Capital Market Development, Competition, Property Rights, and the Value of Insurer Product-Line Diversification: A Cross-Country Analysis

Transcription

1 Capital Market Development, Competition, Property Rights, and the Value of Insurer Product-Line Diversification: A Cross-Country Analysis Thomas R. Berry-Stölzle a Robert E. Hoyt b Sabine Wende c February 2010 a Terry College of Business, University of Georgia, 206 Brooks Hall, Athens, GA 30602, Tel.: , Fax: , trbs@terry.uga.edu b Terry College of Business, University of Georgia, 206 Brooks Hall, Athens, GA 30602, Tel.: , Fax: , rhoyt@terry.uga.edu c Department of Risk Management and Insurance, University of Cologne, Albertus-Magnus- Platz, Cologne, Germany, Tel.: , Fax: , sabine.wende@uni-koeln.de 1

2 Capital Market Development, Competition, Property Rights, and the Value of Insurer Product-Line Diversification: A Cross-Country Analysis ABSTRACT In this paper, we show that the effect of diversification on performance is not homogeneous across countries. Diversified insurance companies perform significantly worse than their focused competitors in countries with well developed capital markets, high levels of property rights protection, and high levels of competition. In addition, we find that the diversification-performance relationship for insurance companies depends on company size. For large insurers operating in countries with less developed capital markets, diversification significantly increases performance. Our results suggest that the optimal organizational structure may be different for insurers operating in emerging economies than for insurers operating in developed countries. (JEL G34, G32, G22) Key words: Diversification, Focus, Organizational Structure, Internal Capital Markets, Firm Performance, International Insurance Markets 1

3 Introduction Theory provides conflicting arguments about the impact of corporate diversification on performance. On the one hand, the creation of an internal capital market without information asymmetries (Williamson, 1975; Stein, 1997), economies of scope (Teece, 1980), and risk reduction (see, e.g., Cummins, Phillips, and Smith, 2001; Cummins and Trainar, 2009, p. 467 ff.) should be beneficial for corporations. On the other hand, diversification may increase agency costs (Harris, Kriebel, and Raviv, 1982; Aron, 1988; Rotemberg and Saloner, 1994) and lead to inefficient allocation of capital among divisions of a diversified firm (Stulz, 1990; Rajan, Servaes, and Zingales, 2000). Thus, the net effect of diversification is an empirical question, and there is a large body of literature examining the relative performance of diversifiers versus specialized firms (see, e.g., Lang and Stulz, 1994; Comment and Jarrell, 1995; Berger and Ofek, 1995; Servaes, 1996; Cummins, Weiss, and Zi, 2003; Laeven and Levine, 2007; Liebenberg and Sommer, 2008; Schmid and Walter, 2009). While most researchers focus on measuring the average effect of diversification on the performance of all firms in their sample, few studies directly address the fundamental question of how the diversification-performance linkage varies across countries. The transaction cost theory proposed by Coase (1937) and Williamson (1985), however, suggests that the optimal structure of a firm depends on its institutional context. Hence, the main argument in our paper is that diversification could systematically increase firm performance in some countries but damage it in others. The goal of our research is twofold. First, we examine the diversification-performance relationship of insurance companies across a broad range of economies including those in the developing markets. Second, we explicitly examine country specific factors and how these factors moderate the diversification-performance linkage. 2

4 The factors we hypothesize to mediate the diversification-performance relationship are a country s degree of capital market development, the level of competition in the country s product markets and the level of property rights protection in the country. We expect that internal capital markets and, hence, diversification are most valuable in countries where it is expensive to raise external capital. The structure-conduct-performance paradigm suggests that effective collusion between firms increases with industry concentration, and collusion positively impacts firm performance (Stigler, 1964). Thus, in concentrated markets where the costs of diversification can be offset by joint monopoly rents, diversification should be more valuable than in markets with strong price competition. La Porta et al. (1997) document that countries with poorer investor protection have smaller capital markets; thus, internal capital markets should have a relatively high value in such countries. Furthermore, if business contracts cannot be predictably enforced outsourcing may not be a viable alternative to diversification. We examine these hypotheses by regressing firm performance on a diversification measure interacted with variables capturing different levels of capital market development, competition, and property rights protection across countries. Using data on 2,164 insurance companies across 76 countries over the period 2004 through 2007, we find that diversified insurance companies perform significantly worse than their focused competitors in countries with well developed capital markets, high levels of property rights protection, and high levels of competition. In addition, we find that the diversification-performance relationship for insurance companies depends on company size. When controlling for firm size, the result that diversified insurance companies perform significantly worse than their focused competitors in countries with well developed capital markets still holds. In addition, for the sub-sample of large insurers we find a positive relationship between diversification and performance in countries with less developed capital markets. 3

5 We are only aware of one paper that has a similar research focus to ours. Fauver, Houston, and Naranjo (2003) examine the value of diversification for non-financial firms across 35 countries for the 1991 through 1995 period. They find that the value of diversification depends on the level of capital market development, the level of capital market integration, and the type of legal system in a country. However, while there were still substantial differences in the level of capital market integration across countries in the early 1990s, virtually all countries have opened their capital markets to foreign investors by the start of our sample period, 2004 through Furthermore, the type of legal system, whether it is of English, German, Scandinavian, or French origin, may explain differences across countries if all countries in the sample have established institutions to actually enforce the law. However, when looking at a broader set of countries, including some in which the enforcement of contracts is unpredictable, the type of legal system does not have explanatory power. Our research extends the existing literature in multiple ways. First, we consider a broader set of countries than has been used in prior studies; we assess the impact of diversification across over 70 countries excluding the U.S. As a result, we consider the value of diversification in a range of economies, not just developed or emerging markets. Second, and importantly for emerging markets, we directly investigate the impact of property rights protection and competition on the value of diversification. We propose that the stability of the government and law enforcement as well as the competitive environment in a country moderate the diversification-performance linkage. Third, prior cross-country studies of the value of diversification have excluded insurers. As a result, our study offers the first comprehensive evidence of the value of corporate diversification for insurers operating in countries ranging from developed to emerging markets. This gap in the literature is all the more important given Santalo and Becerra s (2008) result that the effect of diversification on firm performance is not homogeneous across industries. 4

6 This paper proceeds as follows. In the next section we describe the prior literature on the value of diversification. We then develop testable hypotheses about the mediating effect of capital market development, competition and the level of property rights protection on the diversification-performance relationship. This is followed by a description of the data and methodology used to empirically test our hypotheses, and a section containing our results. The final section concludes. Prior Literature A large body of literature has studied the relationship between diversification and firm value. Lang and Stulz (1994), and Berger and Ofek (1995), for example, find that diversifiers trade at a discount compared to specialized firms. This so-called diversification discount has been interpreted as evidence that diversification destroys value (Martin and Sayrak, 2003). Theoretical explanations of the diversification discount include inefficient allocation of capital among divisions of a diversified firm (Stulz, 1990; Rajan, Servaes, and Zingales, 2000) and increased agency costs (Harris, Kriebel, and Raviv, 1982; Aron, 1988; Rotemberg and Saloner, 1994). However, empirical evidence on the net effect of diversification is far from conclusive. For instance, Villalonga (2004) shows that the diversification discount found in previous studies may be an artifact of segment-level data reported by Compustat. Using data from the U.S. Census Bureau, she is able to construct more precise diversification measures and finds a diversification premium. Schoar (2002) examines plant-level data from the Longitudinal Research Database and reports that conglomerates are more productive than stand-alone firms. In contrast, Maksimovic and Phillips (2002) in their analysis of plant-level data find that conglomerate firms are less productive than single-segment firms of similar size. 5

7 International evidence on the effect of diversification also paints a fuzzy picture. Lins and Servaes (1999) analyze a sample of firms from Germany, Japan, and the U.K. in 1992 and In Japan and the U.K., they estimate diversification discounts that are of similar magnitude to those reported for U.S. firms. For the German firms, they report a discount, however, this discount is not statistically significant. Lins and Servaes (2002) examine the value of diversification for seven Asian countries, namely Hong Kong, India, Indonesia, Malaysia, Singapore, South Korea, and Thailand, in the year Pooling data across all seven countries, they find a significant diversification discount. Khanna and Palepu (2000) on the other hand report that affiliates of the most diversified Indian business groups outperform unaffiliated firms in the year Fauver, Houston, and Naranjo (2003) add to the discussion by pointing out that the value of diversification may depend on the level of capital market development in a country. Santalo and Becerra (2008) argue that the effect of diversification on performance varies across industries. Using Compustat segment data for the period 1993 through 2001, they find that diversified firms perform better in industries in which specialized firms have a small combined market share, and worse in industries in which specialized firms have a large market share. Their findings suggest that the diversification-performance linkage should be examined for each industry separately. Insurance industry specific studies on the effect of corporate diversification on performance are rather scarce. Hoyt and Trieschmann (1991) analyze market risk-return profiles for a sample of publicly traded insurance companies over the 1973 through 1987 period. They report that insurers specialized in either the property-liability or the life-health insurance market segments perform better than insurers serving both market segments. Tombs and Hoyt (1994) use a Herfindahl index of premiums written in different lines of business as a diversification measure, and they find a negative and significant relationship between diversification and insurer stock 6

8 returns. Their sample include 26 publicly traded insurance companies operating in both the property-liability as well as the life-health insurance market segments. Berger et al. (2000) examine relative cost, revenue, and profit scope economies of diversified and specialized insurance companies for the years 1988 through Their results indicate that cost and revenue scope economies differ between diversifiers and specialized insurers; however, they find no significant difference in profit scope economies. Thus, they conclude that both a conglomerate strategy as well as a focus strategy are viable in the long run. Cummins, Weiss, and Zi (2003) use a more recent sample, 1993 through 1997, and data envelopment analysis to measure scope economies. They find only weak support for economies of scope and conclude that strategic focus seems to be a more efficient strategy than conglomeration. Meador, Ryan, and Schellhorn (2000) focus exclusively on life-health insurance companies. Using a sample of 321 insurers over the 1990 through 1995 period, they calculate cost efficiency scores and regress these scores on an Herfindahl index-based measure of diversification. Their findings suggest that diversified life insurers are more cost efficient than their more focused counterparts. In a complementary study, Liebenberg and Sommer (2008) analyze the propertyliability insurance industry. Based on a sample of 914 insurers for the years 1995 through 2004 they find that undiversified insurance companies consistently outperform diversified insurance companies. Elango, Ma, and Pope (2008) examine 1074 property-liability insurers for the 1994 through 2002 period and report that the effect of business line diversification on firm performance depends on an insurer s degree of geographic diversification. Their finding of a more complex relationship between diversification and performance serves as a motivator for additional analysis on the topic. Our study contributes to the literature on the effects of corporate diversification by examining the diversification-performance relationship of insurance companies across a broad range of 7

9 economies including those in the developing markets. In addition, we explicitly examine country specific factors and how these factors moderate the diversification-performance linkage. Conceptual Background and Hypotheses Development The relationship between corporate diversification and performance is an important topic in the finance as well as in the insurance literature (see, e.g., Berger and Ofek, 1995; Liebenberg and Sommer, 2008; Schmid and Walter, 2009). While diversification may increase agency costs, the theoretical work of Williamson (1975) and Stein (1997) suggest that diversification may also have benefits; diversified firms can establish internal capital markets to effectively allocate capital within the firm. In Stein s (1997) model, corporate headquarters can create value through winner-picking and actively re-allocating capital to projects or business units with the highest expected rate of return. For non-financial firms, the empirical evidence on the efficient functioning of such internal capital markets is mixed. 1 For financial intermediaries, however, recent empirical evidence documents active and efficient internal capital markets (see, e.g., Houston and James, 1998; Powell, Sommer, and Eckles, 2008). The differing results between studies of financial intermediaries and of firms in other industries are consistent with the prediction of Stein s (1997) model: The benefits for establishing an internal capital market are highest for firms which are opaque and hard to evaluate by outside investors, and whose business units operate in related fields so that the headquarters has the expertise to pick the winners. Financial intermediaries tend to have both of these characteristics. Supporting the argument that an internal capital market without information asymmetries gives opaque business units easier access to capital, Campello 1 The results of Shin and Stulz (1998) and Rajan, Aervaes, and Zingales (2000), for example, suggest that internal capital re-allocations within conglomerates do not depend on investment opportunities. In contrast, Khanna and Tice (2001) find evidence of efficiently functioning internal capital markets in the discount retail industry. 8

10 (2002) finds evidence that internal capital markets reduce the effect of external financial constraints on investments of financial intermediaries. Obviously, the value of internal capital markets depends on the availability and cost of external capital. In an economy with a well developed capital market where smaller stand-alone firms can relatively easily raise external capital, internal capital markets may have limited value. On the other hand, in an economy where it is more costly to raise external capital, internal capital markets may be more valuable. Since a firm s access to external capital depends on the level of capital market development within the country where the firm operates, we expect that internal capital markets and, hence, diversification is most valuable in countries with relatively low levels of capital market development. In summary, we can state the following testable hypothesis for financial intermediaries: Hypothesis 1. The level of capital market development moderates the relationship between diversification and firm performance; diversified firms perform worse in countries where the level of capital market development is relatively high compared to countries where the level of capital market development is relatively low. The structure-conduct-performance paradigm explains the relationship between market structure and firm performance via the conduct of the firms in the market (Mason, 1939; Bain, 1951). Market concentration facilitates effective collusion between the firms in the market because the cost of collusion is lower in highly concentrated markets. Prices that are less favorable to consumers positively impact firm performance (Stigler, 1964). Recent empirical evidence from the U.S. insurance market does not support the structure-conduct-performance hypothesis (Choi and Weiss, 2005; Weiss and Choi, 2008). However, in their study of international insurance markets, Pope and Ma (2008) find support for the structure-conduct-performance hypothesis 9

11 in countries with low levels of market liberalization. Most of the countries in our sample fall in this category since they still have relatively high barriers to entry for foreign insurers. Therefore, we expect that in highly concentrated markets where the costs of diversification can be offset by joint monopoly rents, diversification should be more valuable than in markets with strong price competition. Furthermore, Berry-Stölzle, Hoyt, and Wende (2010) report that for insurance companies operating in emerging markets with low insurance penetration, a fast growth strategy significantly improves performance. Since broadly diversified insurance companies are in a good position to capture growth opportunities in whichever business line they arise, diversification should be more valuable in economies with low insurance penetration compared to saturated markets where growth is only possible at the expense of other firms market share. Combining these two arguments, we state the following hypothesis: Hypothesis 2. The level of competition in a country s product markets moderates the relationship between diversification and firm performance; diversified firms perform worse in countries with relatively strong competition compared to countries where the level of competition is relatively low. La Porta et al. (1997) document a link between a country s legal system and its level of capital market development. Their cross-sectional analysis of 49 countries reveals that countries with poorer investor protection have smaller capital markets. This finding suggests that the availability of external capital should be lower and, hence, the cost of raising external capital should be higher in countries with relatively low levels of investor rights protection. Therefore, internal capital markets should have a relatively high value in such countries, making a diversification strategy relatively advantageous for a firm. 10

12 In addition, if business contracts cannot be predictably enforced in a country outsourcing may not be a viable alternative to diversification. Firms may diversify and create internal resources and institutions to mitigate external market failures (Khanna and Palepu, 2000). Overall, we expect that the value of corporate diversification depends on the level of property rights protection in the following way: Hypothesis 3. The level of property rights protection moderates the relationship between diversification and firm performance; diversified firms perform worse in countries where the level of property rights protection is relatively high compared to countries where the level of property rights protection is relatively low. Data and Methodology Sample Selection Unlike the previous studies of product-line diversification in the insurance industry, we consider the questions of value in a cross-country analysis. We use company-level data from A.M. Best s Statement File Global for the years 2004 through Our initial sample consists of all listed insurance companies operating in countries other than the U.S. First, we exclude all companies classified as reinsurers or pure holding companies. Second, we exclude companies that report negative direct premiums written, premiums earned, total assets, policyholder surplus or investment positions. Our third screen is to exclude companies with missing data on basic accounting variables, including total assets, policyholder surplus, profit before taxes, life insurance premium earned and non-life insurance premiums earned. We then aggregate affiliated insurers operating in one country, controlling for potential double counting of intra-group shareholding. Such an aggregation is appropriate since diversification decisions are usually made on the group level and not by individual subsidiaries (Berger et al., 2000). Since we use historical 11

13 risk measures which require five years of data, we exclude firm-year observations for which the preceding five years of data are not available. Finally, we exclude extreme outliers. Our first outlier screen is to eliminate firm-year observations with reported life (non-life) insurance premiums in excess of the overall premium volume of the corresponding country s life (non-life) insurance market. 2 Next, we eliminate observations if the return on equity (ROE) has a value above one or below minus one (Berger and Ofek, 1995). Finally, we follow Laeven and Levine (2007) and eliminate observations where the variables are more than four standard deviations from the sample mean. Our final sample consists of 6,353 firm-year observations with a maximum of 1,699 unique insurers in The sample includes insurers from 76 different countries over the period 2004 through Performance Regression Model We analyze the relationship between diversification and insurer performance in two steps. First, we measure the net effect of diversification within the context of a standard performance regression specification (see, e.g., Lai and Limpaphayom, 2003; Liebenberg and Sommer, 2008). In a second step, we then add interaction terms between country characteristics and our diversifi- 2 Country aggregates of life insurance premiums and non-life insurance premiums are obtained from Swiss Re s sigma publications. 3 The 76 countries included in our sample are: Antigua & Barbuda, Argentina, Australia, Austria, Bahamas, Bahrain, Barbados, Belgium, Bolivia, Brazil, Bulgaria, Canada, Chile, Croatia, Cyprus, Czech Republic, Denmark, Dominican Republic, Ecuador, Egypt, El Salvador, Estonia, Finland, France, Germany, Ghana, Greece, Hong Kong, Hungary, Iceland, India, Indonesia, Ireland, Israel, Italy, Jamaica, Japan, Jordan, Kazakhstan, Kuwait, Latvia, Lithuania, Luxembourg, Malaysia, Mexico, Morocco, Netherlands, New Zealand, Nigeria, Norway, Oman, Pakistan, Panama, Peru, Philippines, Poland, Portugal, Qatar, Romania, Russia, Saudi Arabia, Singapore, Slovakia, Slovenia, South Africa, Spain, Sweden, Switzerland, Taiwan, Tanzania, Thailand, Trinidad & Tobago, Turkey, United Kingdom, Uruguay, and Venezuela. 12

14 cation variable to the baseline regression model to measure the moderating effects of these country characteristics. There are two commonly used measures of accounting performance in the insurance literature: Return on assets (ROA) and return on equity (ROE). While studies focusing exclusively on life insurers or exclusively on property-liability insurers may employ both measures (see, e.g., Browne, Carson, and Hoyt, 2001; Greene and Segal, 2004; Liebenberg and Sommer, 2008), we cannot use ROA in our joint analysis of pure life insurers, pure non-life insurers and insurers operating in both market-segments because the asset structure of life and non-life insurers and, hence, their ROA are not comparable. Thus, we use ROE to measure insurer performance. We calculate ROE for each insurer as net income before taxes divided by equity capital as shown on the balance sheet. Since higher performance can simply be driven by higher risk, we follow the approach of Hamilton and Shergill (1993), Lai and Limpaphayom (2003) and Liebenberg and Sommer (2008) and include a risk measure as a control variable in our performance regression model. The main advantage of such a regression specification is that it allows us to directly assess the magnitude of the effect of diversification on insurer performance while controlling for risk. Specifically, we include the SDROE5 variable which measures the standard deviation of an insurer s ROE over the past five years. The first part of our regression analysis focuses on whether diversification is performance enhancing or reducing in our international sample of insurance companies. To construct measures of diversification for an insurance company, we would ideally like detailed data on each business line the company writes. Data availability, however, restricts our ability to measure the diversification across lines of business. Therefore, we adopt the classification of Hoyt and Trieschmann (1991), Berger et al. (2000) and Cummins, Weiss, and Zi (2003) and view insurance 13

15 companies that operate in either the life or the non-life market segments as specialists and companies that are joint producers as diversified. We follow Laeven and Levine (2007) and use a continuous measure to describe the degree of diversification across the life and non-life insurance market-segments. The Diversification variable is calculated as life premiums earned non life premiums earned 1. (1) total premiums earned This premium diversity measure takes values between zero and one with higher values indicating greater diversification. To disentangle the effect of diversification from possible performance differences between the life insurance business and the non-life insurance business, we also include a measure of insurance activity in our regression models. More specifically, we use the ratio of life insurance premiums earned to total premiums earned to measure each insurer s position along the con tinuum from pure life insurance to pure non-life insurance underwriting. Our analysis of the diversification-performance relationship is based on a pooled crosssectional time series regression model. The model is estimated using ordinary least squares (OLS) as well as the fixed-effects estimator (FE). The specification of the model is as follows: ROE = α + β Diversification + β Insurance Activity + β X + ε (2) it, 1 it, 2 it, 3 it, it, where ROE i, t is the return on equity for insurer i in year t, Diversific ation i, t is the premium diversity measure defined in Equation (1), Insurance Activity i, t denotes the ratio of life insurance premiums to total premiums, X is a vector of control variables, and ε i, t is a random error term. 4 4 Note that our analysis focuses on the effect of product-line diversification on performance. There are studies with a focus on geographic diversification across countries that report a nonlinear relationship between diversification and performance (see, e.g., Outreville, 2008). However, we are not aware of any study reporting a quadratic relationship between product-line diversification and performance. Furthermore, when adding Diversification-squared to the 14

16 The vector of control variables includes the SDROE5 variable described above to capture differences in risk across the insurance companies in the sample, as well as other insurer-level and country-level characteristics. We also include a measure of insurer size in the model. All else equal, bigger risk pools should produce less volatile claim payments, lower insolvency risk (see, e.g., Cummins, Harrington, and Klein, 1995; Grace and Harrington, 1998) and should, hence, be able to charge higher prices than smaller risk pools (Sommer, 1996). Consistent with this expectation, a number of empirical studies have documented a positive relationship between insurer size and performance (see, e.g., Browne, Carson, and Hoyt, 1999; Cummins and Nini, 2002). Thus, we include the natural log of total assets as the size variable in our model. Sommer (1996) finds a positive relationship between insurers capitalization and the prices they are able to command. To control for differences in capitalization, we include the surplus to assets ratio in our model. Past performance is often used to proxy growth opportunities (Laeven and Levine, 2007). To control for differences in past performance across insurers, we include the growth rate in total assets over the past three years and the growth rate in profit before taxes over the past three years in our model. Cummins, Weiss, and Zi (1999) compare the cost efficiency of stock and mutual insurers and find support for the expense preference hypothesis, which predicts that mutuals are less cost efficient than stocks because mutuals lack a powerful control mechanisms for managerial perquisite consumption. To capture the difference in performance between mutual and stock insurers, our model includes a dummy variable equal to 1 for mutual insurance companies and 0 for all others. We include dummy variables capturing the number of years an insurance company is observed in the dataset to control for any survivor effect, e.g., long-lived insurance companies may be more profitable (Stiroh and Rumble, 2006). model, the estimated coefficient of the squared term is insignificant. Therefore, we do not consider higher order terms of the diversification measure in our analysis. 15

17 Since we are pooling data across countries and over time, our model includes country and year dummies. We also control for time-varying country characteristics. More specifically, we include the annual growth rate in real Gross Domestic Product (GDP) per capita in constant 2000 US$ in the model to control for overall economic conditions; and we include the annual inflation rate in the model since inflation can affect insurer performance, especially in long-tail business lines. Inflation and GDP data are obtained from the World Bank s World Development Indicators database. It is inappropriate to assume that insurer observations are independent over time. Hence, standard errors in all OLS regressions are adjusted for clustering at the company-level. To control for any possible omitted firm-specific effects, we also estimate Equation (2) with the FE estimator and present both, the OLS and the FE results. Note, that the Mutual variable was dropped from the FE regressions due to a within standard deviation close to zero. Country-Level Factors Moderating the Diversification-Performance Relationship The second part of our analysis focuses on whether the relationship between insurers diversification and their performance depends on the characteristics of the country in which they operate. Following a methodology similar to that used in Arena (2008), the Diversific ation i, t variable and the Insurance Activity i, t variable in Equation (2) are interacted with dummy variables capturing different levels of capital market development, competition, and property rights protection across countries (below the 33 rd percentile value of the distribution across our sample in a given year, between the 33 rd and 66 th value of the distribution, and above the 66 th percentile value of the distribution). The specification of this extended model is as follows: 16

18 ROE = α + β Diversification Moderator + β Diversification Moderator low third middle third it, 1 it, ct, 2 it, ct, + β Diversification Moderator + β Insurance Activity Moderator top third low third 3 it, ct, 4 it, ct, middle third top third + β5insurance Activit it, ct, 6 it, ct, + β X + ε 7 it, it, y Moderator + β Insurance Activity Moderator (3) where ROE i, t is the return on equity for insurer i in year t, Diversific ation i, t is the premium diversity measure defined in Equation (1), Insurance Activity i, t denotes the ratio of life insurance low third premiums to total premiums, Moderator ct, is a dummy variable coded as 1 if country c is below the 33 rd middle third percentile value of the moderator variable in year t, Moderator ct, is a dummy variable coded as 1 if country c is between the 33 rd and 66 th percentile value of the mod- top third erator variable in year t, Moderator ct, is a dummy variable coded as 1 if country c is above the 66 th percentile value of the moderator variable in year t, X is the vector of control variables from Equation (2), and ε i, t is a random error term. Note that we estimate Equation (3) for each country-level moderator variable separately. The following section discusses our proxies to measure the level of capital market development, competition and property-rights protection. Some of these measures are not available for all 76 countries and all four years in our sample. Thus, the number of observations used varies slightly across the estimated models. Ideally, a measure of capital market development should capture the ease with which borrowers and savers can be brought together. Firms can raise external funding either through securities markets or through the banking system. Thus, a traditional measure of capital market development is the ratio of domestic credit plus stock market capitalization to GDP (see, e.g., Rajan and Zingales, 1998). Unfortunately, data on stock market capitalization is not available for a number of countries in our sample. In a study on the effect of capital market development on 17

19 economic growth, Rajan and Zingales (1998) use two measures of financial development: the ratio of domestic credit plus stock market capitalization to GDP, and the ratio of domestic credit to the private sector to GDP. In their sample of 41 countries these two variable are highly correlated. Therefore, we measure the level of capital market development in a country with the ratio of credit to the private sector by deposit money banks to GDP. The data are obtained from the World Bank s World Development Indicators database. Arguably, the availability and cost of external capital in a country also depends on a country s credit rating. Thus, our second measure of the level of capital market development is the country s credit rating. We use the average of the two ratings published semi-annually by Institutional Investor. The ratings are based on surveys of bankers and are on a scale from zero to 100, with higher values indicating a better rating. We also use two measures for the level of competition in a country s insurance industry. Stigler s (1964) model links market concentration to collusion and prices above competitive levels. Hence, we use the market share of the five largest insurance companies in a country as an inverse measure of competition. 5 We calculate this variable by dividing the total premium volume of the country s largest five insurance companies in our sample by the sum of the country s life and non-life industries premium volumes as reported in Swiss Re s Sigma publications for the corresponding year. Since competition is usually tougher in saturated markets than in new and growing markets, we use the insurance penetration in a country as an inverse measure of competition. Insurance Penetration is calculated as the sum of a country s total life and non-life insurance premium volume as a share of GDP. 5 Due to data limitations, we are unable to use the standard measure of market concentration: the Herfindahl index. The Herfindahl index only captures market concentration correctly when its calculation is based on all companies in the market. A.M. Best s Statement File Global, however, only contains the most important market players. A Herfindah index based concentration measure would, hence, lead to a biased view. 18

20 To measure the level of property rights protection in a country, we use two variables that capture whether a country s legal system is stable and whether people are playing according to the rules. Our proxies are a corruption index and political risk which also captures the stability of the government and law enforcement. Data for the Political Risk Index variable is from the PRS International Country Risk Guide Researchers dataset. This index is an assessment of government accountability and stability, quality of bureaucracy and law enforcement, investment climate, and various sources of political and social conflicts. 6 The index takes on values between zero and 100, with lower values representing unstable institutions and higher risk. Data for our second measure, the Corruption Index variable, is from Transparency International s website. Transparency International s corruption perception index is based on a number of polls and surveys and reflects the public s perception about public sector corruption. The index is on a scale from zero to 10, where lower values stand for less transparency and higher corruption. Summary Statistics Table 1 provides summary information of country characteristics, insurer performance and the percentage of diversified insurance company for each country in our sample. All monetary values are inflation adjusted and converted to constant 2000 US$. The country-level variables presented in the first six columns exhibit substantial variation for emerging markets and are, hence, well suited for our research design. The performance measure used in this study is the ROE. The ROE varies substantially across insurers in our sample; the median ROE ranges from % for insurers operating in Romania to 49.54% for insurers operating in Pakistan with a sample median of 13.62%. Overall, 6 For a more indepth description of the methodology underlying the political risk index: see 19

21 19.82% of the insurance companies in our sample write both life and non-life insurance business and are, hence, classified as diversified in our analysis. 7 [ Insert Table 1 Here ] Results Table 2 contains summary statistics for the three groups of life, non-life and diversified insurers, separately. We can see that diversified insurers have on average a higher ROE and a lower earning volatility as measured by the standard deviation of the previous five years ROE. Therefore, based on the univariate comparison, diversified insurers seem to perform better than specialized insurers. However, diversified insurers are also significantly larger than specialized insurers; the average diversified insurer is more than twice as large as the average life insurer, and about 18-times larger than the average non-life insurer in the sample. Since bigger risk pools produce less volatile claim payments, larger insurers should need relatively less capital to back their outstanding obligations than smaller insurers, resulting in a production advantage. Note that the surplus to assets ratio is significantly lower for diversified insurers than for life and non-life insurers. Thus, the performance advantage of diversified insurers in the univariate comparison tests could just be a size effect. Multivariate models with appropriate controls are necessary to assess the impact of diversification across the firms in our sample. [ Insert Table 2 Here ] 7 The classification as a diversified insurer is based on premiums earned. All insurance companies reporting positive premiums earned for life and non-life business are classified as diversified. 20

22 Pooled Performance Regressions The OLS estimation results from Equation (2) are presented in columns 1 and 2 of Table 3. The estimates are based on a pooled regression across all 76 countries in the sample. The two models presented only vary with respect to the control variables included. In both models the coefficient on the Diversification variable is negative and significant, indicating a negative relationship between corporate diversification and performance as measured by ROE. This result is consistent with the findings of most recent studies of the value of diversification for firms in the U.S. (see, e.g., Schmid and Walter, 2009). [ Insert Table 3 Here ] In addition, the coefficient on the Insurance Activity variable is negative and significant, suggesting that for our full sample of insurers internationally, the fraction of life insurance premiums an insurer writes is negatively related to performance. As expected, the coefficient of the SDROE5 variable is positive and significant, confirming a positive risk-return relationship. The coefficient on the ln(total Assets) variable is positive and significant, indicating a positive relationship between insurer size and performance. Capitalization as measured by the Surplus/Assets ratio is negatively associated with an insurers ROE. Consistent with Liebenberg and Sommer (2008), we find that the mutual organizational from is significantly negatively related to performance. There is a small but positive and significant relationship between an insurer s growth in profits before taxes over the past three years as measured by the Growth in Income variable and ROE. Furthermore, a country s Inflation rate positively affects insurer ROE. Columns 3 and 4 of Table 3 present the FE estimation results from Equation (2). By design, the FE estimator controls for any firm-specific effects. Therefore, time-invariant firm characteristics like the Mutual variable cannot be included in the FE model. Recall that our Diversification variable is a continuous measure defined over the range 0 to 1. The variable has a value of 21

23 0 if an insurer operates in either the life or the non-life insurance market segment, and the variable takes on its highest value of 1 if an insurer writes 50% of its business in life and 50% in nonlife. In both model variants presented in Table 3, the coefficient on the Diversification variable is negative and significant. This finding indicates that an increase in an insurer s degree of diversification is negatively associated with its performance as measured by the ROE. Overall, the FE results are consistent with the OLS results. However, this first part of the analysis does not allow us to control for the effect of differences in market environment across the range of countries from developed to emerging markets. In other words, we are unable to discern differences in the diversification relationship across different economies. Country-Level Factors Influencing the Diversification-Performance Relationship To examine the moderating role of country characteristics on the diversificationperformance linkage, we interact our diversification measure in the performance regression model with dummy variables capturing different levels of capital market development, competition, and property rights protection across countries. More precisely, for each of the six country-level measures of capital market development, competition, and property rights protection discussed above, we create three dummy variables capturing whether a country is below the 33 rd percentile value of the distribution across the sample in a given year, between the 33 rd and 66 th value of the distribution, and above the 66 th percentile value of the distribution. These dummies are then interacted with the Diversification and the Insurance Activity variables in Equation (2), leading to the model given by Equation (3). Table 4 shows the OLS and FE estimation results from Equation (3). The six models presented only vary with respect to the sets of country specific moderator dummies included. Columns 1 and 2 in Table 4 show the regression results using the Credit to Private Sector variable as 22

24 moderator. This variable measures the level of capital market development in a country. With respect to the OLS estimation, the results show that the interaction term with Diversification is negative and significant for the level of Credit to Private Sector above the 66 th percentile of its distribution, but insignificant for levels of Credit to Private Sector below this level. The FE estimates show that the interaction term with Diversification is negative and significant for the levels of Credit to Private Sector above the 66 th percentile of its distribution, and for the levels between the 33 rd and 66 th percentile values, but insignificant for the levels below the 33 rd percentile value. These findings indicate that first, insurer diversification is negatively associated with performance in countries with high levels of capital market development as measured by the Credit to Private Sector variable and second, insurer diversification has no significant impact on performance in countries with low level of capital market development. Overall, these results support Hypothesis 1 which states that the level of capital market development moderates the diversification-performance relationship. [ Insert Table 4 Here ] Columns 3 and 4 in Table 4 show the regression results for our second measure of the level of capital market development: the Country Credit Rating variable. In the OLS regression, the interaction term with Diversification is negative and significant for countries with a high Country Credit Ratings (above the 66 th percentile of the distribution), but insignificant for medium and low Country Credit Ratings. In the FE regression, the interaction terms with Diversification is negative and significant for countries with a high or a medium Country Credit Rating, but insignificant for low levels of the Country Credit Rating variable. Overall, these findings are consistent with the results for the Credit to Private Sector variable, and provide additional support for the moderating role of a country s capital market development on the diversification-performance linkage as stated in Hypothesis 1. 23

25 The OLS and FE regression results for the two measures of competition as moderator variables are presented in columns 5 through 8 of Table 4. In the OLS regression with the Market Share of 5 Largest Insurers as moderator, the interaction term with Diversification is negative and significant for countries with low and medium levels of the moderator variable, but insignificant for high levels of the moderator variable. Since the Market Share of 5 Largest Insurers is an inverse measure of competition, this result indicate that insurer diversification is negatively associated with performance in countries with high levels of competition, but not in countries with low levels of competition. In the FE regression, however, all three interaction terms with Diversification are negative and significant, indicating that diversification is negatively related to performance regardless of the level of competition as measure by the Market Share of 5 Largest Insurers variable. Using the Insurance Penetration variable as mediator in the OLS regression, we find that the interaction term with Diversification is negative and significant for countries with a high insurance penetration, but insignificant for medium and low levels of insurance penetration. In the FE regression, the interaction term is negative and significant for countries with a high and medium insurance penetration, but insignificant for countries with low insurance penetration. Except for the FE model with the Market Share of 5 Largest Insurers as moderator variable, these results provide support for Hypothesis 2. We find that diversification is negatively related to performance in countries with high levels of competition, but not in countries with low levels of competition. Columns 9 through 12 of Table 4 present the OLS and FE regression results using the two measures for the level of property rights protection in a country as mediator variables. The two sets of regressions for the Political Risk Index and for the Corruption Index variables as mediators provide qualitatively similar results. In the OLS regressions, the interaction term with Diversification is negative and significant for countries with high levels of property rights protection, 24

26 but insignificant for countries with medium and low levels of property rights protection. In the FE regressions, the interaction term is negative and significant for high and medium levels of property rights protection, but insignificant for low levels of property rights protection. These findings provide support for Hypothesis 3 which states that diversified insurers perform worse in countries where the level of property rights protection is relatively high compared to countries where the level of property rights protection is relatively low. Controlling for the Moderating Effect of Insurer Size The univariate differences in Table 2 provide some evidence that the size of an insurance company is positively related to its performance. To examine whether insurer size also moderates the diversification-performance relation in addition to its direct effect on performance, we estimate Equation (3) with moderator dummies capturing different size groups. Since our analysis focuses on country-level factors and their influence on the diversification-performance relationship, we also code the size group dummies relative to the insurer population in each country. In other words, an insurer coded as large is above the 66 th percentile of the sample distribution of its country, but not necessarily above the 66 th percentile of the overall sample. Table 5 shows the OLS and FE estimation results from Equation (3) with moderator dummies for size groups. In the OLS regression, the interaction term with Diversification is negative and significant for the group of medium size insurers, but not for small and large insurers. In the FE regression, the interaction term with Diversification is negative and significant for large and medium size insurers, but not for small insurers. Overall, these results indicate that the relation between diversification and performance is not the same for insurance companies of different sizes. This finding is interesting because previous studies on industrial firms have not found a size effect (see, e.g., Berger and Ofek, 1995). 25