Market Discipline and Guaranty Funds in Life Insurance

Market Discipline and Guaranty Funds in Life Insurance by Martin Grace, * Shinichi Kamiya, ** Robert Klein, *** and George Zanjani **** Abstract This paper studies the effects of company risk and guaranty funds on life insurance in force using company-by-state level data during the 1985-2010 period. Consistent with market discipline, it finds a negative relation between company risk (measured by changes in financial ratings) and changes in life insurance in force and annuity considerations. Effects are especially large for annuity considerations. The paper finds some evidence of a decline in market discipline following the creation of governmentbacked guaranty funds in 15 states during the sample period, with the most significant effects being observed at downgraded firms with low financial ratings. JEL classification: G22; G28; E53 Keywords: Guaranty Funds; Deposit Insurance; Market Discipline; Regulatory Discipline * Contact info: mgrace@gsu.edu. Department of Risk Management and Insurance, Georgia State University ** Contact info: skamiya@ntu.edu.sg. Division of Banking and Finance, Nanyang Business School *** Contact info: rwklein@gsu.edu. Department of Risk Management and Insurance, Georgia State University **** Corresponding author. Contact info: gzanjani@gsu.edu. Department of Risk Management and Insurance, Georgia State University. An earlier and incomplete version of this paper circulated under the title Market Discipline and Government Guarantees in U.S. Life Insurance. 1

I. Introduction The extent to which lenders discipline financial institutions and how such discipline is affected by the presence of government guarantees are issues of vital importance for regulators. Government deposit insurance and guaranty funds 1 protect consumers in the event of company default and may help to prevent destabilizing runs on financial institutions. However, they may also mitigate incentives for consumers to monitor their financial institutions, leading to excessive risk-taking and other well-known moral hazard problems. If consumers care about company risk, market discipline can be viewed as an important restraint on risk-taking by firms, even in the presence of public guarantees. It is difficult, however, to empirically isolate the effect of public guarantees. Existing studies take a variety of approaches. Some compare the risk sensitivity of insured debtholders with uninsured or partially insured debtholders. Others exploit variation in deposit insurance schemes across countries. Still others examine changes within the same country over time. These approaches have obvious limitations with respect to isolating the effect of the public guarantee. To illustrate, in addition to having different levels of insurance coverage, insured debtholders typically differ from uninsured debtholders in other ways; international comparisons suffer from the difficulties in controlling for social, cultural, or regulatory differences that could also be contributing to any observed changes; within country comparisons suffer from the difficulty of distinguishing effects related to the scheme change from effects owing to other contemporaneous changes. Life insurance in the U.S. offers a unique opportunity to estimate the effect of public guarantees on market discipline. The government institutions that indemnify policyholders in the event of an insurance company default, known as guaranty funds, are organized on a state basis in the U.S., and a number of states were late to adopt consumer protections. Specifically, 15 states did not have life insurance 1 Guaranty funds are set up at the state level in the U.S. for the purpose of paying the claims of policyholders of insolvent insurance companies. 2

guaranty funds in place at the beginning of 1985, and it was not until the end of 1992 that all states had enacted guaranty funds. The timing is important because electronically compiled statutory data filings are available starting in 1985, so it is possible to observe financial data for companies at the state level starting in 1985. We can thus observe the state-level performances of a company selling similar products in multiple states during the 1985-1992 period, where some of the states had guaranty fund protections in place and others did not. This allows a side-by-side comparison of the effects of company default risk changes in environments with guarantees and without guarantees, in a context where the affected policyholders in each environment are plausibly similar. Specifically, we study how life insurance and annuity growth measures responded differently in protected environments versus unprotected environments when an insurer experienced a change in risk as measured by financial rating. Downgraded firms are known to experience declines in premium growth, while upgraded firms can experience increases. Guaranty funds can be expected to mitigate both of these effects: Consumers protected by guaranty funds can be expected to be less risk sensitive than their unprotected counterparts. Briefly, the paper confirms earlier research by finding a statistically significant negative association between increases in company risk (downgrades) and growth rates of life insurance in force, life insurance premiums, and annuity considerations; 2 decreases in company risk (upgrades) are associated with increases in growth rates. The effects are especially large for annuity considerations. We find evidence that guaranty funds mitigated risk sensitivity during the 1985-1992 period. For downgraded firms, we find effects only for firms with very low ratings (in the C-range), with the impact being a 5 to 8 percent increase in life insurance growth in protected environments relative to unprotected environments. For upgraded firms, effects are more widespread across rating classes, with an overall impact being a 2 to 5 percent decrease in life insurance growth in protected environments relative to unprotected 2 Life insurance in force is the sum of life insurance policy face values, while annuity considerations are the sum of annuity premium payments. 3

environments. In summary, there is some evidence of reduced sensitivity to risk in protected environments, although the magnitude is modest. Related Literature The majority of the empirical literature on market discipline has focused on banks and thrifts, but some studies have been done for insurance markets. Flannery (1998) offers a well-known review of market discipline in banking, while Eling (2012) offers a recent survey of market discipline both in banking and insurance. In terms of the measurement of market discipline, the banking literature analyzes both investor-driven market discipline by analyzing capital market assessment of bank risk-taking (see, e.g., King, 2008; Chen and Hasan, 2011) and depositor discipline (e.g., Goldberg and Hudgins, 1996; Park and Peristiani, 1998; Martinez Peria and Schmukler, 2001; Goldberg and Hudgins, 2002). In the insurance industry, many companies are not publicly traded and use little subordinated debt at the level of the company. Consequently, the insurance literature investigates customer-driven market discipline by analyzing pricing, premium growth, and lapse (see, e.g., Sommer, 1996; Epermanis and Harrington, 2006; Eling and Schmit, 2012). Epermanis and Harrington (2006) analyze the relationship between insurance premium growth and changes in A.M. Best ratings for the US property/casualty insurers during 1992-99, and find premium declines following rating downgrades. They find that premium declines are especially severe for commercial insurance, which could be explained by greater risk sensitivity of commercial insurance demand due less complete guaranty fund protection. Baranoff and Sager (2007) analyze the effect of rating downgrades on the demand of life insurance by counting the number of in-force life insurance policies (the year-end and year-beginning policy counts during 1994-2003 are transformed by taking logarithms and then differencing the logs.). They find that a rating downgrade Granger-causes a decline of about 6% in the number of in-force life policies. Using the methodology proposed by Epermanis and Harrington (2006), Eling and Schmit (2012) study the German market and obtain similar findings. They use both premium growth rates and termination rates (lapses 4

with surrender value as a percentage of the total business in force for life insurers) for life, property/casualty and health during 1996-2005, and test for the existence of market discipline. They find premium declines as well as increased lapse rates following rating downgrades. In the US financial services sector, deposit insurance schemes in banking and state government guarantee funds in insurance protect small depositors and individual policyholders and may potentially prevent destabilizing runs on financial institutions (Diamond and Dybvig, 1983). On the other hand, those safetynet programs may affect the extent of market discipline. For example, Merton (1977) shows that deposit insurance increases the bank's risk as it searches for higher profits because the insurance limits the bank's downside risk and therefore encourages risk taking and discourages precaution. If deposit insurance lowers depositor discipline, incentives for the institution to transfer wealth to shareholders by, for example, increasing asset risk and leverage will become more acute. Demirguc-Kunt and Detragiache (2002) find that, on average, the moral hazard effect seems to be the more dominant influence with respect to overall systemic financial stability (see also Anginer, Demirguc-Kunt, and Zhu (forthcoming)). A number of studies have addressed the important question of how deposit insurance affect market discipline. A variety of approaches have been taken, and the findings are mixed. Some studies offer evidence on the impact of deposit insurance through comparisons of uninsured and insured sources of funding. Billet, Garfinkel and O Neal (1998) show that downgraded banks shift toward insured funding, and that those more reliant on insured funding are penalized less heavily by investors after a downgrade. Using international data, Gropp and Vesala (2004) document that the enactment of explicit deposit insurance is associated with a shift toward insured funding. Park and Peristiani (1998) find evidence of market discipline by both insured and uninsured depositors, although the effects are stronger with the latter. Martinez Peria and Schmuckler (2001) find that the behavior of insured and uninsured depositors is largely the same. 5

Other studies rely either on cross-national comparisons or changes in the deposit insurance scheme within a country over time, or both. Demirguc-Kunt and Huizinga (2004) find that explicit deposit insurance lowers market discipline on bank risk taking; specifically, deposit interest rates and growth rates appear to be less sensitive to bank risk factors---in the sense of penalizing risk less heavily---in the presence of deposit insurance. Gropp and Vesala (2004) and Nier and Baumann (2006) find the contrary result that bank riskiness decreases following the enactment of an explicit deposit insurance scheme; both studies interpret the finding as evidence that explicit schemes are actually commitments to limit the scope of intervention, so that enactments may actually reflect a drop in the level of public support. Other studies advance the analysis in various directions, such as the combined effect of deposit insurance coverage and ownership structure (Forssbaeck, 2011) and the existence of market discipline in other countries (Imai, 2006; Hadad et al., 2011). These approaches are useful but have obvious identification problems. For example, it is hard to know how much to attribute a difference between uninsured depositors and insured depositors to the difference in deposit insurance versus other differences in characteristics; similarly, it is hard to disentangle the impact of differing deposit insurance levels across countries from unobserved cultural, social, or legal differences. Natural experiments are difficult to find, though illuminating where they exist. Gropp, Gruendl, and Guettler (2013) studied the responses associated with the recent removal of government guarantees from savings banks in Germany, using commercial banks (who were always unprotected) as a control group. They found that the removal of the (full) guarantee led to a decline in risk taking, an increase in the cost of funding, and a shift toward insured sources of funding. Karas, Pyle, and Schoors (2013) study the enactment of deposit insurance for household deposits in Russia in 2004, using unprotected firms as a control group. Looking at both interest rates and deposit flows, they find a reduction in market discipline. U.S. insurance offers an exciting opportunity to identify how much public protection matters. States vary in the generosity of their insurance guaranty funds, and the timing of guaranty fund adoption was 6

staggered across states. Many companies serving the United States during this period operated in multiple states, serving similar products to protected and unprotected populations. Thus, studying the impact of depositor protection in the context of U.S. insurance offers a promising environment to control for various differences between insured and uninsured groups. Previous studies on guaranty funds in insurance sectors have generally found evidence consistent with market discipline. Some studies have associated guaranty funds with various types of risky insurance company behavior (Lee, Mayers, and Smith, 1997; Brewer, Mondschean, and Strahan, 1997; Downs and Sommer, 1999; Lee and Smith, 1999). Epermanis and Harrington (2006) and Bohn and Hall (1999)} find premium growth patterns across lines and time consistent with guaranty funds having a negative impact on market discipline, yet both are using aggregation at levels similar to those used in the banking studies to draw these conclusions. The insurance studies to date do not use company-by-state level data, and, to our knowledge, this is the first study to do so. II. Data All financial data are taken from the annual statutory reports made by life-health insurance companies for the 1985 through 2010 calendar years. In addition to balance sheet, income, and cash flow statements analogous (though not identical) to GAAP statements, these reports include detailed information on insurance operations by line of insurance. In particular, at the company-by-state level (i.e., the business of each company broken out by state), companies file a state page for each state that contains the number and face value of ordinary, industrial, group, and credit policies in force at the beginning of the year, issued during the year, and at the end of the year, as well as total life insurance premiums and annuity considerations in the state. This data will form the basis of the dependent variables in our analysis, as they reflect policyholder deposits at the life insurance company. We will focus on how these deposits change in response to changes in company risk, and how that response is affected by the presence or absence of guaranty fund protection. We consider only the state pages for the 50 states and the District 7

of Columbia, and the company-level analysis is based on the aggregation of the data on those state pages (rather than company-level aggregates reported elsewhere in the statutory report). The state page breaks out life insurance in force into four component pieces. Ordinary insurance is the largest line and contains the most familiar contracts, including both term insurance and whole life insurance. Group insurance refers to contracts sold to employers and associations that provide coverage for a set of group members such as employees. Industrial insurance refers to small face amount policies that provide coverage, for example, for funeral expenses. Credit life insurance refers to payment protection insurance or loan repayment insurance that guarantees payment of a loan in the event that a borrower dies. The risk measure for each company is a yearly letter rating issued by A.M. Best. The rating assesses the company's ability to pay policyholder claims and is based on analysis of both public and confidential information. The rating methodology considers a wide range of issues, including risk management practices, operating performance, liquidity, and regulation. A.M. Best also considers company affiliations and holding company structure---it issues grouped ratings where appropriate. Thus, analysis done at the individual company (rather than consolidated group) level implicitly incorporates A.M. Best's assessment of group support. Only firm-years that qualified for a letter rating were included in the analysis. In cases where A.M. Best issued more than one rating for a company in a year, the last rating was used. A.M. Best refined its rating system in 1992 with the introduction of A++, B++, C++, D, and E. The last two ratings are rare, with the latter indicating some level of regulatory supervision. We include in the analyses all firm-years where the firm was rated in the A-range, B-range, or C-range. 3 3 State-level analyses will include those explicitly rated NA (not applicable) or NR (not rated), as well as those not covered by A.M. Best. A.M. Best also maintained a parallel rating structure for portions of the sample period for firms that did not qualify for letter ratings, known as Financial Performance Ratings (FPR). Firm-years with these ratings are discarded in all the analyses except those based on state level market share, where they are categorized with unrated companies. 8

Table 1a shows the distribution of ratings and rating actions within the 1985-2010 sample period. More than two-thirds of letter ratings fall somewhere in the A-range. The probability of a downgrade is about seven percent and is roughly the same in the A-range as outside of it. On the other hand, the probability of an upgrade is substantially higher for companies below the A-range (11.4%, versus the sample average of 5.8%). Table 1b shows the distribution of upgrades and downgrades by year for the sample. Downgrade probabilities ranged from a low of 2.4% in 1986 to a high of 21.6% in 1993, while upgrade probabilities ranged from a low of 1.1% in 2009 to a high of 11.6% in 1994. The year of guaranty fund adoption for each state was determined by researching the statutes. Guaranty funds are financed by an association of licensed companies in each state, with membership being mandatory. The liabilities incurred by the association are paid for by premium taxes or assessments. State guaranty fund protections are usually granted only to resident consumers of state-licensed companies, with some highly restrictive exceptions. 4 The identification strategy for guaranty fund effects is driven by the 15 states (listed in Table 2) that adopted guaranty funds after 1/1/1985. Typical enactments provided $300,000 of protection to claimants, although some states opted for levels as low as $100,000 with initial enactments, while others ventured as high as $500,000. III. Company Risk and Market Discipline: Company Level Analysis In Section IV, we will test how the presence or absence of guaranty fund protection affects changes in annuity considerations, life insurance premiums, and life insurance in force after company downgrades. Here, we first verify that these quantities change in the expected direction after a downgrade, thereby confirming and extending earlier results in the literature. To this end, we adopt Epermanis and Harrington s 2006 methodology and start with simple comparisons of growth differences between downgraded firms and firms with similar ratings which were not downgraded. 4 Typical laws extend coverage to non-residents when 1) the company is domiciled in the state, 2) the company has never held a license in the non-resident's state, 3) the non-resident's state has similar guaranty fund protections in place, and 4) the non-resident is not eligible for protection in another state. 9

We consider growth differences for three different variables. The first variable, TOTGROWTH, measures the logarithm of total insurance in force at the end of year t minus the logarithm of total insurance in force at the end of year t-1. The second variable, LFIN_CHG, provides a similar log growth measure for total life insurance premiums. The third, AC_CHG, provides the measure for total annuity considerations. To be included in our sample, a firm-year had to have a calculable dependent variable (for example, TOTGROWTH is calculable if total insurance in force is positive at the end of year t and year t-1), a letter rating, and a letter rating in the previous year. Epermanis and Harrington truncated their premium growth variables at -1 and 1. We will perform our analyses on both a truncated and a non-truncated sample in this section. Table 3 shows the sample statistics. Control Group Tests We start with control group tests by subtracting the mean growth rates in downgraded firms from the mean growth rates in firms that were not downgraded. Both the downgraded firms and the controls are segmented into three rating groups (Above A-, A-, and Below A-) according to the rating held before the downgrade. We calculate the differences in the year before the downgrade (t-1), the year of the downgrade (t), the year after the downgrade (t+1), and for the two years including the year of the downgrade and the year after the downgrade (t, t+1). 5 The results of this analysis are shown in Table 4a. In the year of the downgrade, the point estimates suggest reductions of between about 4% and 8% in total insurance in force (TOTGROWTH), depending on the initial rating. Estimates for life insurance premiums (LFIN_CHG) are roughly consistent, suggesting 5 The A- level is widely recognized as a key rating level, especially in property-casualty insurance. In addition to Epermanis and Harrington (2006), see Bradford, M. (2003), Big Changes at Kemper Prompt Buyer Concerns Business Insurance, January 6; and Souter, G. (2000), Reliance Downgraded Business Insurance, June 12. 10

reductions ranging from about 3% to 8%. Negative effects are also in evidence in the year after the downgrade, and in the year before the downgrade for highly rated firms. Differences are even larger for annuity considerations (AC_CHG), especially at low-rated firms. In the year of and the year following the downgrade, annuity considerations fall about 19% for highly rated firms, 34% for A- rated firms, and 52% for firms initially rated below A-. In Table 4b, we repeat the analysis but sort the firms into size deciles for each year, with the size being determined according to the level of the dependent variable. 6 Thus, the control group mean for each firm-year is adjusted for size and for time. The results are roughly similar. In Table 4c, we repeat the analysis, adjusting for size and for time, but with the truncated sample. The results are comparable for TOTGROWTH and LFIN_CHG. For annuity considerations, the effects are still economically large and statistically significant, though not so large as those produced by the nontruncated sample. Tables 4d through 4f present analogous control group tests for upgraded firms. The findings are directionally consistent with expectations, as upgraded firms generally experience positive growth relative to the control groups. However, in contrast with the results for downgrades, upgrades at low-rated firms do not appear to be associated with significant changes in growth. Instead, growth after an upgrade appears to be confined to firms rated in the A-range. Fixed Effects Regressions 6 For example, when studying TOTGROWTH, the firms would be sorted into deciles based on their total insurance in force at the beginning of the year. 11

We now consider a more rigorous comparison with a larger set of control variables, including firm and year effects. In particular, we now consider the specification: Here, is one of the dependent variables described earlier or this part of the analysis (TOTGROWTH, LFIN_CHG, or AC_CHG), is a firm fixed effect, is a year effect, is the coefficient on the logarithm of the prior year ending total of insurance in force, premiums or annuity considerations (Log TOT, Log LFIN, or Log AC), is a vector of rating action dummies, rating dummies, and interaction of ratings and rating actions. The results for the truncated sample are presented in Table 5a. The estimates are supportive of the initial findings in the control group tests, although they generally indicate lower effects. Downgrade effects appear to be around -2% to -3% for life insurance, and -9% for annuities. Specifications that split out the effects into interactions between ratings and rating actions yield directionally supportive results, although the usual standards for statistical significance for the downgrade interactions are met frequently only in the case of annuities. The effects for annuities are again strikingly large in some cases, specifically for firms rated A- or lower. Results for the non-truncated sample are shown in Table 5b. The estimates largely confirm those from the truncated sample for insurance in force and annuities (TOTGROWTH and AC_CHG), with the annuity effects actually becoming more negative. The results for life insurance premiums in the case of the expanded specification are now weaker, however. Some evidence of positive upgrade effects appear in both the truncated and the non-truncated samples, with the effects appearing in some coefficients relating to firms rated A- or above. 12

We also repeated the analysis for the 1985-1992 and the 1993-2010 subsamples (unreported), thus confining attention first to the period when some states were without guaranty funds and then to the period when all states had some level of guaranty fund protection. The results obtained for the 1993-2010 subsample are comparable to but overall stronger statistically than those obtained for the full sample, while the 1985-1992 subsample yields much weaker evidence overall on market discipline. IV. Guaranty Funds and Market Discipline To study guaranty fund effects, we first analyze aggregated data at the state level by summing up all insurance in force, premiums, and annuity considerations across companies with the same ratings. We then analyze how these aggregated shares by rating change as the late-adopting states approved guaranty funds, relative to the change in shares in states that already had guaranty funds in place. We then turn to company-state-year level data and analyze how responses in insurance in force, premiums, and annuity considerations associated with downgrades vary with the presence or absence of guaranty fund protection. Market Share Analysis Figure 1 shows the market share of companies rated A- or better from 1985 to 2010, both at the national aggregate level and the average across states. 7 7 Alaska is excluded from the analysis due to a dearth of state pages in the 1980 s. 13

Figure 1 - Share of Total Life Insurance Premiums in Companies Rated A- or Better, 1985-2010 1 0.95 0.9 0.85 0.8 0.75 1985 1990 1995 2000 2005 2010 Average of State Shares National Aggregate Share The aftermath of the junk bond crisis and recession in 1990-1991---which contributed to the noteworthy failure of Executive Life in 1991 and led to changes in regulation and other aspects of institutional infrastructure of the industry---appears to have been accompanied by an increase in the overall credit quality of the industry in the early 1990 s. Figure 2 offers a first glance at comparing states with established guaranty fund systems with the 6 states (other than Alaska) that adopted guaranty funds after 1990. 1 0.95 0.9 0.85 0.8 0.75 0.7 0.65 Figure 2 - Share of Total Life Insurance Premiums in Companies Rated A- or Better 0.6 1985 1987 1989 1991 1993 1995 1997 Post 1990 Adopters (Average of 6 states) Pre 1990 Adopters (Average of 44 states) Thus, at a first cut, there is no obvious evidence of any guaranty fund effects. 14

We proceed with some regressions. We use the basic specification of: so that the dependent variable in state s in year t is modeled as a state fixed effect plus a year effect plus a parameter times an indicator variable that takes a value of one if the state had a guaranty fund approved at the beginning of the year and zero if not. For the dependent variable, we consider, for each state-year, the fraction of credit, group, industrial, ordinary, and total insurance in force, as well as total life insurance premiums and total annuity considerations, that reside in companies rated A- or better. Sample statistics are reported in Table 6 and results in Table 7. The results for total premiums and insurance in force show little in the way of guaranty fund effects, but the analysis by sublines suggests a possible richer subtext. The results suggest that guaranty fund approval was associated with an 8.7% decline in the market share of companies rated A- or better in industrial insurance in force, but corresponding market share increases of 6.2% in ordinary insurance in force and 2.4% in annuity considerations. 8 Company-State-Year Level Analysis 8 We considered two robustness checks here, given the unusual sample statistics for the insurance in force variables in Table 6. Specifically, some observations in the late 1980 s indicated very low market shares (<1%) for A- companies, a result that may have been produced by differences across companies in the units used to report in force insurance. We retried the regressions after discarding observations with the dependent variable showing less than 1% market share and obtained similar results. We also tried the regressions on the 1990-2010 subsample: In this case the point estimates for industrial insurance and annuity considerations were similar (though not statistically significant after clustering standard errors), but the ordinary insurance result disappeared. 15

Perhaps the most exciting opportunity afforded by insurance data in the course of exploring guaranty fund effects lies in the company-state-year level data contained in the state page. Here, we can study how responses to distress at particular companies differed between states with guaranty funds and those without. We study the specification: where i indexes companies, s indexes states, and t indexes years. As before, we consider a set of dependent variables---log growth in total insurance in force, log growth in life insurance premiums, and log growth in annuity considerations. We also consider growth of insurance in force in the sublines--- ordinary, industrial, credit and group. An important difference from the previous analysis of companylevel market discipline, as suggested by the subscripts on, is that these growth variables are now measured at the company-state level rather than the company level. As before, we include the level of the prior year dependent variable in the regression. In addition to state effects, denoted by, the key fixed effect is a company-year fixed effect, We also include variables,, aimed at identifying the effect of guaranty funds. In the simplest specifications, we simply include a variable APPROVED indicating whether the state had a guaranty fund in place at the beginning of the year. Since it is possible that the effect of the guaranty fund depends on the rating of the firm, in other specifications we interact firm rating categories with APPROVED---with the coarsest categories being A range, B range, and C range, and the finest categories lumping A++ with A+ and A with A- (and similar groupings for the B and C ranges). To be included in the regression, a company-state-year observation had to be either downgraded or upgraded in that year within Best s letter rating system, and have a nonmissing value for log growth of the dependent variable (which requires strictly positive levels at both the beginning and end of the year). Since guaranty fund effects are likely to be differ depending on the direction of the rating action, we run 16

the analysis separately for downgraded and upgraded firms. As before, we consider both unadulterated and truncated samples, where the truncation is applied to the growth variables at -1 and +1. Table 8a and 8b present sample statistics for the downgrade and upgrade samples, respectively. Table 9a and 9b show the regression results for, respectively, the non-truncated and the truncated sample of downgraded firms. If guaranty funds reduce market discipline, one would expect the interaction effects to be positive---as consumer fear in the aftermath of the downgrade will presumably be reduced if guaranty fund protections are present. Specification 1, which uses a single indicator variable (APPROVED---which indicates the presence of a guaranty fund in the state in that year) offers little evidence to suggest that guaranty fund enactments were associated with market discipline declines. However, in specifications 2 and 3, in which rating level is interacted with the presence of a guaranty fund, we see effects start to emerge at firms with low ratings, especially in the C-range. There, the point estimates suggest that having a guaranty fund in place increases growth of insurance in force in the ordinary and group lines. The total effect among C-rated firms is on the order of +5% to +10%, with point estimates in the sublines ranging higher than +30%. Tables 10a and 10b show the results for, respectively, the non-truncated and the truncated sample of upgraded firms. Since these are upgrades, one would expect the guaranty fund effects to be reversed. At any risk level, the company improvement will presumably produce some level of increased interest and growth from consumers, but any positive effect would be mitigated by the presence of guaranty fund protection. Thus, negative interaction effects would be consistent with a reduction of market discipline following guaranty fund enactments, and these in fact are what we see at some level across the board in all of the dependent variables studied. As with the downgrade sample, there is evidence of substantial responses at low rating levels---notably for group insurance at the B++. But there is also evidence at higher rating levels, with statistically significant negative effects appearing for life insurance premium growth and total insurance in force growth as high as the A/A- level. 17

V. Discussion At the company level, the results for life insurance indicate downgrade effects for premiums and insurance in force in the mid-single-digit range. These are roughly consistent with the findings of Baranoff and Sager (2007), who find a negative effect of about 6% for policy counts. This paper was the first (to our knowledge) to examine the association between downgrades and changes in annuity considerations, and these effects appear to be much larger than those for life insurance. Estimates of annual negative growth associated with downgrades fell in the area of -10% to -20% for companies rated A- or higher, and they reached as low as -50% for low-rated companies. There are several possible explanations for the greater sensitivity of annuities. First, many annuities are single premium annuities. Ordinary and industrial life insurance, on the other hand, feature long payment periods stretching out over 20 years or more, meaning that much more of the annual premium in life insurance flows from legacy contracts rather than new business. Second, the switching costs associated with life insurance are high: In addition to termination fees, the front-loading of premiums means that the typical switcher will be paying much higher rates for a similar period of coverage. Moreover, a new contract will invariably require new underwriting, so a switcher also faces reclassification issues. Although termination fees are also present with annuities, the reclassification and re-rating issues are not---as most annuity pricing is not sensitive to medical condition and, naturally, is more attractive for older people. Finally, many institutional contract holders of group annuities may be especially sensitive to risk, as a number of state guaranty funds do not cover unallocated annuities. The influence of guaranty funds is studied by comparing growth in state-years without guaranty funds against state-years with guaranty funds for firms experiencing a change in risk. For downgraded firms, we did not find much of an overall effect, but, on closer inspection, found some response at firms with 18

very low ratings---with guaranty funds being associated with increased overall growth of insurance in force of 5 to 8 percent at downgraded firms with ratings in the C-range. Effects at upgraded firms were easier to detect: Overall impact on insurance in force was a growth decrease on the order of 2 to 5 percent, with most of the contribution coming from companies with ratings in the B- and A- ranges. To the extent that subline effects can be discerned, guaranty fund effects appeared to be most pronounced for group life insurance. While effects were detected, one might well ask why they are not larger. After, all, during the key period of identification during the late 1980 s and early 1990 s, the typical state enactment featured protection in the range of $200,000 to $300,000, which would seem especially important for smaller policyholders in industrial insurance. However, awareness among consumers was probably quite low in the late 1980 s and early 1990 s. In contrast with the situation for banks, where FDIC protection is openly communicated to depositors, many states at the time barred insurance sellers from advertising the presence of guaranty fund protection to consumers. Even agents themselves may have been lulled to complacency over the perceived security of the industry: A.M. Best did not bother listing state guaranty fund limits in their annual Reports until the early 1990 s, at which point the failures of Executive Life and some others had debunked the notion that major life insurance companies could not fail. While this paper focused on the enactment of guaranty funds, there are other possibilities for future research in both the life and property-casualty industries. Coverage varies across states, and also across lines. For example, annuities typically have different sublimits than death benefits, and, as noted earlier, unallocated annuities are covered in some states but not others. Exploiting such variation, especially during the period of the 1990 s and 2000 s when insolvency risk sensitivity may have been higher than in the 1980 s, offers another opportunity to identify the impact of guaranty funds. 19

VI. References Anginer, D., Demirguc-Kunt, A., and Zhu, M. (forthcoming), How Does Deposit Insurance Affect Bank Risk? Evidence from the Recent Crisis, Journal of Banking and Finance. Baranoff, E.G., and Sager, T.W. (2002), The Relations among Asset Risk, Product Risk, and Capital in the Life Insurance Industry, Journal of Banking and Finance 26, 1181-97. Baranoff, E.G., and Sager, T.W. (2007), Market Discipline in Life Insurance: Insureds Reaction to Rating Downgrades in the Context of Enterprise Risks, working paper. Billett, M.T., Garfinkel, J.A., and O Neal, E.S. (1998), The Cost of Market versus Regulatory Discipline in Banking, Journal of Financial Economics 48, 333-358. Brewer III, E., Mondschean, T.S., and Strahan, P.E. (1997), The Role of Monitoring in Reducing the Moral Hazard Problem Associated with Government Guarantees: Evidence from the Life Insurance Industry, Journal of Risk and Insurance 64, 301-322. Bohn, J.G., and Hall, B.J. (1999), The Moral Hazard of Insuring the Insurers, in The Financing of Catastrophe Risk, Kenneth Froot (ed.), Chicago: University of Chicago Press. Chen, Y., and Hasan, I. (2011), Subordinated Debt, Market Discipline, and Bank Risk, Journal of Money, Credit, and Banking 43, 1043-1072. Cummins, J.D., and Danzon, P.M. (1997), Price, Financial Quality, and Capital Flows in Insurance Markets, Journal of Financial Intermediation 6, 3-38. Demirguc-Kunt, A., and Detragiache, E. (2002), Does Deposit Insurance Increase Banking System Stability? Journal of Monetary Economics 49, 1373-1406. Demirguc-Kunt, A., and Huizinga, H. (2004), Market Discipline and Deposit Insurance, Journal of Monetary Economics 51, 375-399. Diamond, D.W., and Dybvig, P.H. (1983), Bank Runs, Deposit Insurance and Liquidity, Journal of Political Economy 91, 401-419. Downs, D.H., and Sommer, D.W. (1999), Monitoring, Ownership, and Risk-Taking: The Impact of Guaranty Funds, Journal of Risk and Insurance 66, 477-497. Eling, M. (2012), ``What Do We Know about Market Discipline in Insurance?'' Risk Management and Insurance Review 15, 185-223. Eling, M. and Schmit, J.T. (2012), Is There Market Discipline in the European Insurance Industry? An Analysis of the German Insurance Market, Geneva Risk and Insurance Review 37, 180-207. Epermanis, K., and Harrington, S.E. (2006), Market Discipline in Property-Casualty Insurance: Evidence from Premium Growth Surrounding Changes in Financial Strength Ratings, Journal of Money, Credit, and Banking 36, 1515-1544. Flannery, M. (1998), Using Market Information in Prudential Bank Supervision: A Review of the U.S. Empirical Evidence, Journal of Money, Credit, and Banking 30, 273-305. 20