Home State Loyalty: An Examination of Investments in States of Property-Casualty Insurers

Transcription

1 Home State Loyalty: An Examination of Investments in States of Property-Casualty Insurers Dara Marie Marshall, Ph.D. Assistant Professor, Accountancy Miami University 3087 Farmer School of Business 800 E. High St. Oxford, OH and Willie Dion Reddic Joseph I. Lubin School of Accounting Whitman School of Management Syracuse University 721 University Avenue Syracuse, New York

2 HOME STATE LOYALTY: AN EXAMINATION OF INVESTMENTS IN STATES OF PROPERTY-CASUALTY INSURERS ABSTRACT This paper investigates investment decisions of Property-Casualty Insurers in state-level municipal securities. Our primary hypothesis is that insurers are prone to make investment decisions based on familiarity, similar to what has been documented in individual investor behavior in the finance literature. This leads to our prediction that insurers with a higher proportion of direct premiums written and that are domiciled within a particular state are more likely to allocate higher proportions of their state-level investments to municipal securities in that state. We find that property-casualty insurers rationally diversify their state bond portfolio away from states where they write policies. We also find however, similar to individual investors, property-casualty insurers invest more of their state bond portfolio in their domicile states and even more in their domicile state where they write policies. Our results support familiarity theory of investment and contribute to prior literature on municipal bonds. Keywords: Insurance Investments, Government Investments, Regulatory Impacts, Municipal Securities, Familiarity Theory 2

3 I. Introduction The municipal market makes up approximately $3.7 trillion and offers investors high credit quality investments whose income is largely tax-exempt. The U.S. insurance industry s significant funding of municipal entities represented 13.0% of this market as of year-end Municipal securities are issued by states or local governments to fund a variety of government expenditures and investments including transportation and transit, water and sewer, electric and gas utility, higher education, housing and development, school district, and medical facilities (NAIC, 2012). Therefore, why do insurers decide to invest in this particular market? More importantly, are insurers more likely to invest in municipal securities where they conduct most of their business operations in? For example, is there some form of home state bias for insurers when making these investment decisions? The purpose of this study is to explore whether insurers allow their business relationships with states to drive their state-leave municipal investment choices. More specifically we test whether direct premiums written insured in each state affect insurers decisions to invest in that state. Insurers, like other investors, are concerned about the financial attributes of their investments; however, insurers are also concerned about the non-financial attributes of their investments as well. For example, it would seem more likely for an insurer to invest within their community where they conduct business. Consequently, insurers by design are well-diversified in their investment strategies since they have to fall within certain regulatory guideline requirements. By adopting a familiarity theory framework as proposed by Huberman 2001, we find that insurers are more likely to invest in municipal securities where they are home domiciled and where they conduct most of their business operations. The study focuses on insurers decisions of investing in states (i.e., familiarity) where they conduct business. Prior literature suggests that investors gravitate more towards investments 3

4 that the investor is more familiar with in terms of home state familiarity (e.g., Kilka and Weber 1997; Coval and Moskowitz 1999, Campbell, 2000, Huberman 2001, Massa and Simnov 2004). For example, Coval and Moskowitz (1999) show that US investment managers favor locally headquartered firms when deciding on their investment portfolio. Huberman (2001) provides evidence that investors do invest in investments that they are more familiar with; in our case, home state familiarity and business operations. Familiarity theory would predict that the more business an insurer does within a state, the more likely it would invest in that state. By examining the insurance market, we can directly observe the business operations (e.g., investments and underwriting) that insurers have with different states. States/Cities/Counties sell municipal securities to provide services to their residents. These securities are usually issued by states, cities, and school districts and are used for capital improvements such as new buildings, new schools, new roads, etc. This provides an economic advantage to the insurer. For example, if a new building is constructed by a municipal bond then the contractor must be insured. If the insurer is located in the state where the contractor lives then there is a high probability that the insurer provides the insurance for that particular contractor. The insurer will receive both coupon payments on the municipal bond and collect premiums from the contractor. Our contribution is among two dimensions. First, we design and test whether an insurer is less likely to invest in municipal securities where they conduct most of their business operations. In these analyses we regress the proportion of an insurer's total investments in a particular state scaled by the insurer s total investment in all state level securities on the proportion of net premiums written in that state to total net premiums written. Second, we test whether an insurer is more likely to invest in municipal securities where they conduct most of their business and if that insurer is home domiciled in that state. We show 4

5 that insurers with higher proportion of direct premiums written and are home state domiciled are more likely to increase their investment in municipal securities within that state. The remainder of this paper is organized as follows. The next section reviews the related literature. Section 3 states the hypotheses and illustrations, and Section 4 discusses the data and specification. Section 5 reports the results and Section 6 reports sensitivity analyses. Section 7 concludes and discusses the limitations of this study. II. Background Investment Decisions of Insurers Property and casualty insurers derive their income from two sources: underwriting and investing. Income from underwriting can be volatile and is historically negative (Fairley, 1979.) Thus, investment decisions of this industry are paramount. Since we examine insurers state investment decisions in relation to the amount of underwriting activity within a state, we first must understand the relationship between underwriting and investment. Prior research has examined how underwriting may or may not affect investment decisions. This research either attempts to answer questions such as: what should an optimal investment portfolio of a P&C insurer be, what is the performance of P&C investment portfolios and how does it compare (usually to historical market returns or other industry investment portfolio returns,) and do investment portfolio decisions affect shareholder value? The effect of underwriting The examination of the relationship between underwriting and investment within a property and casualty insurer is motivated by the differing preferences between shareholders and policyholders. Lambert and Hofflander (1966) describe the conflict between policyholders and shareholders of P&C insurers as a conflict between two goals: liquidity (for unexpected losses) and higher investment income (which may decrease liquidity.) In their 1966 study, the authors 5

6 attempt to measure underwriting riskiness which could then be used to determine the appropriate level of investment portfolio risk. The authors compute variances of different lines of insurance and covariances between lines of insurance and find that certain lines if provided together by a firm would be more risky. They do not find large covariances between insurance lines and the market as they had predicted. The relationship between underwriting and investment is further examined in Kayane and Nye (1975.) In their study, they model the optimal portfolio (or company composition) which minimizes risk for a given level of return. This solution is achieved through the weighting of different insurance lines and investments. In their optimal solution, only four lines of insurance would be offered and the investment portfolio would be weighted around 85% toward equities. Realizing this does not describe the property and casualty insurance industry, the authors relax some of their model constraints and find solutions that are closer to the mix of investments and insurance lines that make up the investment and underwriting lines of P&C insurers. Instead of developing a model or examining measures from underwriting and investment, Johnson (1972) surveys members of investment departments of P&C insurers in order to determine what affects investment decisions. The survey the author uses was commissioned by the Securities and Exchange Commission and includes 25 insurance groups that represent 58% of the P&C market in their survey period. The majority of respondents did not agree that liquidity needs or the mix of insurance lines offered affected portfolio decisions. However, some respondents felt that increases in underwriting volume affect liquidity needs which could then affect investment decisions. Johnson also found that P&C insurers used simple measures for portfolio performance and did not measure performance regularly. 6

7 From these and other related studies, the relationship between underwriting and investment in P&C insurers is not clear. It would seem that liquidity needs would affect the riskiness of investment portfolios or at least the proportion of securities that can be quickly sold at a minimal loss (Lambert and Hofflander, 1967.) Though these early studies may not have had the power to show a relationship, later studies examine the composition of P&C investment portfolios and their performance. Through these analyses, further insight is gained on the determinants of P&C insurers investment decisions. Investment portfolio composition The preceding studies attempted to prescribe the optimal investment portfolio of P&C insurers. The following studies describe P&C investment portfolios and thus may lend support to theories of what the optimal portfolio should be and the relationship between investment and underwriting within the firm. Monroe and Trieschmann (1972) compare the performance between P&C insurer investment portfolios to the stock market and portfolios of investment companies. P&C investment portfolio returns are lower than both market and investment company portfolio returns. However, the authors fail to reject the null that risk-adjusted returns are equivalent to market portfolio returns or investment company portfolio returns. These results imply that P&C portfolios are less risky which would support Lambert and Hofflander (1967) theory that liquidity drives investment decisions, assuming liquidity needs are higher in P&C insurers compared to the market or investment companies. Petroni and Wahlen (1995) examine the value relevance of P&C investment portfolio fair value disclosures. Though their study s focus is on reporting, they provide descriptive evidence of the composition of P&C investment portfolios. Fixed income investments make up the 7

8 majority of P&C insurers assets in their sample ( ) For fixed income investments, Petroni and Wahlen describe the types of securities held and the likelihood that different securities are held for different time horizons. In their sample, municipal securities made up 40% of the average P&C insurer s fixed income portfolio. Since the authors use P&C insurers 10-K filings in order to measure the effect of reporting, detailed investment portfolio data that show time to maturity for different types of investments is not available. By correlating the disclosed proportions of the fixed income portfolios of insurers by type of investment to disclosures on the proportions of fixed income portfolios by time to maturity, the authors show the likely maturity times for different fixed income investments. Holdings in U.S. treasuries by P&C insurers likely mature in less than one year or in one to five years. Holdings in municipal fixed income investments or corporate investments likely have longer times to maturity (five to ten years or longer.) Similar to the composition of investment portfolios in Petroni and Wahlen s (1995) sample, P&C insurers investment portfolios on average were made up of 70% bonds and debt instruments in Cummins and Grace s (1994) sample. Two possibilities come from the theories and models of what comprises an optimal P&C insurer investment portfolio and the empirical evidence of observed P&C investment behavior: underwriting and investment are negatively correlated or underwriting and investment are not related. Though Lambert and Hofflander (1966) do not find empirical support for a relationship between underwriting and investment, versions of their theory have been modeled by other researchers (Kahane and Nye, 1975 and Fairley 1979.) Vaughn (1998) also models a relationship between underwriting and investment where investment decisions affect premium pricing. Though these studies may differ in the mechanisms on how underwriting affects investment or vise-versa, they all show a negative relationship between underwriting and investment risk or 8

9 underwriting and investment return. This provides us a starting point of what the expected relationship between investment and underwriting should be. Familiarity The previous discussion on P&C investment portfolio literature shows that P&C insurers tend to make less risky investment decisions. This is most likely due to the importance of investment income to total income and the potential that investment income is used to cover underwriting losses. There is evidence that investors do not always make investment decisions based on portfolio theory. In particular, we examine the literature on familiarity, which is a theory that investors overweight their portfolios in investments based on either real or believed greater knowledge of that investment relative to others or simply home bias toward that investment relative to others. There are several reasons why investors may exhibit behavior of familiarity (Anderson et al., 2011) but most researchers recognize two. One reason is based on psychology theories about the use of heuristics in judgment and decision making. Heath and Tversky (1991) conduct four different experiments and review a number of others that examine the tendency of subjects to choose an equivalent bet over another where the only difference between the bets is their belief that they have more knowledge about one of the bets. The authors refer to this as the competence hypothesis and explain that it is driven by attribution. Thus even when provided with additional information on the unfamiliar option, subjects would still exhibit this behavior because they feel more competent in the familiar option and thus can take credit for success and place blame outside of themselves for failure. Later studies distinguish the psychology based reason for familiarity from rational decision making based on imperfect information (Shapiro, 2002; Merton, 2012.) Massa and Simonov (2004) call this pure familiarity and others refer to it by one measure for it, investor 9

10 sophistication (Karlsson and Nordén, 2007; Grinblatt and Keloharju, 2001.) What distinguishes this reason from the psychology based reasoning is that this behavior decreases as more information is provided on the unfamiliar alternative. Massa and Simonov (2004) use a sample of individual investment portfolios in Sweden from combined with demographic data and find that investors rationally hedge their non-investment income by tilting their investment portfolios to be negatively correlated when they are more sophisticated. Unsophisticated investors still exhibit familiarity bias by having their investment portfolios positively correlated with non-investment income. Grinblatt and Keloharju (2001) and Karlsson and Nordén (2007) find similar results using the proportion of home investments in a portfolio as their measure of familiarity. Grinblatt and Keloharju (2001) s analysis is within a country based on distance and language whereas Karlsson and Nordén (2007) compare domestic investment to international investment. There is evidence that institutional or sophisticated investors also exhibit familiarity investing in addition to the studies that examine individual or retail investor behavior. Lütje and Menkhoff (2007) survey fund managers in Germany to determine familiarity based on psychology theory and rational choices based on imperfect information. They find fund managers exhibit familiarity investing behavior based on both reasons measured by return optimism, overconfidence, higher age, lower bonus and less experience. Anderson et al. (2011) examine fund portfolios of institutional investors worldwide and test an alternative theory for familiarity based investing culture. Using Hofsted measures of culture, their study finds that institutional investors over allocate their portfolios to home securities based on culture measures and also based on distance. 10

11 III. Hypotheses We examine whether property and casualty insurers exhibit familiarity based investing. Property and Casualty insurers can only be domiciled in one state and are subject to different regulations and tax policies in each state they can sell insurance (Petroni and Shackelford, 1995.) Also, P&C insurers must file statutory reports that detail all other their investment holdings. P&C Insurers invest in multiple types of securities but given their large participation in the municipal bond market and large allocation of their investment portfolios to fixed-income securities we choose to focus only on U.S. state-level investments. The state level insurance market provides us a unique opportunity to examine familiarity investing in large institutional investors. The prior literature discussed above describes the pressure P&C insurers face to maintain sufficient investment income in order to possibly absorb losses from underwriting and in the case of publicly held insurers maximize firm value (Lambert and Hofflander, 1967.) In order to maximize firm value while minimizing risk, P&C insurers should diversify their various income streams (Lambert and Hofflander, 1966; Kahane and Nye, 1975; Fairley, 1979; and Vaughn, 1998) P&C insurers earn income from investments and underwriting different insurance lines in different states. Similar to prior studies, Chen et al. (2012) finds support to the hypothesis that insurers hedge investment income to underwriting income. They find underwriting characteristics drive investment decisions by modeling and empirically testing the relationship between claim durations and maturities of US treasuries holdings. Their model shows that investment maturities are matched as a hedge against interest rate risk related to claims. We test whether P&C insurers use utility maximizing investment strategies, or invest based on familiarity by examining their state-level municipal securities investment portfolio, their state of domicile, and state level business activity. Assuming that underwriting income and 11

12 investment income are correlated, an insurer could hedge different income streams against one another to maximize value and minimize risk. An insurer trying to hedge against underwriting risk should invest in something that is not correlated with potential claims. Assuming that underwriting losses within a state are correlated to liquidity of municipal securities issued by that state, then the proportion of state securities held relative to the entire state-level fixed income investment portfolio should be negatively related to the proportion of net premiums written within that state to total net premiums. Insurers operate in multiple states and have to learn multiple markets. The more business done within a state relative to other states may increase an insurer s knowledge of that state relative to others. Though the knowledge gained from underwriting within a state is related to that state s regulations, taxes, and insurance market, insurers may also increase their knowledge of that state s investment risk relative to others, whether true or believed. This could lead to insurers using familiarity, which would imply investing more that state relative to others. Familiarity investment and rational hedging present two different possibilities of the relationship between underwriting within a state and investing within a state. The first follows the literature describing the relationship between underwriting and investing which predicts that insurers hedge underwriting income to investment income. Thus in our setting, insurers underwriting income, as proxied by the proportion of direct premiums written in a state relative to other states, is negatively correlated with the proportion of investment in municipal securities in that state relative to other states. The second possibility is that insurers investment decisions are based on familiarity. This would imply a positive relationship between underwriting within a state and investments within that state. Thus, we test the following hypothesis, stated in null form: 12

13 H1: An insurer s proportion of direct premiums written within a particular state has no effect on the insurer s proportion of state investment portfolio allocated to municipal securities in that state. There are some reasons not to expect any relationship between underwriting and investment. Petroni and Wahlen (1995) show that fair value disclosures in municipal fixed income securities are not value relevant. Therefore investors may not pay attention to these disclosures. This could be due to the information content being too noisy because of longer maturities and thinner markets. Thus there may not be an incentive to hedge underwriting and investment income at all. Also, Vaughn s model implies that investment risk drives premium prices. We do not measure overall investment risk and the proportion of state-level investments allocated to a state does not measure the riskiness of the state-level portfolio. One reason to not expect familiarity to influence the state investment decisions is that P&C insurers do not necessarily have to become knowledgeable in every state market. Petroni and Shackelford (1995) document insurers decisions on where to domicile based on state tax and regulatory requirements. They describe that when insurers expand into other states, they can make the choice of either establishing a subsidiary that is domiciled in that state or getting a license to sell insurance in that state. By getting licensed instead of establishing a domiciled subsidiary, insurers may be able to continue to only abide by their home state s regulations in the new state through substantial compliance. Not every state allows this but many do. This implies that insurers may not consider themselves familiar with a state based on doing business in that state alone. Insurers domiciled within a state are subject to all of that state s regulations and taxes. By being domiciled within a state an insurer may believe or may be more competent of that state s 13

14 investment characteristics. We hypothesize that insurers with the same home state domicile as the state-level investment will have a higher allocation of their state-level investment portfolio in municipal securities in that state. Thus, we test the following hypothesis, stated in alternative form: H2: Insurers domiciled in the same state as the investment state have a higher proportion of their state investment portfolio allocated to municipal securities in their domicile state. Hypothesis one implies that insurers will rationally diversify their overall income by investing in state-level securities that are less likely to be correlated with their underwriting income or that insurers will have higher investment holdings in states where they conduct more business due to familiarity. Hypothesis two implies that domiciling within a state does give insurers the belief that they are more competent in the investment characteristics of that state versus others. Given that hypothesis one could lead to a negative relationship between underwriting within a state and investment within that state, we also explore how an insurer s business relationship interacts with where they are domiciled. We predict that familiarity will dominate in this interaction. The third hypothesis, stated in the alternative form, is: H3: Insurers with a higher proportion of direct premiums written in their domicile state have a higher proportion of their state investment portfolio allocated to municipal securities in that state. To illustrate this scenario (insurers with a higher proportion of direct premiums written and home state domiciled are more likely to increase their investment in municipal securities in that state), assume that an insurer [Auto-Owners Insurance Co.] who is home state domiciled in Michigan and writes a major proportion of their business (16.83%) in Michigan. This insurer will 14

15 be more partial to invest in municipal securities in the state of Michigan since they are more familiar with the state. Therefore, based on familiarity theory, we propose that insurers will be more likely to increase their investment exposure in their home state domiciled when that particular state is also where they conduct most of their business. We only focus on one section of the fixed-income investments within an insurers portfolio state-level municipal securities. The amount of premiums written and domicile is measured at a state level so we focus on state-level securities as opposed to corporate, national or international. IV. Data and Specification Specifications We use the following Pooled OLS specifications to test our hypotheses: ActualCostsj,i = β0 + β1directpremiumwrittenj,i + β2samestatedomicilej,i + β3[directpremiumwrittenj,i X SameStateDomicilej,i] + Σ βicontrolsj,i (1) We expect insurers having more business within a state (higher DirectPremiumWritten) will have less invested in that state relative to other states, if they hedge, or that they will investment more in that state relative to other states, if they use familiarity, thus β1 0. Insurers domiciled within a state (SameStateDomicile = 1) will have more invested in that state relative to other states; thus we expect β2 > 0. If an insurer with a higher proportion of direct premiums written in their home state will be more likely to increase their municipal security investments in their home state domicile, then we expect β3 > 0. The dependent variable is the actual cost of investment holdings within a state scaled by the total of all state level investment holdings for an insurer in a particular year. We include two proxies to measure familiarity presence. First, for the subset of entities with available data, we include direct premiums written by an insurer within a 15

16 state scaled by total premiums written by that insurer in a particular year. Second, we set an indicator variable to one if the insurer is home state domiciled and the insurer invest in that particular state, else zero. To control for insurer characteristics, we include NetAdmittedAssets defined as the logarithm value of net admitted assets. This variable controls for insurer size (Beaver and McNichols, 1998). We expect a negative relationship between the size of the insurer and the investment decisions of an insurer to purchase municipal bonds in their home state. We use measures from the government accounting literature to control for the characteristics of state municipal bonds that drive investment decisions. Though there a many prior accounting studies on municipal bonds (Raman,1981; Copeland and Ingram, 1982; Howard and Wilson, 1984; Reck and Wilson, 2006) there currently is not a theoretical basis for including certain fiscal measures instead of others (Reck and Wilson, 2006.) We attempt to control for fiscal capacity by including GFBal which is the scaled general fund balance of a state. Reck and Wilson (2006) examine returns on municipal bonds in the secondary market and use this measure in their model. It is calculated as total general revenue minus total general expenditure scaled by total general revenue as defined by the U.S. Census. We control for size by including the natural log of a state s population (SIZE) following Howard and Wilson (1984.) Prior accounting studies on municipal bond ratings, yields, and interest costs include a measure of debt burden. We include DebtPerCapita calculated as total debt outstanding scaled by population (Copeland and Ingram, 1982; Reck and Wilson, 2006.) To measure fiscal self-reliance we use GeneralRevOwnSourcesPerCapita following Raman (1981) and Copeland and Ingram s (1982) studies on bond rating changes. We calculate this as general revenue from own sources as defined by the U.S. census, scaled by population. We include Surplus as a measure of current 16

17 operating performance, which is an indicator equal to one if there is a current year positive general fund balance and zero otherwise. -Insert Table 1 here- Data We obtain data on insurers and municipalities from two sources: the National Association of Insurance Commissioners (NAIC) and the Census Bureau Annual Survey of State and Local Government Finances. The NAIC is the U.S. standard-setting and regulatory support organization created and governed by the chief insurance regulators from the 50 states, the District of Columbia and five U.S. territories. Through the NAIC, state insurance regulators establish standards and best practices, conduct peer review, and coordinate their regulatory oversight. The Census Bureau conducts an annual survey called the Annual Survey of State and Local Government which includes income statement and balance sheet type items. Participation is required every 5 years, but is voluntary in other years. Because there is a lag between when the Census conducts its survey and when results are available, we use the sample period of Table 2 describes our sample selection process. There are several schedules that make up NAIC statutory reports but we only use two. Schedule T and the Assets page contains data on our DirectPremiumWritten and NetAdmittedAssets variables for 2,821 insurers for all 51 (District of Columbia included) states over a four year period, respectively. This amounts to 575,484 observations. To obtain the variables on insurers investments we use Schedule D. Within schedule D we only use the section on state-level investments. Investments are listed individually thus we aggregate investments by state by insurer only for issuer obligations. These investments have text descriptions which we then match to state names. From the 80,549 individual investments that matched U.S. state names, 36,909 observations represent the aggregate of these 17

18 investments by state by insurer by year. We then merge our investment observations from schedule D to our insurer observations from schedule T and the Assets Pages. From Panel B and D we can see that not every insurer of the 2,821 insurers in the NAIC database invests in statelevel municipal securities. Because not investing is a valid observation, we only eliminate insurer-state-years where state-level investment is zero and direct premiums written is zero or insurer-state-years where both state-level investment and direct premiums writing are missing. This leaves us with 102,757 insurer-state-years which we then merge with census data. Our final sample size with no missing values is 90,495 insurer-state-years. -Insert Table 2- Descriptive Statistics Table 3 presents state descriptive statistics. Column 1 shows all 50 states and column 2-4 shows the average of the main variables of interests. The 10 most invested states among all insurers in our sample are Massachusetts ($8,231,627), California ($6,638,349), Washington ($5,907,727), Florida ($5,446,744), Illinois ($5,285,365), Georgia ($5,181,979), Pennsylvania ($4,608,276), Connecticut ($4,564,252), North Carolina ($3,718,173), and Wisconsin ($3,645,332). The 10 largest amounts of policies written by insurers in the following states are California ($77,896,680), New York ($45,211,460), Florida ($41,020,850), Texas ($30,840,040), Michigan ($24,137,840), New Jersey ($22,888,030), Pennsylvania ($22,716,030), Illinois ($22,048,600), Ohio ($16,703,930), and Georgia ($15,990,860). The most invested states in which an insurer invests (by percent) in and by an insurer home state domicile are Illinois (0.06), Connecticut (0.05), New York (0.05), Ohio (0.05), Pennsylvania (0.05), Wisconsin (0.05), California (0.04), Louisiana (0.04), Texas (0.04), and Florida, Michigan, North Carolina (0.03). Insurers that are located in the Midwest and South are more likely to invest back into their state of domicile, and more likely to write a larger portion of their business in that state. Table 3 18

19 presents descriptive statistics for state level investments by year. From an initial glance at the TotalActualCosts_AllStates, the mean suggest that a vast amount of insurers are investing in certain states. However, insurers direct premium written decreased from 2006 to Insert Table 3 here- A correlation table is presented in Table 4. The correlation between ActualCostsj,i and DirectPremiumWrittenj,i is positive and significant which shows that insurers that have a higher proportion of direct premiums written tilt their state bond portfolio toward states where they write policies. The correlation between ActualCostsj,i and SameStateDomicilej,i is positive and significant which indicates that insurers invest more of their state bond portfolio in their domicile states. -Insert Table 4 here- Table 5 provides descriptive statistics of variables used in analyses. We have 90,495 insurer-state-year observations. The means for ActualCosts, DirectPremiumsWritten, and SameStateDomicile are 0.044, 0.042, and 0.017, respectively. As stated earlier, the ActualCosts variable is our dependent variable. On average, insurers invest 4.4 percent of their investments in municipal securities. The DirectPremiumsWritten and SameStateDomicile are our independent variables of interest. On average, insurers write approximately 4.2 percent of their direct premium written in each state and 1.7 percent of insurers are home state domiciled as the state where they invested (SameStateDomicile.) -Insert Table 5 here- Table 6 provides a comparison of State Investments by Insurers domicile state. We have 1,512 insurer-state-year observations. The same domicile state mean for our dependent variable ActualCosts is 28.3 percent and the independent variable is 48.6 percent. -Insert Table 6 here- 19

20 V. Results In our sample, state_actual_costs and its scaled version, ActualCosts only have positive values. Also, 60,357 of our 90,495 insurer-state-year observations have a value of zero. One of the assumptions in the ordinary least squares model is normality of the dependent variable. In our sample, if we assume our dependent variable, ActualCosts, is really a transformation of an underlying normally distributed variable that can take on all values, positive and negative (i.e. the value of a state s municipal securities that satisfies the solution to maximizing an insurer s utility function) then our sample is censored. This means that our zero observations represent real investment choices and we can observe some of the exogenous variables that determine that decision. (Amemiya, 1984) When the dependent variable is censored, OLS estimates will be biased and inconsistent. Amemiya (1984) shows that if all of the explanatory variables in the model are normally distributed the OLS estimator will be consistent but still biased. In the case of censored samples, a type 1 Tobit (or standard Tobit) will yield a consistent estimator of beta. Table 7 presents analyses of the sample using a standard Tobit model. We perform these analyses as an additional sensitivity check because of the properties of our dependent variable. The coefficients are interpreted similarly to OLS except they measure the effect on the latent variable. For each model (Base Model and Base Mode with Interaction) the coefficient β1<0, the effect of a DirectPremiumsWritten is negative and significant, implying that insurers with a higher proportion of direct premium written have lower investment allocations in municipal securities in that state. Therefore, the null hypothesis stated in H1 is rejected. The economic intuition is that insurers with a large proportion of direct premiums written within a state (i.e., a large proportion of their business operations is within a particular state) should diversify their state municipal securities away from that particular state to lessen the chance of experiencing losses from both the underwriting and investment sides of their business simultaneously. We do 20

21 not examine diversification within investments but rather diversification across different streams of income - underwriting and investment income. To test H2, we observe the coefficient on SameStateDomicile is positive and significant β2>0, which is consistent with our alternative hypothesis stated in H2. Insurers allocate more of their state-level investment portfolios to municipal securities issued by their state of domicile. To test H3, we observe the coefficient β3>0 for the base models with interaction, which rejects the null hypothesis stated in H3. An insurer that writes a large portion of their business within their domicile state has more incentive to increase their investment in municipal securities in that state since a majority of their business operation flows through that specific economy. -Insert Table 7 here- VI. Sensitivity Table 8 presents sensitivity analyses of the sample using a standard Tobit mode where we use bootstrap procedures. We perform these analyses as an additional sensitivity check because we are pooling our data. Our results remain unchanged from Table 7. -Insert Table 8here- VII. Conclusion This study explores whether direct premiums written insured in each state affect insurers decisions to invest in that state. We determine that insurers with a large proportion of their direct written premium in within a particular state are less likely to increase their municipal securities investment in that state by 6.4 percent. Also, we find that insurers with a large proportion of their direct premiums written and are home state domiciled are more likely to increase their municipal securities investments by approximately 30 percent, when including state and year effect To our knowledge, this is first study to examine this relationship for a sample of US insurers and the 21

22 first to incorporate familiarity theory for this purpose. One caveat to this study is that there is some form of home state bias for insurers when making municipal investment decisions in that state. If regulators, investors, and politicians fully understood the reasons of why insurers decide to invest in that particular state, then more firms could perhaps follow the same suit. This story should not be confused with optimal portfolio strategies story among insurers, investors, or firms, which is what we have not explicitly modeled here. The evidence of familiarity theory of investment has not been brought into the firm level setting. Although investment decisions among insurers can in some ways be quite similar to investors. We are able to use familiarity theory on a state level in an insurer platform. Moving forward, by incorporating optimal investment portfolio theory we could probable determine if familiarity theory is provides a stronger determinant of investment strategies among insurers. 22

23 Variable ActualCosts DirectPremiumsWritten SameStateDomicile = = = Table 1 Variable Definitions Definition The actual cost of investment holdings within a state scaled by the total of all state level investment holdings for an insurer in a particular year direct premiums written by an insurer within a state scaled by total premiums written by that insurer in a particular year Indicator variable with a value of 1 if an insurer's domicile state is the same as the state in which it invests and 0 otherwise NetAdmittedAssets GFBal = Log of total net admitted assets = SIZE = Log of total population within a state DebtPerCapita = Total General Revenue - Total General Expenditure scaled by Total General Revenue as defined by the U.S. Census Ratio of total debt outstanding to total population calculated as Account Type 1/ population as defined by the U.S. Census GeneralRevOwnSources PerCapita Surplus = = General Revenue from own sources as defined by the U.S. census scaled by population Indicator variable with a value of 1 if General Revenues - General Expenditure is positive or zero and 0 otherwise state_actual_costs = The acutal cost of investment holdings within a state in a particular year state_fair_value = The fair value of investment holdings withing a state in a particular year dpw = Direct premiums written within a state within a given year TotalActualCosts_AllStates = Total of all state level investment holdings for an insurer within a given year at actual cost TotalFairValue_AllStates = Total of all state level investment holdings for an insurer within a given year at fair value 23

24 Table 1 continued Variable Definitions debt_outs = Total debt outstanding as defined by the U.S. Census total rev = Total revenue outstanding as defined by the U.S. Census total_ expenditure = Total expenditures as defined by the U.S. Census population = Estimated population from the U.S. Census Source: Assets Page, Schedule T, Schedule D, and the U.S. Census Bureau 24

25 Table 2 Sample Selection Panel A: Full Sample Insurer-state-years in Schedule T 575,484 Insurer-state-years in Schedule D 36,909 Insurer-state-years where both DirectPremiumsWritten is not 0 and ActualCosts are > Insurer-state-years where DirectPremiumsWritten = 0 and ActualCosts are > Insurer-state-years where both DirectPremiumsWritten is not 0 and ActualCosts are = Insurer-state-years where DirectPremiumsWritten = 0 and ActualCosts is missing Insurer-state-years where both DirectPremiumsWritten and ActualCosts are missing 0 Panel B: Insurers per year Schedule T Schedule D ,821 1, ,821 1, ,821 1, ,821 1,288 Panel C: Insurer Individual-State-Level Investments (Schedule D) Descriptions Match With Descriptions US State ,985 17, ,011 19, ,919 20, ,595 22,912 Total 86,510 80,549 25

26 Table 2 cont. Sample Selection Panel D: Insurer-state-years (Schedule D and Schedule T) Schedule D Schedule T , , , , , , , ,871 Total 36, ,484 Panel E: Insurer-state-years (Schedule D and Schedule T) After D and T Merge All merged no missing values Change analysis ,901 21, ,955 21,767 19, ,935 23,119 20, ,966 24,516 21,508 Total 102,757 90,495 60,651 26

27 Table 3 State Descriptive Statistics Direct Premiums Abbreviation ActualCost FairValue Written (in 000 s) SameState Domicile debt_outs total rev total_ expenditure population AK 980, , ,655,190 10,465,061 8,191, ,691 (6,012,389) (232.83) (12,533.05) (0.04) (1,009,257) (1,578,917) (804,303) (8,004) AL 476, , ,259,014 23,763,376 20,837,419 4,603,775 (2,037,013) (261.39) (46,401.01) (0.07) (2,401,207) (2,234,019) (2,048,552) (54,770) AR 547, , ,848,722 16,032,997 13,487,385 2,802,546 (2,837,299) (51.82) (25,826.12) (0.08) (1,100,671) (1,547,416) (1,119,570) (37,847) AZ 359, , ,933,692 27,880,978 24,312,143 5,933,732 (1,816,172) (63.02) (39,185.28) (0.02) (3,253,614) (3,250,671) (4,161,336) (194,321) CA 6,994, , ,617, ,971, ,518,727 35,935,040 (28,088,228) (445.22) (253,612.23) (0.21) (34,801,419) (25,247,722) (26,087,254) (249,848) CO 338,824 (1.08) 13, ,041,720 24,058,249 19,137,018 4,688,394 (1,526,649) (224.69) (46,654.81) (0.06) (2,988,854) (1,776,924) (1,865,178) (87,483) CT 4,658,906 (12.52) 10, ,034,364 22,223,222 19,564,718 3,512,621 (14,853,339) (386.61) (23,719.36) (0.22) (6,676,820) (2,633,899) (2,176,896) (11,653) DE 906, , ,076,121 6,579,649 6,106, ,515 (3,349,799) (13.32) (16,298.61) (0.08) (1,157,722) (665,282) (600,453) (15,357) FL 5,463, , ,124,711 84,729,727 68,300,046 17,969,391 (21,496,397) (201.14) (150,808.69) (0.16) (12,090,895) (9,285,707) (8,606,616) (357,832) GA 5,387, , ,114,131 38,545,059 36,245,709 9,061,841 (19,103,512) (290.11) (66,547.89) (0.11) (3,517,570) (4,308,148) (3,751,465) (222,340) HI 3,226, , ,709,205 9,684,310 8,697,103 1,298,779 (10,269,653) (262.91) (13,319.37) (0.04) (2,274,518) (1,037,864) (828,381) (16,309) IA 88,743 (5.59) 8, ,378,724 16,791,088 14,312,153 2,976,144 (497,685) (151.78) (25,340.96) (0.08) (1,544,064) (1,488,330) (958,540) (17,440) ID 28, , ,400,393 7,844,919 6,198,781 1,451,074 (390,148) 0.00 (11,146.00) (0.05) (349,756) (808,446) (543,389) (42,602) 27

28 Table 3 cont. State Descriptive Statistics Direct Premiums Abbreviation ActualCost FairValue Written (in 000 s) SameState Domicile debt_outs total rev total_ expenditure population IL 5,194,720 (41.41) 23, ,934,081 63,265,491 54,769,765 12,636,731 (15,996,398) (1,915.63) (99,428.43) (0.23) (13,205,230) (5,041,073) (3,847,939) (40,434) IN 213, , ,696,912 30,024,144 26,261,164 6,308,814 (1,152,316) (487.69) (41,464.72) (0.06) (4,805,110) (2,871,192) (1,962,539) (55,172) KS 133,314 (1.86) 8, ,474,396 13,287,403 12,229,615 2,757,863 (676,752) (70.58) (30,159.04) 0.00 (2,067,359) (1,678,698) (1,123,995) (18,897) KY 106,666 (0.14) 8, ,713,986 22,682,826 21,270,365 4,203,598 (814,019) (5.06) (27,683.55) 0.00 (2,684,793) (2,007,723) (1,993,739) (41,280) LA 2,737,573 (11.86) 13, ,479,746 27,539,620 23,160,181 4,447,951 (12,385,912) (311.30) (56,696.69) (0.20) (3,711,321) (3,858,738) (3,336,052) (115,338) MA 8,687,840 (0.54) 14, ,621,057 44,863,102 38,222,533 6,414,898 (30,013,041) (307.28) (51,457.80) (0.15) (16,244,219) (3,110,856) (4,340,787) (10,811) MD 2,938,782 (16.51) 13, ,267,744 30,986,572 26,683,945 5,607,482 (10,850,852) (547.28) (50,411.92) (0.13) (4,842,025) (2,739,818) (4,671,303) (39,807) ME 308,819 (0.02) 3, ,450,921 8,789,559 7,480,911 1,321,020 (1,259,538) (15.55) (12,299.78) (0.05) (1,011,239) (465,449) (431,935) (5,038) MI 1,615,915 (2.85) 26, ,873,946 57,713,128 50,999,111 10,034,787 (5,499,481) (135.51) (103,064.60) (0.18) (5,576,525) (2,662,548) (3,466,526) (21,645) MN 2,481,834 (4.73) 13, ,470,590 33,882,932 28,924,877 5,146,772 (8,525,683) (317.44) (50,520.34) (0.10) (2,632,047) (3,433,437) (2,942,667) (45,312) MO 643, , ,498,005 28,768,964 23,202,591 5,820,068 (2,262,905) (76.61) (55,378.43) (0.10) (3,870,307) (2,583,118) (1,696,743) (52,858) MS 2,336, , ,026,097 17,867,074 15,608,696 2,907,579 (8,839,457) (30.34) (22,873.26) (0.08) (1,893,405) (2,930,477) (2,191,920) (14,094) MT 109, , ,924,201 6,108,001 4,974, ,607 (707,634) (1.79) (9,569.99) (0.05) (886,555) (680,256) (458,932) (13,103) 28

29 Table 3 cont. State Descriptive Statistics Direct Premiums Abbreviation ActualCost FairValue Written (in 000 s) SameState Domicile debt_outs total rev total_ expenditure population NC 3,815,406 (0.08) 16, ,951,967 46,546,209 37,373,125 8,835,041 (12,508,810) (191.92) (57,897.41) (0.14) (5,032,388) (2,706,780) (6,683,025) (214,874) ND 74,866 (2.91) 2, ,533,671 4,558,437 3,457, ,351 (786,673) (107.59) (8,998.27) (0.05) (385,848) (489,159) (310,444) (3,162) NE 53, , ,898,870 9,017,324 7,309,822 1,767,433 (431,110) (79.94) (20,896.12) (0.03) (327,896) (547,332) (508,675) (12,690) NH 587,325 (9.99) 4, ,583,817 6,390,198 5,839,263 1,302,991 (3,400,447) (367.08) (10,205.96) (0.04) (1,435,930) (527,679) (297,522) (8,677) NJ 2,570,148 (5.72) 21, ,908,436 57,160,985 51,553,151 8,657,477 (8,188,653) (197.42) (61,732.72) (0.11) (15,404,350) (5,875,702) (5,240,160) (15,746) NM 285,338 (0.40) 4, ,542,863 14,352,050 12,698,062 1,948,752 (1,555,828) (53.47) (13,491.78) 0.00 (1,860,203) (1,961,488) (1,761,616) (32,350) NV 2,367, , ,371,602 12,144,980 9,427,048 2,481,055 (9,483,020) (395.18) (19,564.38) (0.06) (1,314,240) (1,448,215) (1,286,638) (95,340) NY 1,833, , ,639, ,622, ,272,590 19,134,278 (7,358,494) (284.47) (117,019.17) (0.20) (34,046,808) (15,829,192) (18,691,560) (23,373) OH 3,767,314 (4.64) 17, ,808,323 78,253,700 61,159,161 11,475,413 (12,223,808) (365.86) (67,867.80) (0.23) (7,201,072) (5,753,154) (5,203,401) (18,372) OK 283, , ,565,452 19,452,821 16,241,915 3,578,165 (1,113,983) (567.58) (32,130.42) (0.13) (2,860,242) (1,845,326) (1,441,509) (42,156) OR 1,459, , ,139,986 26,024,521 19,169,211 3,647,113 (6,592,687) (47.64) (29,088.95) (0.06) (3,197,923) (3,044,915) (1,387,469) (58,103) PA 4,732, , ,065,879 74,175,758 61,824,191 12,487,843 (18,550,764) (7,963.65) (99,607.26) (0.23) (8,759,995) (5,716,072) (6,343,819) (58,742) RI 865,405 (0.26) 3, ,655,989 7,651,137 6,447,153 1,065,341 (5,021,402) (9.70) (8,986.38) 0.00 (1,533,983) (477,543) (650,097) (6,339) 29