Credit Risk Determinants of Insurance Companies *

Transcription

1 Credit Risk Determinants of Insurance Companies * LILIANA GONZALEZ ESSEC Business School LORENZO NARANJO ESSEC Business School March, 2014 ABSTRACT This paper investigates the determinants of credit risk in insurance companies in the U.S. and Europe. Consistent with recent results for non-financial firms in the U.S., we find that equity volatility is a major determinant and predictor of CDS spreads for both U.S. and European insurers, even after controlling for the composition of their investment portfolios and other firm-specific characteristics such as leverage and macro controls. Furthermore, we find macroeconomic factors to affect the credit risk of European but not U.S. insurers, whereas cash holdings seem to be relevant in explaining the credit spreads of U.S. insurance companies. We find that cash holdings and credit spreads of U.S. insurers are positively correlated. However, the availability of cash reduces the credit risk of firms experiencing positive solvency shocks. Overall, our results are economically significant and suggest that equity and credit markets incorporate quickly relevant information on the creditworthiness of large insurers. Keywords: Insurance Companies, Credit Risk, Credit Default Swaps, Financial Crisis JEL Codes: G11, G12, G22 * We would like to thank Andras Fulop, Rik Sen, Marcela Valenzuela, Patricio Valenzuela, and seminar participants at Universidad de Chile (CEA) for helpful comments and suggestions. We also thank the ESSEC Research Center for financial support. Corresponding Author: Avenue Bernard Hirsch, Cergy, France. naranjo@essec.edu 1

2 1 Introduction The subprime crisis that started in 2007 put the financial sector at risk. Because of their leverage, insurance companies endured severe market stress, and major insurers such as AIG had to be rescued by the U.S. government as they were considered too big to fail. The financial crisis of proved that major insurers can be the source of financial fragility and systemic risk at the global level. The situation has created pressure on both academics and regulators to understand the causes of such credit events. However, little work has been done so far in understanding the drivers of credit risk for large insurance corporations. We contribute to this debate by analyzing the determinants of credit spreads for insurance companies in the U.S. and Europe. Past work on the credit risk of insurance companies has mostly focused in predicting defaults using financial characteristics or Best s ratings 1, but no studies have analyzed the risk of the debt issued by large insurers. Moreover, even though some authors have analyzed the determinants of banks credit spreads (e.g. Annaert et al., 2013), to the best of our knowledge we are the first to look at the determinants of credit spreads of insurance companies. Finally, existing research on CDS spreads for non-financial firms has studied mostly the U.S. market (e.g. Zhang et al., 2009), while research on banks has focused mostly on Europe (e.g. Annaert et al., 2013). In this paper we look both at major U.S. and European insurers together. 1 See for example Trieschmann and Pinches (1973), Shaked (1985), Ambrose and Carroll (1994), Carson and Hoyt (1995), and Lee and Urrutia (1996). 2

3 Insurance companies possess firm-specific characteristics that differentiate them from other firms in the economy, such as the structure of their balance sheet, the regulatory environment, and the way in which they transfer risk to other sectors of the economy. As many financial firms, insurance companies operate under high leverage as the result of their large fraction of insurance liabilities. The resulting leverage on the balance sheet makes their debt risky, increasing the possibility of default and bankruptcy. Such risks are usually exacerbated in times of market stress. As a consequence, the cost of debt is a widely used indicator to assess the financial health of insurers. We believe that it is important to understand the main drivers of credit risk. Our main findings can be summarized as follows. First, we find that that equity volatility is the most important determinant and predictor of credit spreads for U.S. and European insurers, a result that is consistent with recent findings for CDS spreads of U.S. firms (Zhang et al., 2009). We believe that this result is relevant for investment professionals and macro-prudential regulators, because it suggest an additional simple tool to monitor and assess the financial health of large insurers. From a market efficiency perspective, this finding suggests that both debt and equity markets quickly incorporate relevant information on credit events for such companies. The effect we uncover is also economically significant. According to our estimates, a one standard deviation increase in the firm equity volatility can increase CDS spreads by around 1.5% for U.S. insurers and by 0.9% for European insurers. The fact that volatility affects CDS spreads of insurance companies challenges the commonly held view that variables that are known to explain credit spreads of non-financial firms usually 3

4 lose their explanatory power when applied to financial firms (see e.g. Boss and Scheicher, 2002; Raunig and Scheicher, 2009; Grammatikos and Vermeulen, 2012). For example, research on European banks by Annaert et al. (2013) finds no direct relation between firm-specific volatility and credit spreads. Furthermore, we find that the ratios of debt and insurance liabilities to total assets, respectively, and the distance-to-default measure of Vassalou and Xing (2004), are also important determinants of credit spreads for insurance companies. The results hold after controlling for the composition of their portfolio investments, and when we separate the sample between U.S. and European insurance companies. Since the effect of volatility holds after controlling for leverage, this provides indirect evidence that the volatility of assets is relevant in explaining the level of credit risk for insurance entities. We also uncover differences between the credit spread determinants of U.S. and European insurers. First, we find that macroeconomic factors such as the risk-free rate and the swap spread are important in explaining the credit spreads of European insurers. Hence, credit spreads of European insurers comove more with the business cycle than spreads of U.S. insurers. Second, we uncover a strong positive correlation between cash reserves and credit spreads for U.S. insurers. This positive comovement is consistent with the recent findings of Acharya et al. (2012) who show that companies holding cash at optimal levels will do so for precautionary motives. We argue in the paper that regulation might explain why cash holdings are more relevant in explaining the variation of CDS spreads for U.S. insurers. In the U.S., insurance companies are 4

5 required to maintain their adjusted capital above a minimum required level that depends on the risk of their assets and their insurance liabilities. In contrast, under the E.U. directive Solvency I, which was still under place at the end of our sample, insurers are required to keep their adjusted capital above a regulatory level that depends on future premium and claims liabilities. As a consequence, we can expect cash holdings of U.S. insurers to be more informative about their credit condition than cash reserves of European insurers, and to vary more in response to market events. Empirically, we observe that U.S. insurers hold on average four times less cash to total asset than European insurance companies, and that cash reserves for U.S. insurers vary more over time than the ones of their European counterparts. To gain further understanding on the role that cash holdings play for U.S. insurers, we study how credit spreads react for firms that hold a larger proportion of cash after they experience an unexpected improvement in their solvency. To achieve this, we use a differences-in-differences approach in which we perform a cross-sectional comparison of cash holdings of firms which have unexpectedly improved their financial position compared to those who have not. Our results are robust to several specifications, and confirm the intuition that companies holding more cash become safer if their financial health suddenly improves. The rest of the paper is organized as follows. In Section 2 we describe the data and variables that we use in the empirical analysis. In Section 3 we analyze the determinants of CDS spreads for U.S. and European insurance firms. Section 4 analyzes in detail the relation between cash holdings and credit risk of U.S. insurance companies. Section 5 concludes. 5

6 2 Data and Variables 2.1 Data Sources Our data sample covers the period from July 2002 to June We collect quarterly data on CDS spreads and balance sheet information for twelve U.S. insurance companies and eight European insurers. The data is obtained from Bloomberg, Compustat and official regulatory filings. To measure credit risk, we focus on credit-default swaps (CDS) spreads rather than corporate bond yields. The CDS is an agreement in which the seller of the contract compensates the buyer if a credit event occurs. The buyer of the CDS makes a series of periodic payments to the seller and, in exchange, receives compensation if the underlying security defaults. Such periodic payments are called the spread of the contract. In the case of corporate debt, investors use default swaps to express their views about the creditworthiness of the firm, and to protect themselves in the event of default, debt restructuring, or a drop in the credit rating. Even though in a frictionless world CDS and bond spreads should be closely related to each other (Duffie and Singleton, 1999), in practice we observe significant differences that are due in part to the illiquidity of corporate bonds (Sarig and Warga, 1989; Chen et al., 2007), and different tax treatments of coupon payments (Elton et al., 2001). Furthermore, by focusing on CDS rather than bond spreads we avoid the problem of arbitrarily choosing the risk-free benchmark (Houweling and Vorst, 2005). Finally, CDS spreads react more quickly to new information compared to corporate bond yields (Hull et al., 2004; Blanco et al., 2005; Zhu, 2006), seem to anticipate changes in corporate bond ratings (Norden and Weber, 2004), and 6

7 incorporate information faster than bond yields in periods of market stress (Delatte et al., 2012). Our dataset is limited because of CDS data availability. For the U.S., there are 28 insurance companies in the S&P 500 among which only 12 have sufficient data on CDS spreads. For Europe, we select insurance companies that are traded in the most representative stock market indices of each country. Among these insurance companies, we keep only firms having enough CDS data. Table I presents the list of insurance companies that we use in the paper and summarizes relevant balance sheet information. The table presents time-series averages of assets (total assets, cash and investments) and liabilities (insurance liabilities, debt and equity) for insurance companies in the U.S. and Europe. On average, total assets of U.S. companies are smaller than their European counterparts. Similarly, average cash and investments 2 held by U.S. companies are smaller than for European companies. For liabilities, U.S. companies have lower insurance liabilities as well as debt than European insurers. However, there is more dispersion in size for U.S. than for European companies. Except for Scor, all other European insurers in our sample are of the same order of magnitude in terms of total asset size. We sample the CDS data quarterly because this is the frequency at which balance sheet information is available for companies, and in particular insurers. In contrast, some authors like Acharya et al. (2012) use monthly credit spreads combined with quarterly balance sheet data. 2 Investments include long positions in financial securities other than cash such as short-term debt, fixed-income, equity, loans, and others. 7

8 The disadvantage of this alternative method is that each balance sheet observation is kept constant for three months, introducing serial correlation in the regression residuals. In order to avoid this problem, we choose to perform our analysis using quarterly data. Our results, however, are robust to conducting the empirical analysis at the monthly frequency. 2.2 Variables CDS Spreads Our data set is composed of end of quarter CDS quotes obtained from Bloomberg for five year single name CDS, which are known to be the most liquid contracts. CDS quotes available for the sample period are richer for the U.S. than for Europe. Table II reports descriptive statistics on CDS spreads for U.S. and European insurance companies. In the table, CDS spreads are reported for each firm and also aggregated by region (U.S. and Europe). The mean spread for U.S. firms is bp, while the standard deviation is bp. The mean spread of bp and standard deviation of bp for European insurers are lower than the ones for U.S. insurance companies, respectively Investment Variables We include in our analysis different types of investments in financial instruments made by insurers, as a percentage of total assets: cash, short-term investments, fixed-income, equity and loans. These variables represent the asset allocation performed by insurers in order to maximize the profitability of the funds obtained by selling insurance policies and issuing debt. All variables are scaled by total assets. 8

9 In our analysis, cash represents the total amount of cash held by the company. We do not consider as cash short-term investments that could easily be liquidated and turned into cash. The intuitive prediction is that firms holding more cash should be safer, and hence have lower CDS spreads. However, recent research (Acharya et al., 2012) suggests that if cash levels are determined endogenously, a regression of CDS spreads on cash holdings might reveal a positive correlation between cash and credit risk. In other words, an increase in cash might be interpreted as a negative signal by market participants since the insurer could be increasing the cash for precautionary motives. In such case, we should expect a positive coefficient for cash when regressed with CDS spreads. Nevertheless, exogenous variations in cash should correlate negatively with credit risk. Furthermore, cash levels might also be constrained by country regulations and not impact CDS spreads. Figure 1 plots cash reserves to total assets and quarterly CDS spreads for U.S. insurers from July 2002 to June Figure 2 plots the same variables for European insurers. We observe that overall cash holdings of U.S. insurers display a strong positive correlation with CDS spreads, especially during the 2007 subprime crisis. The correlation between cash holdings and credit spreads seems weaker in Europe, and the time variation in cash reserves is less pronounced. Table III shows cash holdings as percentage of total assets for U.S. and European insurers. We observe that U.S. firms hold on average 0.77% of their total assets as cash, while European firms hold 3.63% of their total assets as cash holdings. Therefore, European insurers hold on average four times more cash than their U.S. counterparts. 9

10 Besides cash, we also include in the analysis investments in corporate debt that we divide by maturity. Short-term investments represent the total amount invested in deposits and investments with original maturities within one year, such as commercial paper. On the other hand, fixed-income represents the total amount invested in fixed-income securities with maturities over one year. Both items were hand-collected from official filings. Given the low risk of these securities, we should expect a negative effect on the credit risk of insurers. However, for the same reasons stated for cash holdings, investment on such securities could also be perceived by market participants as way to anticipate future losses. Insurers also take direct equity stakes in other companies. Given that these securities increase the risk of the portfolio, we hypothesize that this variable should correlate positively with the level of CDS spreads. Finally, we include loans that correspond to mortgage loans issued by insurers. As the period is usually associated with the bursting of a real estate bubble, we expect this variable to have a positive effect on CDS spreads, especially after the collapse of Lehman Brothers. Table III, Panel A displays summary statistics for these variables. For both U.S. and European insurers, investments in fixed-income securities represent the largest share with respect to total investments, although the figure is larger in the U.S. (49%) than in Europe (38%). Shortterm investments are more predominant in the U.S. than in Europe, whereas the opposite is true for equity investments. There is, however, some dispersion around the mean as shown by the standard deviation of these variables. 10

11 2.2.3 Firm-Specific Variables We also collect from Bloomberg and 10-Q forms company specific variables that allow us to compute quantities that are known to impact CDS premia, such as leverage, equity volatility and distance-to-default. First, we want to analyze to what extent leverage is an important determinant of CDS premia for insurance companies since it is known to explain credit risk premia for companies in general. We construct two measures of leverage, one representing just the long-term debt as a percentage of total assets, and one representing the proportion of insurance liabilities of each insurer to total assets. The reason for doing this relies on the fact that insurance firms have a large exposure to the assets that they insure (casualty, life and property) and we want to understand which form of leverage matters most for CDS premia. This is something specific to the insurance industry that, to the best of our knowledge, has not been analyzed in previous literature. The data for debt was hand collected from regulatory filings whereas insurance liabilities were obtained from Bloomberg. Second, we also want to analyze the impact of equity volatility on the CDS spread of insurers. We use a 90 day historical volatility that we obtain from Bloomberg. This quantity is calculated as the annualized standard deviation of the stock returns for the 90 most recent trading days. The measure uses closing prices for its computation. Our choice aims to be consistent with the fact that we use quarterly data in our regressions. Finally, we also include in our analysis Vassalou and Xing (2004) distance-to-default measure. This variable represents the distance, measured in standard deviations, from default in a 11

12 Merton (1974) setup. A higher distance-to-default translates in a lower probability of default. We compute our measure of distance-to-default (DD) using the method of Bharath and Shumway (2008). The naïve distance-to-default measure of Bharath and Shumway (2008) is defined as: where represents the value of the market equity calculated as the product of the stock price at the end of each quarter and the number of shares outstanding; is the face value of debt; is the return of equity of firm in the previous period; ; ; and is the forecasting horizon of 1 year. We collect the inputs to the distance-to-default model of Bharath and Shumway (2008) from different sources. The volatility of stock returns is obtained from Bloomberg and estimated as the annualized standard deviation of the relative price change for the 30 most recent trading days, expressed as a percentage. The market value of equity for each insurer is calculated as the product of the stock price at the end of the month and the number of shares outstanding using data from Bloomberg. As in Bharath and Shumway (2008), the face value of debt is estimated to be the short-term debt plus one-half of long-term debt that we obtain from COMPUSTAT for U.S. insurers and from regulatory filings for European insurers. Table III, Panel B presents summary statistics of these four variables. On the one hand, we can observe that corporate debt is relatively low as a percentage of total assets both for U.S. (5%) and European (7%) insurers. On the other, insurance liabilities represent a large share of the 12

13 balance sheet and are quite similar in the U.S. (71%) and Europe (69%). Volatility is also quite similar on average in the U.S. (37%) and Europe (37%), although there is more cross-sectional variation in the U.S. Finally, in terms of distance-to-default, U.S. insurers appear safer than European insurers Macro Variables The literature has also determined the importance of common macro factors in determining the level of CDS spreads. In our empirical analysis, we will alternatively use time-effects to capture any common trend in the series that is not captured by our macro factors. Our macroeconomic control variables were obtained from Bloomberg for the U.S. and Europe, and include the risk-free rate, the yield slope, the implied stock market volatility and stock market skew, and the swap spread. As pointed out by Collin-Dufresne et al. (2001), an increase in the risk-free rate should produce a decrease in CDS spreads. Following Longstaff et al. (2005) and Raunig and Scheicher (2009), we use the five-year swap rate in USD and EUR as a proxy for the risk-free rate. Collin-Dufresne et al. (2001) also show that an increase in the slope of the yield curve should decrease CDS spreads. We compute the slope of the yield curve in the U.S. and Europe as the difference between the ten and one-year USD and EUR swap rate, respectively. We also know from previous literature that an increase in stock market volatility should affect positively CDS spreads to compensate investors for more expected losses from default. We use the VIX implied volatility index (Coudert and Gex, 2008; Raunig and Scheicher, 2009) to proxy 13

14 for this variable in the U.S and use the V2X index to proxy for the volatility of the Euro STOXX 50 index. We also include a measure of tail risk since the period studied in our paper covers the recent financial crisis. For this we consider the implied stock market skew. Similarly to stock market volatility, an increase in the skew should proxy for a higher probability of a market crash, hence increasing CDS spreads. We use the SKEW index provided by the CBOE (SKEW) to proxy for the implied skew. There is no such index for Europe, so we use the same index for both U.S. and European insurers. Finally, we follow Longstaff et al. (2005) and include the swap spread between swap rates and government bond yields to proxy for flight-to-liquidity. We use the Bloomberg 2-year swap spread in USD and EUR. Table III, Panel C presents summary statistics for these variables. The values reported in the table are just time-series averages since the variables are common to all insurers in their respective region. We find that all macro variables are similar for both the U.S. and Europe, except for the implied volatility that is higher in Europe. 3 Credit Spreads Determinants of Insurance Companies 3.1 Methodology In this section we analyze which variables are significant determinants of CDS spreads for all insurance companies in our sample: investment portfolio variables, firm-specific variables or macroeconomic factors. 14

15 As mentioned before, existing empirical studies have found that macro variables such as the level and the slope of the term-structure of interest rates, or the implied market volatility are important determinants of CDS spread changes. Furthermore, these studies have also confirmed that firm-specific characteristics such as leverage and idiosyncratic volatility also matter for the level and changes of CDS spreads. To test which variables impact CDS spreads of insurance companies, we run first a basic panel regression model allowing for fixed-effects and firm-specific variables (equity volatility, debt, insurance liabilities and distance-to-default). We also include macro-controls that have been shown to matter for credit risk (risk-free rate, slope of the yield curve, stock market volatility, stock market skew and swap spread), or alternatively time-effects, to remove unwanted systematic trends. We also add a set of investment portfolio variables such as cash, short-term and fixed-income investments, equity and loans, to test whether the portfolio risk of these investments has an effect on credit risk. The empirical specification of the analysis is as follows: (1) where represents the CDS quote for entity i at the end of period t; is is a set of investment portfolio variables; is a set of firm-specific variables; and is a vector of macroeconomic or time-effects variables. Since the previous regression might reflect equilibrium between credit risk and firmcharacteristics, we also test for Granger causality between credit spreads and firm-specific variables. Hence, we run the same panel regression allowing for fixed-effects, lagged firmspecific variables, and lagged macro-controls. In some specifications we replace the macro 15

16 controls with time fixed-effects to remove unwanted systematic trends. The empirical specification of the analysis is as follows: (2) where represents the CDS quote for entity i at the end of period t; is is a set of lagged investment portfolio variables; is a set of firm-specific variables with lags; and is a vector of lagged macroeconomic or time-effects variables. 3.2 Results for the Full Sample We run the panel-data model in equation (1) for the full sample at the quarterly frequency. Table IV presents the regression results. All p-values are calculated using robust standard errors. Column (1) presents the results for a regression of CDS spreads on investment variables allowing only for fixed-effects. Column (2) adds macro variables, while column (3) uses timeeffects instead of macro controls. We repeat the same process but this time including other firm-specific variables in columns (4), (5) and (6). We do not find that portfolio investment variables have a significant effect in explaining CDS spreads. Even though cash, fixed-income and equity come significant under some specifications, they lose their explanatory power when macro variables and time-effects are included. On the contrary, we find that firm-specific variables are significant determinants of credit spreads. The coefficients on volatility and insurance liabilities are positive as expected. Distance-to-default and debt are significant under most specifications, except when we include time-effects. As expected, debt has a positive coefficient while distance-to-default has a negative coefficient. Finally, we find that macroeconomic factors such as the risk-free rate, the 16

17 yield slope and the stock market skew are significant determinants of credit spreads. The swap spread loses explanatory power when we include firm-specific variables in the regression. We also analyze if lagged values of these variables can predict the variation of CDS spreads. As explained before, we run the panel-data model in equation (2) for the full sample at the quarterly frequency. Table V presents the results. All p-values are calculated using robust standard errors. Column (1) presents the results for a regression of CDS spreads on investment variables allowing only for fixed-effects. Column (2) adds macro variables, while column (3) contains time-effects. We repeat the same process but this time including firm-specific variables in columns (4), (5) and (6). Among the portfolio investment variables, we find that these variables are not significant predictors of CDS spreads. Even though lagged values of short-term investments, fixed-income and equity come significant under some specifications, they lose their explanatory power when firm-specific variables, macro variables and time-effects are included. In terms of firm-specific variables, we find that debt and volatility are significant in predicting CDS spreads. They have both positive coefficients as expected. Lagged values of distance-to-default and insurance liabilities are significant under most specifications, except when we include time-effects. Finally, we find that past values of macroeconomic factors such as the swap rate, the yield slope and stock market skew are also significant predictors of CDS spreads for the full sample. Overall, our results confirm the findings of Hang et al. (2009) in that equity volatility risk predicts a large part of the variation in CDS spreads for non-financial firms. This result is also consistent with Campbell and Taksler (2003) that find that equity volatility is a significant 17

18 determinant of corporate bond yields. Hence, our results challenge the view that variables that are known to explain credit spreads of non-financial firms usually lose their explanatory power when applied to financial firms (see e.g. Boss and Scheicher, 2002; Raunig and Scheicher, 2009; Grammatikos and Vermeulen, 2012). Equity volatility seems to be an important predictor of credit spreads for insurance companies as it is for non-financial firms. We also find that the level of debt is a crucial predictor of credit spreads. However, our results show that it is important to distinguish between the level of debt and the level of insurance liabilities, which does not have the same strong forecasting power. Contemporaneously, we find that all firm-specific variables other than investment variables including cash are significant in explaining CDS spread variation. Since in our regressions we have pooled together both U.S. and European insurers, in the next section we analyze the differences in credit spreads determinants when we separate the insurance companies by region. 3.3 Results by Region We run the panel-data model in equation (1) for the U.S. and Europe separately. Table VI presents the regression results for U.S. firms, and Table VII reports the results for European firms. All p-values are calculated using robust standard errors. Column (1) in both tables presents the results for a regression of CDS spreads on investment variables allowing only for fixed-effects. Column (2) adds macro variables, while column (3) contains time-effects. We repeat the same process but this time including firm-specific variables in columns (4), (5) and (6). 18

19 Among the portfolio investment variables for U.S. firms, we find that cash, short-term investments and fixed-income have positive signs in all specifications, which means that they are positively correlated with CDS spreads. Among these variables, cash come significant in all specifications, fixed-income is significant in four out of six specifications, and short-term investments are significant in only two specifications. The results for cash are counter intuitive since more cash should be associated with a lower probability of default and hence lower credit risk. However, recent work by Acharya et al. (2012) indicates that standard OLS regressions used in empirical studies of credit spreads should predict a positive correlation between cash holdings and credit risk. Furthermore, as predicted by Acharya et al. (2012), we find that the economic significance of the coefficient is stronger when no credit-risk controls are included in the regression since cash holdings proxy for credit risk. However, this significance decreases by more than half when credit-risk controls are included. Our results suggest that one of the most important determinants of CDS spreads for U.S. insurers are their cash reserves. Short-term investments, on the other hand, do not play such a prominent role even though they could be seen as close substitutes. In terms of firm-specific variables, we find that the amounts of debt and equity volatility are significant determinants of the credit risk of U.S. insurers. They have both positive coefficients as expected. The distance-to-default coefficient is negative in all specifications although is only significant when we include time-effects. On the other hand, credit risk of U.S. insurers seems to be insensitive to the level of their insurance liabilities. Finally, we do not find macroeconomic 19

20 factors to be significant determinants of CDS Spreads for U.S. insurers when we control for firmspecific factors. Table VII presents the results for credit spreads regressions of European insurance firms. First, we can observe that contrary to U.S. firms, investment portfolio variables do not seem to consistently explain CDS spreads across all specifications. The statistical significance of cash holdings, short-term investments and equity disappears when we allow for time-effects. For firm-specific variables we find that equity volatility and distance-to-default have a significant effect for European insurers, even after controlling for time-effects. As expected, volatility has a significant positive coefficient, while distance-to-default has a negative correlation with CDS spreads. Contrary to what was observed for U.S. firms, two macroeconomic factors seem to explain CDS spreads under all specifications: the risk-free rate and the swap spread. Interest rates are high when the economy is booming and low in recessions, suggesting a negative correlation with credit spreads. Conversely, the swap spread widens in periods of market stress because of flight-to-liquidity, suggesting that credit spreads should also increase in such periods. As expected, the risk-free rate has a negative coefficient while the swap spread has a positive correlation with credit risk. Hence, credit spreads of European insurers are more sensitive to the business cycle than spreads of U.S. insurers. 3.4 Comparison between CDS Determinants in the U.S. and Europe Our results suggest that there is a difference between U.S. and European insurers in terms of determinants of credit spreads. CDS spreads in the U.S. seem to be driven more by individual characteristics such as cash, debt and equity volatility, rather than observable macroeconomic 20

21 factors. However, CDS spreads in Europe seem to be explained better by equity volatility, distance-to-default, and macroeconomic factors such as the risk-free rate and the swap-spread. Investments in financial assets do not seem to matter for the credit spread of European insurance companies. The results show that equity volatility is the only firm-specific variable that comes significant in all specifications for both U.S. and European insurers. The effect is also economically significant for both U.S. and European insurers. This finding confirms the results of Zhang et al. (2009) in that equity volatility is the most important determinant of credit spreads. We can observe that the effect of equity volatility on credit spreads relates to individual characteristics not related to market wide volatility as the coefficients on the VIX and V2X are not significant, and that the effect holds after controlling for leverage. We also find interesting that cash holdings are of such importance in explaining the credit spread variation of U.S. insurers. Interestingly, very little is known about the determinants of cash holdings for U.S. insurers. A notable exemption is Colquitt et al. (1999), who investigate the differences in cash holdings across property-liability insurers. They conclude that larger and high quality insurers hold less cash, and that insurers with a higher variance of cash flows tend to hold more cash. We believe that the differences of the impact of cash reserves on the credit risk for U.S. compared to European insurers might be due to existing regulation during our sample period. In the U.S., insurance companies are required to maintain their adjusted capital above a minimum required level called risk-based capital (RBC), which depends on several risk factors, 21

22 as established by the National Association of Insurance Commissioners (NAIC). The risk factors for the NAIC s RBC formulas focus on three major areas: asset risk, underwriting risk, and other risks. The weight of each factor in the RBC formula differs depending on the type of insurance, but asset risk remains an important determinant of minimum capital requirements. In E.U. countries, insurance entities are required to maintain minimum solvency margins according to the existing Solvency I legislation. The minimum capital is calculated as a fixed percentage of premiums, claims, reserves and/or net amounts at risk. The required minimum solvency margin for general insurers depends on premiums written for the year, or the threeyear average of claims incurred. Life insurance companies are required to maintain a minimum solvency margin generally of 4 percent of insurance reserves, plus 0.3 percent of the amount at risk under insurance policies. The same minimum capital requirements are applicable for insurance entities operating in Switzerland. 3 As a consequence, in our sample the regulatory capital for U.S. insurers is more sensitive to the risk of assets than for European insurers, which may explain why cash reserves of U.S. entities are more informative about their credit condition than the cash holdings of European entities, and vary more in response to market events. Empirically, we observe that U.S. insurers hold on average four times less cash to total assets than European insurance companies, and that cash reserves for U.S. insurers vary more over time than the ones of their European counterparts. 3 The new Solvency II legislation was scheduled to replace Solvency I on January 2013, but was recently postponed until January The new Swiss Solvency Test became fully mandatory in January

23 4 Cash Holdings and CDS Spreads for U.S. Insurers 4.1 Methodology Given the importance of solvency for insurance companies, we explore in more detail the relationship between cash holdings and credit risk. Acharya et al. (2012) show that cash holdings affect credit risk through two channels: a direct one given the endogenous nature of cash holdings, and an indirect channel through exogenous variations in cash levels. On the one hand, they find that endogenous variations in cash result in a robust positive correlation of corporate cash holdings with credit spreads, and with the long-term probability of default. On the other hand, exogenous variations in cash levels unrelated to credit risk factors should be negatively correlated with credit spreads since in that case the firm becomes effectively safer. In this section we focus on U.S. insurers given that cash holdings do not impact significantly the credit risk of European insurance companies. We do not proceed to directly identify exogenous variations in cash levels, but rather we study how credit spreads react for firms that hold a larger proportion of cash after they experience an improvement in their solvency. To achieve this, we use a differences-in-differences approach in which we perform a cross-sectional comparison of cash holdings of firms which have unexpectedly improved their financial position compared to those who have not. We test three related empirical specifications in which we study the interaction of unexpected improvements in solvency with contemporaneous cash levels. We use unexpected increases in cash reserves, as well as unexpected reductions of insurance liabilities, as proxies for improvements of insurers solvency. Our first empirical specification is as follows: 23

24 (3) where represents the CDS quote for entity i at the end of period t; is the cash holdings to total assets ratio; is an indicator function equal to one if and zero otherwise; is a set of firm-specific variables that includes all other investment variables; and is a vector of macroeconomic or time fixed-effects variables. We use the interaction of cash holdings with changes in cash with respect to the previous period as our identification strategy. Since cash holdings changes are difficult to predict, the interaction term captures the relative relevance of cash holdings after a positive solvency shock. We also use a reduction in insurance liabilities as a proxy for an unexpected improvement in the liabilities of the firm: (4) as well as their combined effect: (5) where is an indicator function equal to one if and zero otherwise; and is an indicator function equal to one if both and. In the last specification the combined effect of an improvement in cash reserves and a reduction of insurance liabilities should yield the strongest results. 24

25 4.2 Results Table VIII reports the results of the estimation. We first estimate all three specifications allowing for firm fixed-effects and macro controls or time-effects in columns (1) to (6). We then re-estimate all three specifications including the other investment portfolio variables and individual characteristics in columns (7) to (12). For the specifications described in equations (3) and (4), we find a negative coefficient for the interaction term, although the coefficient fails to be statistically significant except in one case. For our final specification in equation (5), we find the interaction coefficient to be negative and statistically significant. Therefore, the results show that firms with larger amounts of cash that experience an unexpected improvement in their solvency become safer. The results are also economically significant since they suggest that firms who hold one standard deviation more of cash see their credit spreads reduced between 30 to 42 bp after a positive solvency shock. 5 Concluding Remarks Our analysis on the determinants of credit risk for insurance companies in the U.S. and Europe reveals three main results: i) consistent with recent findings for non-financial firms in the U.S., we find that equity volatility is a major determinant and predictor of CDS spreads for both U.S. and European insurers, even after controlling for the composition of their investment portfolios and other firm-specific characteristics such as leverage and macro controls; ii) when analyzing if other determinants differ for U.S. and European insurers, we find macroeconomic factors to affect the credit risk of European but not U.S. insurers, whereas cash holdings seem to be 25

26 relevant for the credit spreads of U.S. insurance companies; and iii) we find that in equilibrium, cash holdings of U.S. insurers and credit spreads are positively correlated, even though the availability of cash reduces the credit risk of firms experiencing positive solvency shocks. We believe that our results are relevant for practitioners, investment professionals and macroprudential regulators worldwide. We show that minimum capital requirements can have a substantial effect on the cash reserves of insurance entities. This effect is captured by the credit spread of insurers when cash reserves are informative about the credit situation of the insurer. When cash holdings are above this informative level, financial markets tend to focus more on the macro-economic environment when assessing the creditworthiness of insurance companies. In conclusion, we find that equity and credit markets are quite efficient in incorporating quickly information about the financial solvency of large insurance companies. 26

27 References Acharya, V., Davydenko, S.A. and I.A. Strebulaev, 2012, Cash Holdings and Credit Risk, Review of Financial Studies, 25, Ambrose, J.M. and A.M. Carroll, 1994, Using Best's Ratings in Life Insurer Insolvency Prediction, Journal of Risk and Insurance, 61, Annaert, J., De Ceuster, M., Van Roy, P. and C. Vespro, 2013, What Determines Euro Area Bank CDS Spreads?, Journal of International Money and Finance, 32, Bharath, S.T. and T. Shumway, 2008, Forecasting Default with the Merton Distance to Default Model, Review of Financial Studies, 21, Blanco, R., Brennan, S. and I.W. Marsh, 2005, An Empirical Analysis of the Dynamic Relation between Investment-Grade Bonds and Credit Default Swaps, Journal of Finance, 60, Boss, M. and M. Scheicher, 2002, The Determinants of Credit Spread Changes in the Euro Area, BIS Papers, 12, Campbell, J.Y. and G.B. Taksler, 2003, Equity Volatility and Corporate Bond Yields, Journal of Finance, 58, Carson, J.M. and R.E. Hoyt, 1995, Life Insurer Financial Distress: Classification Models and Empirical Evidence, Journal of Risk and Insurance, 62,

28 Chen, L., Lesmond, D.A. and J. Wei, 2007, Corporate Yield Spreads and Bond Liquidity, Journal of Finance, 62, Collin-Dufresne, P., Goldstein, R.S. and J.S. Martin, 2001, The Determinants of Credit Spread Changes, Journal of Finance, 56, Colquitt, L.L., Sommer, D.W. and N.H. Godwin, 1999, Determinants of Cash Holdings by Property-Liability Insurers, Journal of Risk and Insurance, 66, Coudert, V. and M. Gex, 2008, Does Risk Aversion Drive Financial Crises? Testing the Predictive Power of Empirical Indicators, Journal of Empirical Finance, 15, Delatte, A.-L., Gex, M. and A. López-Villavicencio, 2012, Has the CDS Market Influenced the Borrowing Cost of European Countries During the Sovereign Crisis?, Journal of International Money and Finance, 31, Duffie, D., and K.J. Singleton, 1999, Modeling Term Structures of Defaultable Bonds, Review of Financial Studies, 12, Elton, E.J., Gruber, M.J., Agrawal, D. and C. Mann, 2001, Explaining the Rate Spread on Corporate Bonds, Journal of Finance, 56, Grammatikos, T. and R. Vermeulen, 2012, Transmission of the Financial and Sovereign Debt Crises to the EMU: Stock Prices, CDS Spreads and Exchange Rates, Journal of International Money and Finance, 31,

29 Houweling, P. and T. Vorst, 2005, Pricing Default Swaps: Empirical Evidence, Journal of International Money and Finance, 24, Hull, J., Predescu, M and A. White, 2004, The Relationship Between Credit Default Swap Spreads, Bond Yields, and Credit Rating Announcements, Journal of Banking and Finance, 28, Lee, S.H. and J.L. Urrutia, 1996, Analysis and Prediction of Insolvency in the Property-Liability Insurance Industry: A Comparison of Logit and Hazard Models, Journal of Risk and Insurance, 63, Longstaff, F.A., Mithal, S. and E. Neis, 2005, Corporate Yield Spreads: Default Risk or Liquidity? New Evidence from the Credit Default Swap Market, Journal of Finance, 60, Merton, R.C., 1974, On the Pricing of Corporate Debt: The Risk Structure of Interest Rates, Journal of Finance, 29, Norden, L and M. Weber, 2004, Informational Efficiency of Credit Default Swap and Stock Markets: The Impact of Credit Rating Announcements, Journal of Banking and Finance, 28, Raunig, B and M. Scheicher, 2009, Are Banks Different? Evidence from the CDS Market, Working Paper 152, Oesterreichische Nationalbank. Sarig, O. and A. Warga, 1989, Some Empirical Estimates of the Risk Structure of Interest Rates, Journal of Finance, 44,

30 Shaked, I., 1985, Measuring Prospective Probabilities of Insolvency: An Application to the Life Insurance Industry, Journal of Risk and Insurance, 52, Trieschmann, J.S. and G.E. Pinches, 1973, A Multivariate Model for Predicting Financially Distressed PL Insurers, Journal of Risk and Insurance, 40, Vassalou, M and Y. Xing, 2004, Default Risk in Equity Returns, Journal of Finance, 59, Zhang, B.Y., Zhou, H. and H. Zhu, 2009, Explaining Credit Default Swap Spreads with the Equity Volatility and Jump Risks of Individual Firms, Review of Financial Studies, 22, Zhu, H., 2006, An Empirical Comparison of Credit Spreads Between the Bond Market and the Credit Default Swap Market, Journal of Financial Services Research, 29,

31 Sep-02 Feb-03 Jul-03 Dec-03 May-04 Oct-04 Mar-05 Aug-05 Jan-06 Jun-06 Nov-06 Apr-07 Sep-07 Feb-08 Jul-08 Dec-08 May-09 Oct-09 Mar-10 Aug-10 Jan-11 Jun-11 Nov-11 Apr-12 CDS Spreads (%) Cash to Total Assets (%) Sep-02 Feb-03 Jul-03 Dec-03 May-04 Oct-04 Mar-05 Aug-05 Jan-06 Jun-06 Nov-06 Apr-07 Sep-07 Feb-08 Jul-08 Dec-08 May-09 Oct-09 Mar-10 Aug-10 Jan-11 Jun-11 Nov-11 Apr-12 CDS Spreads (%) Cash to Total Assets (%) Figure 1: CDS Spreads and Cash Holdings for U.S. Insurance Companies. The figure plots the cross-sectional average CDS spread and cash to total assets for all U.S. insurance companies. The sample covers the period July 2002 to June CDS Spreads Cash Figure 2: CDS Spreads and Cash Holdings for European Insurance Companies. The figure plots the cross-sectional average CDS spread and cash to total assets for all European insurance companies. The sample covers the period July 2002 to June CDS Spreads Cash 31

32 Table I: Balance Sheet Composition by Company and Region. The table reports balance sheet information on selected insurance companies in U.S. and Europe. The sample covers the period July 2002 to June All figures are time-series averages denominated in millions of USD. Region Company Type Total Assets Cash Investments Insurance Liabilities U.S. Ace Ltd. MULTI-LINE 68, ,657 42,316 2,915 15,194 Allstate Corp. MULTI-LINE 140, , ,755 5,421 19,295 American International Group Inc. MULTI-LINE 798,324 2, , ,831 68,320 81,124 Chubb Corp. PROPERTY/CASUALTY 47, ,175 27,725 3,261 12,737 Hartford Financial Services Group Inc. MULTI-LINE 283,827 1, , ,415 5,201 16,104 Lincoln National Corp. LIFE/HEALTH 156,589 2,532 60, ,698 4,053 9,606 Loews Corp. MULTI-LINE 74, ,308 41,307 7,188 16,782 Metlife Inc. MULTI-LINE 514,353 9, , ,385 14,465 33,132 Prudential Financial Inc. LIFE/HEALTH 453,417 10, , ,235 21,134 24,796 Travelers Companies Inc. PROPERTY/CASUALTY 99, ,898 64,788 5,648 21,766 Torchmark Corp. LIFE/HEALTH 14, ,505 9, ,323 Unum Group LIFE/HEALTH 52, ,249 39,996 2,801 7,536 Average 225,339 2, , ,646 11,771 21,783 Europe Axa S.A. MULTI-LINE 831,656 29, , ,941 13,408 78,704 Allianz SE MULTI-LINE 1,134,270 26, , , ,856 51,501 Swiss Re REINSURANCE 250,668 14, , ,219 17,597 26,326 Zurich Insurance Group MULTI-LINE 353,237 13, , ,061 10,809 26,351 Swiss Life Holding LIFE/HEALTH 172,771 8, , ,868 4,839 7,574 Legal General Group LIFE/HEALTH 231,886 1, , ,049 12,708 7,324 Muenchener REINSURANCE 289,438 3, , ,242 43,058 28,647 Scor REINSURANCE 28,944 2,073 17,232 22,664 3,497 3,762 Average 411,609 12, , ,120 30,347 28,774 Debt Equity 32