurer A-1 becomes insolvent due to the collapse of Swiss Re (direct effect), and it has assumed reinsurance from its affiliated insurer A-2 and unaffiliated insurer C, then the total effect of Swiss Re on insurer A-2 becomes b + c, and the total effect on insurer C is f + e. If insurer C becomes insolvent as a result of the first-round chain effect, and if it had assumed reinsurance from insurer D, then the final effect on insurer D is g + h.
For simplicity, we assume that the same proportion of recoverable can be collected from all insolvent reinsurers. For example, under the 30 percent loss scenario, we assume that insurer A-l can recover 70 percent of recoverable from Swiss Re when Swiss Re is insolvent. If this puts insurer A-l into insolvency, insurer A-2 and insurer C, which had ceded business to insurer A-l, can now only collect 70 percent of reinsurance recoverable from insurer A-l. If this also puts insurer C into insolvency, we assume insurer D can only collect 70 percent from insurance C. We repeat this process until we reach a point where the number of insolvent insurers does not increase any more.
In each analysis, we did not count the downgrades or insolvency of the subsidiaries of Munich Re (Swiss Re or Berkshire Hathaway) under Munich Re's (Swiss Re's or Berkshire Hathaway's) insolvency scenarios, because we already assumed that all insurers of these groups have defaulted. Therefore, the total insolvent insurers in United States as a result of Swiss Re's default will be the presented number in Table 8 plus the number of Swiss Re's U.S. subsidiaries.
Table 8 shows the number of insolvent insurers resulting from reinsurance recoverable default. The study sample includes all U.S. P/C insurers that have a surplus greater than 200 percent RBC in 2009. If using negative surplus to define insolvency, fewer than 9 insurers out of 2,492 would become insolvent even under the extreme assumption of 100 percent reinsurance loss from Swiss Re. The chain effect is also minimal. Only one more insurer would become insolvent when assuming 100 percent recoverable default from both Swiss Re and the nine additional insolvent insurers resulting from Swiss Re's insolvency. The number of insolvent insurers doubles if we apply a more conservative criterion--200 percent RBC--but this number is still small relative to the size of the sample. Fewer than 30 insurers would become insolvent with and without the chain effect considered in all three cases, even with the assumption of 100 percent recoverable default. In unreported analysis, we also track the total assets of the insolvent insurers. In any one of the major reinsurer insolvency scenarios, the total assets of the resulting insolvent firms are smaller than 1 percent of total industry assets.
One major concern we have before we can conclude that the systemic risk caused by reinsurer collapse seems to be minor is that we have not considered reinsurance spiral cases, in which multiple reinsurers' financial conditions deteriorate simultaneously due to the complex retrocession transactions among reinsurers. The risk of reinsurance spiral has been pointed out as a possible source of systemic risk (Cummins and Weiss, 2014). We consider two extreme cases: all three big reinsurers become insolvent altogether, and the most extreme case where all unaffiliated reinsurers become insolvent at the same time. The last two categories of Tables 7 and 8 show the number of downgraded and insolvent insurers for these two extreme cases.
In addition, we add one more hypothetical scenario where 30 percent of primary insurer's surplus is depleted due to adverse macroeconomic shocks. This is to address the concern that certain macroeconomic condition or major loss shocks that put major reinsurers in a default situation will also affect primary insurers, unless the reinsurer becomes insolvent due to a purely operational reason. We find that before any reinsurer insolvency is considered, this 30 percent surplus shock would cause 294 insurers to be downgraded, and 74 insurers' surplus to fall below the 200 percent RBC level, but no insurers would experience a negative surplus due to this depletion. Using this as a starting situation, we report in Tables 7 and 8 the incremental number of downgrades and insolvency resulting from reinsurers' insolvency (the "30 percent shock" rows). For example, when assuming 100 percent reinsurance recoverable default from Munich Re and considering chain effect, the "RBC 200 percent, 30 percent shock" reports 101 firms. This means that the total number of insurers whose surplus would go below 200 percent of RBC is 175, which is 101 plus 74. With 100 percent or 50 percent recoverable default assumption, and the collapse of three reinsurers altogether, which is only an apocalyptic scenario, the number of downgraded or insolvent insurers could cause quite a large shock to the economy. However, with a more realistic assumption of either a 30 or 10 percent recoverable default rate, the number of insurers that would become insolvent becomes much smaller. The impact on the economy as a whole would be manageable.
CONCLUSION AND DISCUSSION
In this article, we examine systemic risks posed by the insurance sector through global reinsurers. Our goal is twofold. The first is to provide empirical evidence of the interconnectedness between reinsurers and U.S. P/C insurers. The second is to present the first detailed examination on the likely impact of major global reinsurer insolvency on the U.S. P/C insurance industry.
There have been concerns about the complexity of the reinsurance transaction network and its resulting opacity, but our results suggest that the risk transitions from reinsurers to primary insurers are fairly well recognized by both rating agencies and capital market participants. We document that the downgrade of reinsurers increases the likelihood of downgrading for counterparty primary insurers. We also find that primary insurers' stock prices react negatively to the downgrade of reinsurers in the event study framework. These results provide evidence that there is a close interconnectedness between the insurance sector and the reinsurance sector, and the market has well recognized it.
The next question we address is how bad things could get if major global reinsurer(s) collapse. We consider multiple scenarios where top global reinsurers become insolvent. The results suggest that it is reasonable to conclude that the systemic risk caused by reinsurance transactions is relatively small. Even under an extreme assumption of a 100 percent reinsurance recoverable default by one of the top three global reinsurers, only about 2 percent of insurers would be downgraded, and 1 percent of insurers would become insolvent. We also examine a chain effect where we evaluate the risk spreading through both unaffiliated and affiliated reinsurers. Despite the high dependency and concentration of affiliated reinsurance portfolios found in this study, the increased risk from this chain effect turns out to be minimal. It seems that a parent or large company in group providing affiliated reinsurance to its subsidiaries is reasonably diversified in its unaffiliated reinsurance portfolio, so the hypothetically dangerous affiliated reinsurance default chain is not triggered by the insolvency of major reinsurance companies. The shock will be larger when we assume the failure of reinsurer is added to an already bad situation. However, our result suggests that the incremental shock due to the interconnectedness of insurer and reinsurer will not be very large unless the three top reinsurers fail concurrently and can only pay less than 50 percent of recoverable.
Our study of interconnectedness and scenario analyses only serves as the first step in analyzing possible systemic risk imposed by the reinsurance sector. There are other factors that should be considered when reaching final conclusions. First, the negative effects detected from past downgrading events may only serve as a lower limit of major reinsurer solvency. The market in the past has only experienced the insolvency of small reinsurers. Shocking news, such as major global reinsurer(s) failure, could panic the market, magnifying the contagion effect even further, as we have seen in the recent financial crisis. Second, the impact of affiliated insurer insolvency on other affiliated insurers within the same group is not fully addressed in this article. Although we include affiliated reinsurance transactions in our scenario analysis, firms within the same group are connected through many channels other than reinsurance transactions. Collapse of affiliated reinsurers may have a more significant impact on a primary insurer than unaffiliated reinsurers because of the concentration of intragroup reinsurance arrangements and the sharing of the same corporate culture, risk preferences, and corporate governance mechanisms.
APPENDIX The Impact of Reinsurer Rating Downgrades on Primary Insurer Rating Downgrades--Robustness Checks, 2003-2009 (1) Random (2) Fixed Variables Effects Effects (3) AR(1) RDownRec_UN -0.0288 -0.0309 -0.0234 (0.0457) (0.0503) (0.0456) RDownRec_AU 0.0737 *** 0.0674 *** 0.0647 *** (0.0185) (0.0201) (0.0179) Single 0.0019 0.0001 (0.0091) (0.0075) Single * -0.0132 -0.1334 RDownRec_UN (0.1969) (0.1990) Single * 0.1274 ** 0.1392 ** RDownRec_AU (0.0550) (0.0545) [DELTA]Investment yield -0.0010 -0.0012 -0.0011 (0.0010) (0.0011) (0.0010) [DELTA]CAT risk 0.0785 0.0638 0.0690 (0.0566) (0.0620) (0.0551) [DELTA]NPW/PHS 0.0002 *** 0.0002 *** 0.0002 *** (0.0001) (0.0001) (0.0001) [DELTA]Reinsurance 0.0001 ** 0.0001 ** 0.0001 ** recoverable/PHS (0.0001) (0.0001) (0.0000) [DELTA]Reserve/PHS 0.0007 *** 0.0006 *** 0.0007 *** (0.0001) (0.0001) (0.0001) [DELTA]Junk bond/PHS -0.0010 * -0.0009 -0.0009 * (0.0005) (0.0006) (0.0005) [DELTA]BCAR -0.0001 ** 0.0000 -0.0001 ** (0.0000) (0.0000) (0.0000) [DELTA]Log -0.0264 ** -0.0110 -0.0235 * (Asset) (0.0130) (0.0147) (0.0129) [DELTA]Combined ratio 0.0003 *** 0.0002 * 0.0003 *** (0.0001) (0.0001) (0.0001) Best rating -0.0007 0.0548 *** -0.0027 * (t--1) (0.0019) (0.0050) (0.0016) RRUN TTREC 0.0356 -0.1868 ** 0.0573 * (0.0385) (0.0889) (0.0340) RDownRec Aff -0.0112 -0.0161 -0.0246 (0.0161) (0.0175) (0.0163) Ind CombinedR 0.1598 *** 0.1723 *** 0.1409 *** (0.0482) (0.0503) (0.0479) Number of 7,289 7,289 7,289 observations Chi-squared 284.1 284.4 F-value 23.6 (4) Driscoll &Kraay (5) Level (6) Lagged Variables Error Downgrade Control RDownRec_UN -0.0309 -0.5802 -1.4296 (0.0391) (1.3900) (1.6475) RDownRec_AU 0.0674 ** 0.6377 ** 0.5900 ** (0.0342) (0.2911) (0.3008) Single 0.132 -0.0674 (0.2032) (0.2051) Single * 1.7025 3.0781 RDownRec_UN (3.9165) (3.8762) Single * 1.3785 * 1.4377 * RDownRec_AU (0.7096) (0.7935) [DELTA]Investment yield -0.0012 ** -0.0402 -0.1109 ** (0.0005) (0.0278) (0.0518) [DELTA]CAT risk 0.0638 ** 2.2347 0.0022 (0.0277) (1.5645) (0.0109) [DELTA]NPW/PHS 0.0002 *** 0.0059 *** -0.0084 (0.0001) (0.0017) (0.0065) [DELTA]Reinsurance 0.0001 *** 0.0025 * 0.0030 *** recoverable/PHS (0.0000) (0.0014) (0.0009) [DELTA]Reserve/PHS 0.0006 *** 0.0177 *** -0.0036 *** (0.0001) (0.0022) (0.0012) [DELTA]Junk bond/PHS -0.0009 ** -0.0438 ** 0.0119 (0.0005) (0.0190) (0.0142) [DELTA]BCAR 0.0000 -0.0021 *** -0.0033 *** (0.0000) (0.0008) (0.0009) [DELTA]Log -0.0110 -1.3416 *** 0.0011 (Asset) (0.0177) (0.3724) (0.0045) [DELTA]Combined ratio 0.0002 ** 0.0079 *** 0.0015 ** (0.0001) (0.0026) (0.0007) Best rating 0.0548 *** -0.0929 ** -0.0706 (t--1) (0.0197) (0.0432) (0.0451) RRUN TTREC -0.1868 ** 1.6139 * -1.7997 (0.0937) (0.8954) (1.1960) RDownRec Aff -0.0161 ** -5.7264 -4.7918 (0.0071) (9.9109) (7.6382) Ind CombinedR 0.1723 *** 5.2212 *** 9.7256 *** (0.0439) (1.7273) (1.7798) Number of 7,289 7,289 7,289 observations Chi-squared 235.4 104.6 F-value 105.1 (8) Alternative (7) Sub Affiliates Variables Sample Downgrade RDownRec_UN 0.1632 -0.5089 (1.3672) (1.4292) RDownRec_AU 0.7474 ** 0.6937 ** (0.2953) (0.2930) Single 0.2189 0.1182 (0.2090) (0.2053) Single * 1.1179 1.7616 RDownRec_UN (3.8576) (3.9044) Single * 1.4915 * 1.5980 ** RDownRec_AU (0.7759) (0.7763) [DELTA]Investment yield -0.0452 -0.0333 (0.0323) (0.0273) [DELTA]CAT risk 2.2630 2.1448 (1.5847) (1.5782) [DELTA]NPW/PHS 0.0053 *** 0.0059 *** (0.0018) (0.0017) [DELTA]Reinsurance 0.001 0.0018 recoverable/PHS (0.0015) (0.0014) [DELTA]Reserve/PHS 0.0175 *** 0.0169 *** (0.0025) (0.0024) [DELTA]Junk bond/PHS -0.0442 ** -0.0432 ** (0.0195) (0.0191) [DELTA]BCAR -0.0025 *** -0.0021 *** (0.0008) (0.0008) [DELTA]Log -1.2701 *** -1.2435 *** (Asset) (0.3916) (0.3757) [DELTA]Combined ratio 0.0069 *** 0.0074 *** (0.0026) (0.0025) Best rating -0.0654 -0.0834 * (t--1) (0.0450) (0.0434) RRUN TTREC 1.8616 ** 1.6303 * (0.8787) (0.8688) RDownRec Aff -5.0202 -0.0405 (9.0538) (0.1718) Ind CombinedR 5.6754 *** 5.4019 *** (1.8051) (1.7358) Number of 6,987 7,289 observations Chi-squared 207.4 216 F-value Note: Standard errors are in parentheses. Significant at the *** 1%, ** 5%, and * 10% levels.
Acharya, V. V., J. Biggs, M. Richardson, and S. Ryan, 2009, On the Financial Regulation of Insurance Companies, NYU Stern School of Business, Working Paper.
A.M. Best Company, 2009, Best's Credit Rating Methodology, Global Life and Non-Life Edition (Oldwick, NJ: A.M. Best Company).
A.M. Best Company, 2010, Global Reinsurance: 2009 Financial Review, Best's Special Report, April 12 (Oldwick, NJ: A.M. Best Company).
Bell, M., and B. Keller, 2009, Insurance and Stability: The Reform of Insurance Regulation, Zurich Financial Services Group (Zurich, Switzerland).
Billio, M., M. Getmansky, A. W. Lo, and L. Pelizzon, 2012, Econometric Measures of Systemic Risk in the Finance and Insurance Sectors, Journal of Financial Economics, 104(3): 535-559.
Boehmer, E., J. Musumeci, and A. Poulsen, 1991, Event-Study Methodology Under Conditions of Event-Induced Variance, Journal of Financial Economics, 30(2): 253-272.
Brown, S. J., and J. B. Warner, 1980, Measuring Security Price Performance, Journal of Financial Economics, 8(3): 205-258.
Chandra, R., S. Moriarity, and L. G. Willinger, 1990, A Reexamination of the Power of Alternative Return-Generating Models and the Effect of Accounting for Cross-Sectional Dependencies in Event Studies, Journal of Accounting Research, 28(2): 398-408.
Chen, H., J. D. Cummins, K. S. Viswanathan, and M. A. Weiss, 2014, Systemic Risk and the Interconnectedness Between Banks and Insurers: An Econometric Analysis, Journal of Risk and Insurance, 81(3): 623-652.
Cole, C. R., and K. A. McCullough, 2006, A Reexamination of the Corporate Demand for Reinsurance, Journal of Risk and Insurance, 73(1): 169-192.
Cole, C. R., and K. A. McCullough, 2008, A Comparative Analysis of US Property and Casualty Reinsurers and Insurers, Risk Management and Insurance Review, 11(1): 179-207.
Cole, C., K. McCullough, and L. Powell, 2010, Collateralization of International Reinsurance Liabilities in the U.S. Insurance Industry, Insurance Markets and Companies: Analyses and Actuarial Computations, 1(2): 32-38.
Cowan, A., 1992, Nonparametric Event Study Tests, Review of Quantitative Finance and Accounting, 2: 343-358.
Cummins, J. D., 2007, Reinsurance for Natural and Man-Made Catastrophes in the United States: Current State of the Market and Regulatory Reforms, Risk Management and Insurance Review, 10(2): 179-220.
Cummins, J. D., S. E. Harrington, and R. Klein, 1995, Insolvency Experience, Risk-Based Capital, and Prompt Corrective Action in Property-Liability Insurance, Journal of Banking and Finance, 19(3/4): 511-527.
Cummins, J. D., C. M. Lewis, and R. Wei, 2006, The Market Value Impact of Operational Loss Events for US Banks and Insurers, Journal of Banking and Finance, 30: 2605-2634.
Cummins, J. D., and M. A. Weiss, 2000, The Global Market for Reinsurance: Consolidation, Capacity, and Efficiency, Brookings-Wharton Papers on Financial Services, 2000: 159-222.
Cummins, J. D., and M. A. Weiss, 2004, Consolidation in the European Insurance Industry: Do Mergers and Acquisitions Create Value for Shareholders? The Brookings/Wharton Conference: Public Policy Issues Confronting the Insurance Industry.
Cummins, J. D., and M. A. Weiss, 2014, Systemic Risk and the U.S. Insurance Sector, Journal of Risk and Insurance, 81(3): 489-527.
Cummins, J. D., and X. Xie, 2008, Mergers and Acquisitions in the US Property-Liability Insurance Industry: Productivity and Efficiency Effects, Journal of Banking and Finance, 32(1): 30-55.
Doherty, N. A., and R. D. Phillips, 2002, Keeping Up With the Joneses: Changing Rating Standards and the Buildup of Capital by U.S. Property-Liability Insurers, Journal of Financial Services Research, 21: 55-78.
Doherty, N. A., A. V. Kartasheva, and R. D. Phillips, 2012, Information Effect of Entry Into Credit Ratings Market: The Case of Insurers' Ratings, Journal of Financial Economics, 106(2): 229-446.
Driscoll, J. C., and A. C. Kraay, 1998, Consistent Covariance Matrix Estimation With Spatially Dependent Panel Data, Review of Economics and Statistics, 80: 549-560.
Feldblum, S., 2002, Reinsurance Accounting: Schedule F, Casualty Actuarial Society Forum, Fall 2002: 685-776.
Geneva Association, 2010, Special Report of the Geneva Association Systemic Risk Working Group, Systemic Risk in Insurance: An Analysis of Insurance and Financial Stability, Geneva, Switzerland, March 2010.
Ghosh, C., and J. I. Hilliard, 2012, The Value of Contingent Commissions in the Property-Casualty Insurance Industry: Evidence From Stock Market Returns, Journal of Risk and Insurance, 79(1): 165-192.
Grace, M. F., 2010, The Insurance Industry and Systemic Risk: Evidence and Discussion, Networks Financial Institute Policy Brief No. 2010-PB-02.
The Group of Thirty, 2006, Reinsurance and International Financial Markets (Washington, DC: The Group of Thirty).
Halek, M., and D. Eckles, 2010, Effects of Analysts' Ratings on Insurer Stock Returns: Evidence of Asymmetric Responses, Journal of Risk and Insurance, 77(4): 801-827.
Harrington, S. E., 2009, The Financial Crisis, Systemic Risk, and the Future of Insurance Regulation, Journal of Risk and Insurance, 76(4): 785-819.
Harrington, S. E., G. Niehaus, and T. Yu, 2012, Insurance Price Volatility and Underwriting Cycles, in: D. Georges, ed., Handbook of Insurance, 2nd Edition (Boston, MA: Kluwer).
Hoechle, D., 2007, Robust Standard Errors for Panel Regressions With Cross-Sectional Dependence, The Stata Journal, 7(3): 281-312.
International Association of Insurance Supervisors, 2012, Ending "Too-Big-To-Fail": FSB Progress Report to the G20. World Wide Web: http://recoveryandresolutionplans. wordpress.com/category/international-association-of-insurance-supervisors/ (accessed August 4, 2012).
International Monetary Fund, 2004, Global Financial Stability Report, World Economic and Financial Surveys: Market Developments and Issues (Washington, DC: International Monetary Fund.
Kartasheva, A. V., and S. Park, 2011, Real Effects of Changing Rating Standards for Catastrophic Risks, Working Paper, The Wharton School, University of Pennsylvania.
Kessler, D., 2014, Why (Re)insurance Is Not Systemic, Journal of Risk and Insurance, 81(3): 477-487.
The London Working Group, 2002, Report: Prudential Supervision of Insurance Undertakings, Conference of the Insurance Supervisory Services of the Member States of the European Union (London, UK).
MacKinlay, A. C., 1997, Event Studies in Economics and Finance, Journal of Economic Literature, 35: 13-39.
McDonnell, W., 2002, Why Some Insurers Fail: Practical Lessons From Recent Cases in Europe, FSA Occasional Paper, December 2002, London.
McHugh, C. C., 2012, NAIC Reinsurance Collateral Reform, Sidley Austin LLP Insurance and Reinsurance Law Report No. 01/2012: 22-28.
Reynolds Porter Chamberlain LLP, 2007, "Hasta la vista baby"--Special Termination Clauses, Reinsurance Update, August: 1-6.
Rossi, M.-L., and N. Lowe, 2002, Regulating Reinsurance in the Global Market, Geneva Papers on Risk & Insurance--Issues & Practice, 27(1): 122-133.
Schwartzman, J. A., 2008, The Game of "Pass the Risk": Then and Now, in: D. Mango, M. Altschull, D. Ingram, and W. Fisher, eds., Risk Management: The Current Financial Crisis, Lessons Learned and Future Implications. World Wide Web: www.soa.org/library/essays/rm-essay-2008.pdf.
Swiss Re, 2003, Reinsurance--A Systemic Risk? Sigma No. 5/2003 (Zurich, Switzerland: Swiss Re).
Weil Financial Regulatory Reform Center, 2012, FSB Identifies Global SIFIs. World Wide Web: http://financial-reform.weil.com/prudential-regulation/fsb-identifiesglobal-sifis/#axzz25ut7AdEU (accessed August 4, 2012).
Zellner, A., 1962, An Efficient Method of Estimating Seemingly Unrelated Regressions and Tests for Aggregation Bias, Journal of the American Statistical Association, 57: 348-368.
(1) Munich Re, Swiss Re, Berkshire Hathaway Reinsurance, Hannover Re, and XL Capital.
(2) Our analysis starts from 2003, but we use 2002 data to create lagged and difference variables.
(3) Professional reinsurers underwrite little to no direct business and tend to contract with other reinsurance firms. We adopt A.M. Best's definition in defining professional reinsurers; that is, if a firm's reinsurance assumed from unaffiliated firms is more than 75 percent of the sum of reinsurance assumed from affiliates and its direct premiums written, then it is classified as a professional reinsurer (Cole and McCullough, 2008).
(4) An earlier version of the paper had included these firms in the downgrading regression analyses, and we reached the same conclusion.
(5) We have also examined which ceding firms tend to utilize more of the unrated reinsurance transactions. The 300 firms out of 7,289 have more than 50 percent of their reinsurance recoverable unrated. We performed a mean test (f-test) to compare the differences in firm characteristics between insurers with more than 50 percent unrated reinsurance and those with less than 50 percent unrated reinsurance. We find that insurers with more unrated reinsurance are smaller in asset size, hold smaller amount of policyholders' surplus, incur a lower combined ratio, and have a slightly lower financial strength rating. The test results are available from authors.
(6) The result is reported in the Appendix (Model 7--Subsample).
(7) We perform a separate analysis for professional reinsurers, which shows that, unlike nonreinsurers, professional reinsurers cede their premiums mostly to alien affiliated reinsurers (e.g., 58.5 percent in 2009) and spread the rest almost equally to the other three types of reinsurers. A similar pattern is observed for net recoverable, with the percentage of alien affiliated reinsurers increasing over time. Major reinsurance groups usually operate globally, and many professional reinsurers in the United States are subsidiaries of these groups. As a result, it is not surprising that these reinsurers cede their business to their non-U.S. affiliates to seek intragroup risk diversification.
Compared to nonreinsurer ceding insurers, professional reinsurers have less aggregated reinsurance (retrocession) exposure. For example, the net reinsurance recoverable over surplus ratio for this group of firms is only 63.26 percent in 2009, suggesting that U.S. professional reinsurers in aggregate maintain strong capital ability. However, these professional reinsurers are exposed to higher credit risks from unaffiliated reinsurers than the nonreinsurer ceding companies. For example, in nonreinsurer ceding companies' portfolios, about 15.6 percent of net reinsurance recoverable is from unaffiliated reinsurers, whereas the percentage for the professional reinsurer was 30.7 percent. It is also worth noting that the credit risk for professional reinsurers is sensitive to catastrophic losses. The net reinsurance recoverable over surplus ratio was as high as 105 percent in 2002 and 195.49 percent in 2005, following the huge losses from the September 11th terrorist attacks and Hurricane Katrina, respectively, which depleted reinsurers' capital significantly.
(8) On November 6,2011, the NAIC Executive Committee and Plenary adopted the revised Credit for Reinsurance Model Law that intends to reduce the collateral requirements for unauthorized reinsurance (McHugh, 2012). Despite the consistency it will bring across the U.S. insurance-reinsurance market and the alignment with the international insurance market, this reform, from the perspective of solvency regulation, may increase the interconnectedness between ceding insurers and reinsurers. The analyses in our current article will not be undermined by the new regulation; instead, it will be reinforced under the new rules.
(9) For the sake of brevity, the summary statistics tables are not shown in the article but are available from the authors upon request. Generally, we find that the nonprofessional reinsurer ceding companies depend very little on unauthorized reinsurance. The amount of total ceded premiums to unauthorized reinsurers averages about only 9 percent of the total direct premiums written, and the total net reinsurance recoverable from unauthorized reinsurers averages about 16 percent of the policyholder surplus of the U.S. P/C industry. In addition, a very small amount of unauthorized reinsurance recoverable is from U.S. unaffiliated reinsurers. An average 43 percent of unauthorized reinsurance recoverable is from alien unaffiliated insurers (with a decreasing trend over time, e.g., the percentage for 2009 is only 26.5), which averages about 7 percent of the industry surplus (4.44 percent in 2009).
The analysis of professional reinsurers shows that they are more likely to cede business to unauthorized reinsurers. The net unauthorized reinsurance recoverable over surplus ratio for this group of firms is 39.91 percent in 2009. A detailed examination of the type of unauthorized reinsurers shows that about 84.6 percent of unauthorized reinsurance recoverable is from alien affiliated reinsurers, and about 14.1 percent is from alien unaffiliated reinsurers.
(10) We calculate the Pearson and Spearman correlation of the two types of indices and find the correlation is above 80 percent for all reinsurers.
(11) The U.S. Department of Justice generally consider markets in which the Herfindahl index is between 1,500 (0.15 in our context) and 2,500 points (0.25 in our context) to be moderately concentrated, and consider markets in which the Herfindahl index is in excess of 2,500 points to be highly concentrated.
(12) We also analyze the diversification of reinsurance portfolios for U.S. professional reinsurers only. This mostly reflects the retrocession activities of these companies, since such firms have little direct business. Overall, professional reinsurers' retrocession portfolios are more diversified (with a mean Herfindahl index 0.533 in 2009) than those of ceding insurers that are not professional reinsurers. Only a small proportion of professional reinsurers (about 35 percent) retrocede to their affiliates (domestic or alien), and the affiliated reinsurance portfolio is very concentrated (mean Herfindahl index close to 0.9, with an upward trend from 2002 to 2009). More than 90 percent of professional reinsurers retrocede to U.S. unaffiliated reinsurers, and that portfolio is more diversified, with a downward trending Herfindahl index (0.506 in 2002 and 0.448 in 2009). Percentage-wise, an increasing number of U.S. professional reinsurers (63 percent in 2002 and 77 percent in 2009) retrocede to alien unaffiliated reinsurers, and this type of reinsurance portfolio is the most diversified (mean Herfindahl index 0.426 in 2009). Overall, professional reinsurers tend to be more diversified in unaffiliated reinsurance but very concentrated in affiliated reinsurance usage.
(13) As many insurers in the same group receive the pooled group rating, most rating changes of affiliated reinsurers and primary insurers occur on the same day. Out of concern that affiliated reinsurance downgrade variable in such cases may capture other effects, we do not count these cases as reinsurer downgrades when calculating RDownRec_Aff. However, such treatment may have ignored the quality of reinsurance. As a robustness check, we count such cases as "reinsurer downgrade" and perform the regression analysis (see the Appendix, model 8). Our main conclusion still holds.
(14) We also run all the regressions with two separate control variables of authorized affiliated reinsurance and unauthorized affiliated reinsurance. As there is a very limited number of unauthorized affiliated reinsurer downgrade cases, the unauthorized affiliated reinsurance variable drops out from logit regression due to perfect prediction. As a result, we could only keep the authorized affiliated reinsurance variable in the regression model. This variable is statistically insignificant and our main results remain robust. We also perform analyses without affiliated reinsurance variables. The results are robust as well.
(15) We convert the A.M. Best rating as follows: A++ = 13, A+ = 12, A = 11, ..., and D = 1.
(16) We also run a regression using lagged control variables instead of the difference variables as a robustness check in case the level of these variables matters more than the change. This result is reported in the Appendix. Our main conclusion remains the same in both specifications.
(17) There can be more explanatory variables associated with ratings such as all other Financial Analysis and Surveillance Tracking (FAST) scores. We included all FAST scores in the first-round ordered probit model but many of these variables are highly correlated with each other, creating a multicollinearity problem. We only keep the significant variables and delete similar variables. For example, we keep net premiums written (NPW) to surplus ratio and discard direct premiums written to surplus ratio. We also conducted the rest of our analysis with a few more control variables, such as [DELTA]2 year development/PHS (policyholders' surplus), [DELTA]other investment/PHS, [DELTA]NPW growth, and [DELTA]affiliated investment/PHS. Adding more control variables does not change the main results. The results are available from the authors upon request.
(18) A total number of 1,671 reinsurer-year observations were recorded as downgrading during the sample period, and these reinsurers affected 5,183 primary insurer-year observations.
(19) One observation turned out to be an outlier with a too large residual. We exclude this one observation in all analyses. However, including or excluding this one observation does not affect our main variable of interest.
(20) The average (median) length between reinsurer downgrades and primary insurer downgrades is 169 (150) days.
(21) Due to the very limited number of downgrades occurring within 30 days, RDownRec_UN* Single variable drops out in the logit regression model. There are only three observations, which have positive RDownRec_UN*Single and all three predict no downgrade. RDownRec_UN also causes quasi-complete separation problem in the logit regression due to the very limited number of positive observations. We therefore drop RDownRec_UN variable as well. However, including this variable in the regression shows almost the same result except that the coefficient of RDownRec_UN becomes very large. OLS regression with all variables also shows qualitatively the same result.
(22) We also tried 50 and 100 days, and the results are qualitatively the same.
(23) As another robustness check, we additionally conduct a regression with leading firms only and a newly constructed affiliated reinsurance downgrade variable that includes affiliated reinsurers downgrades on the same day. The result is robust, and available from authors upon request.
(24) We do not present the fixed effects model as a main result because the logit regression model with the fixed firm effects removes 75 percent of our observations. We instead run the GLS model with fixed effects (column 1) and random effects (column 2) and present the result in the Appendix. Random effects could be run with a logit regression, and the results are consistent with our main conclusion. We also run a regression with AR(1) model (column 3) and with Driscoll and Kraay (1998) robust standard errors (column 4) adjusting for both heteroskedasticity and autocorrelation. The Driscoll and Kraay error model should be applied cautiously when a panel contains a large cross-section but a very short time series (Hoechle, 2007), which is the case in our panel data, so we present this result in the Appendix and do not use it as a main result.
(25) We have recorded 238 different downgrading date announcements that include publicly traded counterparty primary insurers, so SUR regression is run individually for the 238 events. The number of individual companies involved in each event varies. On average, 18 publicly traded counterparty primary insurers are affected by each event, and the average number for noncounterparty primary insurers is 103.
(26) An earlier version of the paper conducted a standard event study utilizing the market model (MacKinlay, 1997) to measure abnormal returns. The CAR results are quantitatively similar. See Boehmer, Musumeci, and Poulsen (1991) and Cowan (1992) for the explanation of event study methodology and statistical significance tests. To address the concern of the cross-sectional correlation caused by clustering of firms around a single event date, we also constructed the Portfolio Time-Series CDA t-test (Brown and Warner, 1980; Chandra, Moriarity, and Willinger, 1990). As an additional robustness check, we also ran an event study by forming a portfolio of firms for each downgrade announcement and used portfolio returns instead of individual stock returns (see Ghosh and Hilliard, 2012, for more discussion). The results are quantitatively similar.
(27) We focus on the longer window of CAR, such as a (-15, +15) window, following the work of Cummins, Lewis, and Wei (2006), which finds that insurers usually experience an impact over a longer window regarding operational loss announcements (they use a (--20, +20) window in the analyses). In addition, the longer window can accommodate the possibility that, before an official announcement of downgrades, rumors may already spread in the market. For example, on April 15,2003, Standard & Poor's Ratings Services said that "... today it lowered its long-term counterparty credit and insurer financial strength ratings on Germany-based reinsurer Hannover Ruckversicherungs AG (Hannover Re) and other core members of the Hannover Re group to 'AA-' from 'AA'. At the same time, all ratings were removed from CreditWatch ..." That is, for some rating downgrades, even before the announcements, the market may already have some negative expectations.
(28) We conducted a robustness check regarding whether the above results hold or become stronger for "threshold rating downgrades." Following Halek and Eckles (2010), we define a threshold rating downgrade as losing an A- (A.M. Best), an Aa3 (Moody's), or an AA- (S&P). For ratings downgrade analyses, we find no significant result for this particular set of threshold downgrading. In regard to the stock market reaction, for counterparty primary insurers, the announcing effect of threshold downgrading is similar to the overall sample, but we do find a stronger contagion effect for threshold downgrading announcements.
(29) In reality, the real loss will vary firm by firm. We assume the same percentage of loss for simplicity.
(30) The hypothesized explanatory variables only adjust for the defaulted recoverable. However, it is possible that other insurer characteristics could change if the cause of reinsurer collapse is due to a catastrophic or macroeconomic event, which affects the entire insurance industry. In this analysis, we cannot incorporate that point. Therefore, the result should be interpreted with caution.
(31) We compare the hypothetical ratings with the original estimated ratings instead of the actual ratings because the difference between actual and hypothetical estimated ratings contains both the increased risk and unavoidable modeling error. A comparison of estimated ratings both before and after the reinsurer insolvency event will return a more consistent result.
(32) Since we can only include those insurers with an A.M. Best rating and no missing explanatory variables in the ratings regressions, the total number of insurers used in this analysis is 1,367.
Sojung Carol Park is an Assistant Professor of Finance at the College of Business Administration, Seoul National University, 1 Gwanak-ro, Gwanak-gu, Seoul 151-916, Korea. Park can be contacted via e-mail: email@example.com. Xiaoying Xie is an Associate Professor of Finance at the Mihaylo College of Business and Economics, California State University, Fullerton, Fullerton, CA 92834-6848. Xie can be contacted via e-mail: firstname.lastname@example.org. Sojung C. Park acknowledges support from the Institute of Management Research at Seoul National University, the Institute of Finance and Banking at Seoul National University, and the Research Settlement Fund for the new faculty of SNU. Xiaoying Xie acknowledges support from the GCAT research fund of the Center for Insurance Studies, and a research grant from Mihaylo College of Business and Economics of California State University, Fullerton. We are very grateful to the two anonymous referees for their valuable comments and suggestions.
TABLE 1 Dependence of U.S. P/C Insurers on Reinsurance by Reinsurer Type Panel A: Ceding Insurers, All Types of Reinsurers Direct Total Premiums Ceded Total Net Written Year Premiums Recoverable (DPW) 2002 320,464 561,798 402,471 2003 349,209 610,754 443,484 2004 364,980 640,440 463,514 2005 388,210 729,528 476,461 2006 395,824 716,290 494,105 2007 402,798 717,355 496,606 2008 405,683 739,650 486,857 2009 400,790 735,134 473,167 Total Ceded Total Net Premiums/ Recoverable/ Year Surplus DPW (%) Surplus (%) 2002 304,803 79.62 184.32 2003 365,349 78.74 167.17 2004 413,152 78.74 155.01 2005 481,048 81.48 151.65 2006 516,245 80.11 138.75 2007 554,372 81.11 129.40 2008 506,222 83.33 146.11 2009 559,895 84.70 131.30 Panel B: U.S. Affiliated Reinsurer Total Ceded Net Ceded Year Premiums Recoverable Premiums (%) 2002 236,331 372,242 73.7 2003 257,192 405,367 73.6 2004 276,220 430,375 75.7 2005 300,778 501,696 77.5 2006 307,760 508,772 77.8 2007 313,543 519,032 77.8 2008 313,967 531,621 77.4 2009 309,899 535,319 77.3 Total Net Ceded Net Recoverable Premiums/ Recoverable/ Year (%) DPW (%) Surplus (%) 2002 66.3 58.72 122.13 2003 66.4 57.99 110.95 2004 67.2 59.59 104.17 2005 68.8 63.13 104.29 2006 71.0 62.29 98.55 2007 72.4 63.14 93.63 2008 71.9 64.49 105.02 2009 72.8 65.49 95.61 Panel C: Alien Affiliated Reinsurer Total Ceded Ceded Net Premiums Year Premiums Recoverable (%) 2002 16,031 24,499 5.0 2003 20,688 31,578 5.9 2004 22,518 36,492 6.2 2005 25,198 44,670 6.5 2006 26,247 46,405 6.6 2007 26,261 47,534 6.5 2008 26,606 54,464 6.6 2009 27,941 53,867 7.0 Total Net Ceded Net Recoverable Premiums/ Recoverable/ Year (%) DPW (%) Surplus (%) 2002 4.4 3.98 8.04 2003 5.2 4.66 8.64 2004 5.7 4.86 8.83 2005 6.1 5.29 9.29 2006 6.5 5.31 8.99 2007 6.6 5.29 8.57 2008 7.4 5.46 10.76 2009 7.3 5.91 9.62 Panel D: U.S. Unaffiliated Reinsurer Total Ceded Ceded Net Premiums Year Premiums Recoverable (%) 2002 40,519 92,469 12.6 2003 40,887 97,311 11.7 2004 36,519 95,243 10.0 2005 32,062 96,503 8.3 2006 31,755 90,075 8.0 2007 30,504 82,647 7.6 2008 32,272 82,258 8.0 2009 31,896 79,970 8.0 Total Net Ceded Net Recoverable Premiums/ Recoverable/ Year (%) DPW (%) Surplus (%) 2002 16.5 10.07 30.34 2003 15.9 9.22 26.64 2004 14.9 7.88 23.05 2005 13.2 6.73 20.06 2006 12.6 6.43 17.45 2007 11.5 6.14 14.91 2008 11.1 6.63 16.25 2009 10.9 6.74 14.28 Panel E: Alien Unaffiliated Reinsurer Total Ceded Ceded Net Premiums Year Premiums Recoverable (%) 2002 18,985 47,240 5.9 2003 22,338 49,310 6.4 2004 21,409 47,677 5.9 2005 21,956 49,738 5.7 2006 21,520 40,617 5.4 2007 23,962 38,525 5.9 2008 24,750 40,637 6.1 2009 23,312 34,395 5.8 Total Net Ceded Net Recoverable Premiums/ Recoverable/ Year (%) DPW (%) Surplus (%) 2002 8.4 4.72 15.50 2003 8.1 5.04 13.50 2004 7.4 4.62 11.54 2005 6.8 4.61 10.34 2006 5.7 4.36 7.87 2007 5.4 4.83 6.95 2008 5.5 5.08 8.03 2009 4.7 4.93 6.14 Notes: Based on industry aggregates, but professional property-casualty reinsurers are excluded from the analysis. We define "professional reinsurer" using A.M. Best's definition. That is, if a firm's reinsurance assumed from unaffiliated firms is more than 75 percent of the sum of the reinsurance assumed from affiliates and its direct premiums written, then it is defined as a professional reinsurer (Cole and McCullough, 2008). Unit: $ million. TABLE 2 Diversification of Reinsurance Portfolios--Herfindahl Index by Type of Reinsurer for U.S. P/C Firms (Professional Reinsurers Excluded) By Reinsurance Premiums Ceded U.S. Alien U.S. Alien All Affiliated Affiliated Unaffiliated Unaffiliated Year Reinsurers Reinsurer Reinsurer Reinsurer Reinsurer Mean 2002 0.679 0.924 0.893 0.569 0.426 2003 0.670 0.928 0.870 0.571 0.409 2004 0.662 0.930 0.880 0.564 0.384 2005 0.650 0.928 0.872 0.551 0.387 2006 0.641 0.922 0.888 0.543 0.371 2007 0.637 0.924 0.894 0.543 0.361 2008 0.636 0.927 0.897 0.536 0.359 2009 0.636 0.929 0.877 0.548 0.352 Median 2002 0.814 1 1 0.506 0.311 2003 0.778 1 1 0.502 0.280 2004 0.771 1 1 0.496 0.256 2005 0.737 1 1 0.473 0.252 2006 0.731 1 1 0.463 0.226 2007 0.725 1 1 0.461 0.219 2008 0.707 1 1 0.467 0.215 2009 0.718 1 1 0.483 0.216 Number of ceding firms 2002 2163 1264 203 1571 967 2003 2159 1235 197 1544 1021 2004 2180 1227 215 1520 1095 2005 2202 1235 239 1499 1164 2006 2233 1235 250 1532 1233 2007 2259 1257 248 1558 1257 2008 2301 1267 293 1564 1298 2009 2313 1299 297 1566 1272 By Net Reinsurance Recoverable U.S. Alien U.S. Alien All Affiliated Affiliated Unaffiliated Unaffiliated Year Reinsurers Reinsurer Reinsurer Reinsurer Reinsurer Mean 2002 0.657 0.925 0.880 0.543 0.484 2003 0.653 0.924 0.889 0.533 0.486 2004 0.643 0.924 0.888 0.529 0.468 2005 0.635 0.926 0.868 0.524 0.440 2006 0.636 0.923 0.882 0.525 0.462 2007 0.639 0.921 0.877 0.528 0.466 2008 0.637 0.923 0.879 0.529 0.442 2009 0.639 0.925 0.872 0.532 0.449 Median 2002 0.729 1 1 0.490 0.395 2003 0.719 1 1 0.471 0.405 2004 0.696 1 1 0.461 0.359 2005 0.683 1 1 0.444 0.333 2006 0.687 1 1 0.448 0.348 2007 0.685 1 1 0.457 0.350 2008 0.680 1 1 0.446 0.315 2009 0.686 1 1 0.440 0.349 Number of ceding firms 2002 2168 1298 221 1571 861 2003 2178 1284 231 1563 920 2004 2212 1291 249 1567 988 2005 2263 1298 264 1597 1087 2006 2298 1304 279 1623 1174 2007 2303 1321 279 1618 1162 2008 2353 1335 311 1661 1259 2009 2357 1360 312 1645 1215 Table 3 Rating Determinants--Ordered Probit Regression Model, 2002-2009 Expected Variables Definitions Sign Estimates Investment Annualized return on + 0.0271 *** yield investment based on (0.0055) average invested assets Junk bond/PHS Total junk bonds in asset - -0.0068 *** to surplus ratio (0.0017) CAT risk The proportion of - -0.4114 *** catastrophic risk (0.0677) exposure: defined as direct premiums written in homeowners, farm owners, auto physical damage, commercial multiperil, or inland marine in AL, FL, MS, SC, or TX to total premiums written NPW/PHS Net premiums written to - -0.0018 *** surplus ratio (0.0002) Reserve/PHS Reserve to surplus ratio - -0.0033 *** (0.0002) Combined ratio Underwriting expenses/net - -0.0008 *** premiums written + loss (0.0001) and loss adjustment expenses incurred / premiums earned Reinsurance Reinsurance recoverable to - -0.0023 *** recoverable/ surplus ratio (0.0001) PHS BCAR Best's Capital Adequacy + 0.0009 *** Ratio (0.0001) Log(Asset) Log value of insurer's + 0.3624 *** admitted assets (0.0068) Public Dummy variable equal to 1 + 0.5758 *** if the insurer (or its (0.0226) parent) is publicly traded, 0 otherwise Single A dummy variable equal to - -0.2778 *** 1 for single (0.0306) unaffiliated company, 0 otherwise Age Firm age + 0.0010 *** (0.0003) Intercepts Not reported Number of 11,808 observations Likelihood 5,673.4 ratio Notes: Standard errors are in brackets. Significant at the *** 1% level. TABLE 4 Summary Statistics of Variables Affecting Rating Downgrades Variables Definitions Mean STD PDown 1 if a primary insurer's 0.032 0.177 A.M. Best rating downgrades in year t, 0 otherwise RDownRec_AU The proportion of authorized 0.0275 0.126 reinsurance recoverable from the downgraded unaffiliated reinsurers to the surplus of the primary insurer RDownRec_UN The proportion of 0.0056 0.0487 unauthorized reinsurance recoverable from the downgraded unaffiliated reinsurers to the surplus of the primary insurer Single 1 if a primary insurer is an 0.162 0.368 unaffiliated single insurer, 0 otherwise Single * Interaction term of 0.0032 0.0411 RDownRec_AU RDownRec_AU and Single Single * Interaction term of 0.0003 0.0103 RDownRec_UN RDownRec_UN and Single [DELTA]Investment Change in investment yield -0.111 2.005 yield (%) from year t--1 to year f [DELTA]Junk bond/ Total junk bonds in assets -0.043 3.643 PHS (%) to surplus ratio change from year t--1 to year t [DELTA]CAT risk Change in the proportion 0.0002 0.035 of catastrophic risk exposed lines of business from year t--1 to t [DELTA]NPW/PHS (%) Net premiums written to -6.719 40.03 surplus ratio change from year t--1 to year f [DELTA]Reserve/ Reserve to surplus ratio -1.706 33.11 PHS (%) change from year t--1 to year f [DELTA]Combined Combined ratio change -0.227 24.33 ratio (%) from year t--1 to year t [DELTA]Reinsurance Reinsurance recoverable to -2.153 42.76 recoverable/ surplus ratio change from PHS (%) year t--1 to year t [DELTA]BCAR Best's Capital Adequacy 7.348 72.96 Ratio change from year t--1 to year t [DELTA]Log(Asset) Log(Asset) change from year 0.076 0.174 t--1 to year t Best ratingff--1) Numerical conversion of 10.22 1.556 Best's rating in year t--1 RRUN_TTREC Unauthorized reinsurance 0.1075 0.0032 recoverable to total reinsurance recoverable RDown Rec_Aff The proportion of 0.0057 0.1250 reinsurance recoverable from the downgraded affiliated reinsurers to the surplus of the primary insurer Ind_CombinedR Industry combined ratio 0.977 0.409 Variables Min. Max. 1% 99% PDown 0 1 0 1 RDownRec_AU 0 3.939 0 0.562 RDownRec_UN 0 1.505 0 0.118 Single 0 1 0 1 Single * 0 1.706 0 0.0574 RDownRec_AU Single * 0 0.8398 0 0.0004 RDownRec_UN [DELTA]Investment -39.7 52.8 -3.2 3.3 yield (%) [DELTA]Junk bond/ -66.6 133.99 -8.76 9.69 PHS (%) [DELTA]CAT risk -0.877 0.864 -0.064 0.066 [DELTA]NPW/PHS (%) -531 686 -126 107 [DELTA]Reserve/ -637 1285 -78 76 PHS (%) [DELTA]Combined -114 120 -114 113 ratio (%) [DELTA]Reinsurance -659.9 604.9 -135.56 126.5 recoverable/ PHS (%) [DELTA]BCAR -842.9 872 -198.3 197.2 [DELTA]Log(Asset) -1.625 2.062 -0.338 0.684 Best ratingff--1) 2 13 5 13 RRUN_TTREC 0 1 0 1 RDown Rec_Aff 0.0000 5.2012 0.0000 0.0003 Ind_CombinedR 0.92 1.04 0.92 1.04 Table 5 The Impact of Reinsurer Ratine Downgrades on Primary Insurer Rating Downgrades, 2003-2009 Expected (1) No (2) Main Variables Sign Interaction Model RDownRec_UN +/- -0.4757 -0.5675 (1.3247) (1.4097) RDownRec_AU + 0.8580 *** 0.6877 ** (0.2761) (0.2929) Single 0.1202 (0.2052) Single * RDownRec_UN +/- 1.8175 (3.8978) Single * RDownRec_AU + 1.5967 ** (0.7766) [DELTA]Investment yield - -0.0349 -0.0335 (0.0275) (0.0275) [DELTA]CAT risk + 2.1964 2.1810 (1.5734) (1.5758) [DELTA]NPW/PHS + 0.0059 *** 0.0059 *** (0.0017) (0.0017) [DELTA]Reinsurance recoverable/PHS + 0.0020 0.0018 (0.0014) (0.0014) [DELTA]Reserve/PHS + 0.0165 *** 0.0169 *** (0.0024) (0.0024) [DELTA]Junk bond /PHS + -0.0415 ** -0.0429 ** (0.0193) (0.0192) [DELTA]BCAR - -0.0021 *** -0.0021 *** (0.0008) (0.0008) [DELTA]Log(Asset) - -1.2306 *** -1.2458 *** (0.3729) (0.3754) [DELTA]Combined ratio + 0.0076 *** 0.0075 *** (0.0026) (0.0026) Best rating - -0.0896 ** -0.0818 * (f-1) (0.0422) (0.0434) RRUN_TTREC + 1.4649 * 1.7018 * (0.8395) (0.8704) RDownRec_Aff + -5.1806 -5.1725 (9.2935) (9.3083) Ind_CombinedR + 5.4031 *** 5.3668 *** (1.7305) (1.7351) Number of observations 7,289 7,289 Chi-squared 211.9 217.2 (3) (4) Leading Variables 30 Days Only RDownRec_UN -0.6459 (1.9281) RDownRec_AU -0.8805 0.7036 ** (2.0641) (0.3528) Single 0.143 0.0219 (0.2002) (0.2257) Single * RDownRec_UN 2.234 (4.1677) Single * RDownRec_AU 13.3257 ** 1.6738 ** (6.3350) (0.8206) [DELTA]Investment yield -0.0326 0.0311 (0.0276) (0.0843) [DELTA]CAT risk 2.1371 0.7347 (1.5666) (2.3909) [DELTA]NPW/PHS 0.0059 *** 0.0125 *** (0.0017) (0.0025) [DELTA]Reinsurance recoverable/PHS 0.0018 0.0031 (0.0014) (0.0019) [DELTA]Reserve/PHS 0.0170 *** 0.0193 *** (0.0024) (0.0036) [DELTA]Junk bond /PHS -0.0431 ** -0.025 (0.0192) (0.0287) [DELTA]BCAR -0.0020 ** -0.0020 ** (0.0008) (0.0009) [DELTA]Log(Asset) -1.1868 *** -2.3283 *** (0.3761) (0.5335) [DELTA]Combined ratio 0.0075 *** 0.0067 * (0.0026) (0.0035) Best rating -0.0946 ** -0.0063 (f-1) (0.0430) (0.0624) RRUN_TTREC 1.9954 ** 1.4209 (0.8432) (1.4566) RDownRec_Aff -5.2772 -10.047 (9.4839) (34.6006) Ind_CombinedR 5.5702 *** 0.1077 (1.7329) (2.3359) Number of observations 7,289 3,581 Chi-squared 214.9 132.7 Notes: Standard errors are in parentheses. We run 2SLS regressions, treating RRUN_TTREC as an endogenous variable. The first-stage regression result is not reported, but is available from authors upon request. Significant at the *** 1%, ** 5%, and * 10% levels. Table 6 The Impact of Reinsurer on Primary Insurer Stocks Panel A: Counterparty Primary Insurers Mean Median Days N CAR (%) CAR (%) f-Test Sign Test (0,0] 4,279 -0.24 -0.08 -6.86 *** -142.5 *** (-5, +5] 4,279 -0.61 -0.32 -4.99 *** -97.5 *** (-15, +15] 4,279 -1.74 -1.01 -7.72 *** -184.5 *** (-15, -1] 4,279 -0.82 -0.42 -5.55 *** -130.5 *** Panel B: Contagion Effects--Noncounterparty Primary Insurers Mean Median Days N CAR (%) CAR (%) t-Test Sign Test (0, 0] 24,664 -0.19 -0.05 -12.59 *** -769.5 *** (-5, +5] 24,664 -0.26 -0.23 -4.88 *** -504 *** (-15, +15] 24,664 -0.59 -0.27 -6.17 ** -265 *** (-15, -1] 24,664 0.03 -0.15 0.48 -238 ** Panel C: Regression for Counterparty Primary Insurers Variables CAR (-15, +15] RDownRec AU -0.1184 ** (0.0512) RDownRec UN 0.2793 ** (0.1370) % Business with catastrophic risk exposure -0.0874 ** (0.0377) Combined ratio -0.0041 *** (0.0015) % Premium in long-tail lines -0.0309 ** (0.0124) Log (market equity) -0.0035 *** (0.0013) Tobin's Q -0.0586 ** (0.0231) Leverage 0.0286 (0.0230) Industry combined ratio -0.4574 *** (0.0454) Constant 0.5606 *** (0.0568) Number of observations 4,007 Adj. [R.sup.2] 0.0303 Notes: This table shows mean and median cumulative abnormal returns (CARs) of primary insurers in response to reinsurer downgrade announcements. In total, 238 downgrading event announcements are included in the analysis. The impact of reinsurers' downgrades on their counterpart primary insurers is shown in Panel A. The impact of reinsurers' downgrades on noncounterparty primary insurers is shown in Panel B. Day 0 is the day a reinsurer downgrade is announced by a ratings agency. Abnormal returns are calculated using the seemingly unrelated regression (SUR) method. The factors affecting counterparty primary insurers' CAR is presented in Panel C, with standard errors in parentheses. Significant at the *** 1 %, ** 5%, and * 10% levels. TABLE 7 Scenario Analysis: Number of Hypothetically Downgraded Insurers BCAR Constant 100% 50% 30% 10% Loss Loss Loss Loss Munich Re Same surplus 16 11 7 1 30% shock 21 10 6 4 Swiss Re Same surplus 31 12 4 1 30% shock 41 26 19 9 Berkshire Same surplus 16 9 7 0 30% shock 21 9 5 0 All three Same surplus 67 35 18 3 30% shock 66 40 25 7 Any- Same surplus 212 128 82 28 unaffiliated 30% shock 269 179 122 42 BCAR_new = BCAR*new Surplus/Surplus 100% 50% 30% 10% Loss Loss Loss Loss Munich Re Same surplus 22 13 8 1 30% shock 23 11 9 6 Swiss Re Same surplus 36 15 8 1 30% shock 45 28 20 12 Berkshire Same surplus 18 11 8 1 30% shock 29 14 10 6 All three Same surplus 70 42 25 4 30% shock 72 42 27 11 Any- Same surplus 240 147 96 36 unaffiliated 30% shock 279 187 129 51 Notes: Total number of insurers with financial strength rating is 1,367. The "30% shock" rows report the incremental number of downgrades caused by reinsurers' insolvency. The total number of "30% shock" should be 294 firms (downgrades caused by the 30% surplus depletion itself) plus the number reported in the table. TABLE 8 Scenario Analysis: Number of Hypothetically Insolvent Insurers 100% Loss 50% Loss Direct Chain Direct Chain Effect Effect Effect Effect Munich Re Negative surplus 5 5 3 3 RBC 200% 13 16 8 8 Negative surplus, 7 7 4 4 30% shock RBC 200%, 30% 27 101 13 30 shock Swiss Re Negative surplus 9 10 4 4 RBC 200% 25 28 17 19 Negative surplus, 15 17 7 7 30% shock RBC 200%, 30% 47 144 21 47 shock Berkshire Negative surplus 5 7 1 1 RBC 200% 17 22 7 8 Negative surplus, 11 15 3 3 30% shock RBC 200%, 30% 18 80 8 24 shock All three Negative surplus 26 29 8 8 RBC 200% 55 94 28 30 Negative surplus, 44 69 15 15 30% shock RBC 200%, 30% 91 207 44 99 shock All Negative surplus 154 214 60 61 unaffiliated RBC 200% 235 338 129 154 Negative surplus, 307 460 154 189 30% shock RBC 200%, 30% 441 830 259 433 shock 30% Loss 10% Loss Direct Chain Direct Chain Effect Effect Effect Effect Munich Re Negative surplus 3 3 1 1 RBC 200% 5 5 2 2 Negative surplus, 3 3 1 1 30% shock RBC 200%, 30% 7 15 5 5 shock Swiss Re Negative surplus 0 0 0 0 RBC 200% 6 6 1 1 Negative surplus, 2 2 0 0 30% shock RBC 200%, 30% 15 23 5 6 shock Berkshire Negative surplus 1 1 0 0 RBC 200% 5 6 2 3 Negative surplus, 1 1 0 0 30% shock RBC 200%, 30% 4 10 1 2 shock All three Negative surplus 6 6 1 1 RBC 200% 17 19 6 7 Negative surplus, 8 8 2 2 30% shock RBC 200%, 30% 29 43 10 11 shock All Negative surplus 27 27 5 5 unaffiliated RBC 200% 80 91 26 26 Negative surplus, 72 79 8 8 30% shock RBC 200%, 30% 180 245 61 66 shock Notes: Total number of insurers is 2,492. The "30% shock" rows report the incremental number of insolvency caused by reinsurers' insolvency. The total number of "RBC 200%, 30% shock" should be 74 firms (< 200% surplus/RBC ratio caused by the 30% surplus depletion itself) plus the number reported in the table.
|Printer friendly Cite/link Email Feedback|
|Author:||Park, Sojung Carol; Xie, Xiaoying|
|Publication:||Journal of Risk and Insurance|
|Date:||Sep 1, 2014|
|Previous Article:||Forecasting mortgage securitization risk under systematic risk and parameter uncertainty.|
|Next Article:||Systemic risk and the interconnectedness between banks and insurers: an econometric analysis.|