études et Dossiers No. 369 World Risk and Insurance Economics Congress July 2010 Singapore Working Paper Series of The Geneva Association

Transcription

1 International Association for the Study of Insurance Economics études et Dossiers études et Dossiers No. 369 World Risk and Insurance Economics Congress July 2010 Singapore Working Paper Series of The Geneva Association The Geneva Association - General Secretariat - 53, route de Malagnou - CH-1208 Geneva Tel.: Fax: secretariat@genevaassociation.org -

2 International Association for the Study of Insurance Economics Études et Dossiers Études et Dossiers No. 369 World Risk and Insurance Economics Congress July 2010 Singapore February 2011 Working Paper Series of The Geneva Association This document is free to download from The Geneva Association website, Association Internationale pour l'etude de l'economie de l'assurance The Geneva Association - General Secretariat - 53, route de Malagnou - CH-1208 Geneva Tel.: Fax: secretariat@genevaassociation.org -

3 The Geneva Association Working Paper Series Études et Dossiers appear at irregular intervals about times per year. Distribution is limited. The Études et Dossiers are the working paper series of The Geneva Association. These documents present intermediary or final results of conference proceedings, special reports and research done by The Geneva Association. Where they contain work in progress or summaries of conference presentations, the material must not be cited without the express consent of the author in question. This document is free to download from The Geneva Association website, please visit: Layout & Distribution: Valéria Kozakova The Geneva Association - Association Internationale pour l'etude de l'economie de l'assurance

4 Cost Efficiency and Productivity Growth of Life Insurance Distribution Channels Jennifer L. Wang National Chengchi University Jin Lung Peng Shih Chien University Larry Y. Tzeng National Taiwan University Lih Ru Chen (contact person) Shih Chien University Abstract Using a unique panel data set of insurance agents and brokers in Taiwan, we analyze the cost efficiency (CE) and productivity growth of life insurance distribution channels by means of Data Envelopment Analysis (DEA) and Malmquist indices. We find that insurance agencies consistently outperform insurance brokers. The empirical evidence also confirms that diversified brokers are more cost efficient than specialized brokers. After decomposing CE, we find that the efficiencies of both insurance agencies and diversified brokers result from technical efficiency (TE). Robustness tests using subsamples, alternative measures of performance and regression analysis provide consistent results. We also find that size is significantly positively related to allocative efficiency (AE) but negatively related to TE. Furthermore, the evidence shows that both market share and leverage are significantly positively related to CE (for both AE and TE). Key words: Life Insurance; Distribution Channels; Product Diversification; Data Envelopment Analysis; Cost Efficiency.

5 1. Introduction In the insurance literature, cost efficiency has received increasing attention over the last two decades. Although many papers have proposed ingenious empirical methodologies or provided intriguing empirical findings 1, relatively few papers have investigated the cost efficiency of distribution channels employed by insurance firms due to there being no suitable data for insurance distribution channels. In this paper, we intend to fill this gap in the literature. To this end, we hand collect a unique data set of distribution channels employed by insurance firms in Taiwan over the period from 2004 to Our sample comprises 83 insurance agencies and 197 insurance brokers 2, which together account for about 74% of total insurance premiums in the life insurance agency and brokerage market. All 83 agencies sell only life insurance and do not sell property liability (P-L) insurance. On the other hand, among the 197 brokers, 80 brokers also have a license to sell P-L insurance. 3 In this paper, we first examine whether life agencies are more efficient than life brokers. In general, agencies perform more service functions than brokers, and thus insurance agencies are expected to generate more output than brokers. In addition, agencies may be more efficient than brokers due to different incentive conflicts between insurance firms and agencies and between insurance firms and brokers. Secondly, we analyze whether a diversified firm is more efficient than a specialized firm. Specifically, we test whether a broker selling both life insurance and P-L insurance 1 For instance, Cummins and Weiss (1993); Gardner and Grace (1993); Yuengert (1993); Fecher, Kessler, Perelman, and Pestieau (1993); Kim and Grace (1995); Grace (1995); Regan (1997); Cummins, Weiss, and Zi (1999); Berger, Cummins, Weiss, and Zi (2000); Meador, Ryan, and Shellhorn (2000); Cummins, Tennyson, and Weiss (1999); Cummins and Rubio-Misas (1998); Berger, Cummins, and Weiss (1997); Cummins and Zi (1998); Cummins and Rubio-Misas (2006); Weiss and Choi (2008); Cummins and Xie (2008, 2009); Eling and Luhnen (2009); Leverty and Grace (2009); Cummins, Dionne, Gagné, and Nouira (2009); Park, Lee, and Kang (2009); Luhnen (2009); Leverty, Lin, and Zhou (2009). 2 In this paper, an insurance agency is a distribution channel whose goal is to sell and service insurance products and is an agency company under common management. In this paper, an insurance agency refers to an agency company. By an insurance broker is meant an insurance brokerage company which searches the marketplace on behalf of its customers. 3 By law, insurance agencies may only choose to sell either P-L insurance or life insurance in Taiwan. Therefore, our agency samples only refer to specialized agencies that only sell life insurance. The insurance brokers can obtain licenses to sell both P-L insurance and life insurance or else they can choose to sell either P-L insurance or life insurance. 2

6 products is more efficient than a broker and/or an agency that sells only life insurance products. Studies on the relationship between efficiency and product diversification have yielded mixed results. 4 Product diversification may be efficiency enhancing due to an internal capital market, risk diversification, and economies of scope (e.g., Williamson, 1970; Lewellen, 1971; Teece, 1980; Myers and Majluf, 1984; Gertner, Scharfstein, and Stein, 1994). On the other hand, product diversification may lead to a deterioration in efficiency because of the existence of coordination problems or agency problems between managers and stockholders. The empirical evidence supports our first hypothesis that agencies are more cost efficient than brokers. This implies that, given the same level of inputs, agencies could produce more output than brokers. After decomposing the cost efficiency, we find that the source of the cost efficiency is technical efficiency rather than allocative efficiency. Furthermore, our empirical evidence also supports the second hypothesis that diversified brokers are more efficient than specialized brokers. This means that a broker who can sell both life insurance and P-L insurance can benefit in terms of cost efficiency through the use of fewer inputs given the same level of outputs. The analysis of decomposition also shows that the source of the cost efficiency is technical efficiency rather than allocative efficiency. In addition, the findings also suggest that efficiencies are greater for distribution channels that are large, use greater financial leverage, and/or are more profitable. The remainder of this paper is organized as follows. Section 2 reviews the prior literature. In Section 3, we provide the background to life distribution channels in Taiwan. Section 4 discusses and develops two hypotheses that we investigate in this article. Section 5 describes our methodology and data. Section 6 presents our empirical results, and Section 7 concludes. 2. Literature Review In recent years, the measurement of efficiency has received much research attention. So far, there have been more than 90 efficiency studies on the insurance industry. 5 A number of these papers have examined the relationship between diversification and the 4 For example, Panzar and Willig (1981), Lang and Stulz (1994), Berger and Ofek (1995), Comment and Jarrell (1995), Servaes (1996), Liebenberg and Sommer (2007), and Santalo and Becerra (2008). 5 For surveys on efficiency studies in the insurance industry, see Cummins and Weiss (2000) and Eling and Luhnen (in press). 3

7 CE performance of insurance firms. The results of these studies, however, provide mixed evidence regarding the cost effect of diversification across product lines within the life or P-L insurance segment. Using Canadian data, Kellner and Mathewson (1983) found evidence of cost economies of scope in the insurance industry. Toivanen (1997) also provided evidence in support of cost economies of scope in the non-life insurance industry. More recently, by using U.S. life insurance industry data during the period, Meador, Ryan and Shellhorn (2000) found cost economies of diversification across different lines of business. By contrast, other empirical studies have suggested that diversification decreases CE performance or that there are no global cost economies of scope (e.g., Fields and Murphy, 1989; Grace and Timme, 1992; Worthington and Hurley, 2002). Grace and Timme (1992) found no cost complementarities in the U.S. life insurance industry. Using data for forty-six Australian insurers, Worthington and Hurley (2002) suggested that product diversification has no effect on CE. Some other papers also found support for diversification having a negative effect on the performance of insurance firms (e.g., Tombs and Hoyt, 1994; Jeng and Lai, 2005; Liebenberg and Sommer, 2008). Using data for P-L and life-health insurers, Tombs and Hoyt (1994) reported that stock returns are positively related to product focus, where the product focus is measured by the firm s line of business Herfindahl index. Jeng and Lai (2005) suggested that there are cost diseconomies of scope for Japanese keiretsu nonlife insurance firms. Liebenberg and Sommer (2008) found that undiversified insurers consistently outperform diversified insurers. Their findings on the measurement of accounting performance further indicated that the product diversification is associated with a penalty of at least 1 percent of the return on assets or 2 percent of the return on equity. A few prior studies have examined economies of diversification across life and P-L insurance segments, rather than within the same insurance segment. Berger et al. (2000) examined the conglomeration hypothesis, which suggests that offering both life and P-L insurance products can add value by exploiting economies of scope, versus the strategic focus hypothesis, which suggests that firms can maximize value by focusing on their core business. They found that the conglomeration hypothesis is valid for some types of insurers and that the strategic focus hypothesis holds for other types. Cummins et al. (2003) extended Berger et al. (2000) and used DEA to examine the conglomeration hypothesis and strategic focus hypothesis. They concluded that the strategic focus hypothesis appeared to be a more efficient strategy than the conglomeration strategy in 4

8 the U.S. insurance industry. Using more recent data (i.e., ), Elango et al. (2008) found that P-L insurers that also sell life insurance products have better financial performance than P-L insurers that do not sell life insurance. Most of the earlier empirical studies have focused on the effect of product diversification on the financial performance of the insurance firms, whereas relatively few studies have investigated insurance distribution channels. One exception is the study by Fields and Murphy (1989), who examined the cost of life insurance distribution channels. They found that there are positive as well as negative cost complementarities for different pairs of life insurance products, which provide mixed results of product diversification for insurance agencies. However, their study did not analyze the CE and productivity of these distribution channels. Their study did not examine economies of diversification across life and P-L insurance segments for distribution channels either. Other strands of the literature on distribution channels have compared the costs or profits of insurance firms using alternative distribution channels (e.g., Joskow, 1973; Cummins and Vanderhei, 1979; Barrese and Nelson, 1992; Berger, Cummins, and Weiss, 1997; Regan, 1997, 1999; Park et al., 2009). Such studies focus on the question as to which model is more efficient in terms of distributing insurance products. They found that P-L insurance firms that use independent agencies incur higher costs than those that use direct writers 6. Some studies have attempted to explain or interpret this coexistence of low-cost and high-cost distribution systems. The first group of studies has argued that the incentive conflicts between insurance firms and direct writers and independent agencies may be one possible explanation for the coexistence of low-cost and high-cost distribution systems. The other group of studies has contended that market imperfections, such as search costs of customers, lead to the coexistence of the two distribution systems. Although the previous insurance literature on distribution channels has focused more attention on the efficiency of insurance firms, there are no studies that directly examine the CE of insurance distribution channels. To fill this gap in the literature, our analysis extends the prior literature by providing the first analysis of cost frontier efficiency and the relationship between CE and product diversification in regard to distribution channels. 6 The classification of direct writers and independent agencies is based on the degree of vertical control in regard to salesmen. The direct writer system includes mass marketing, the employee sales force, and the exclusive agencies. The independent agency system comprises the independent agencies of marketing and insurance brokers. 5

9 3. Life Insurance Distribution Channels in Taiwan The insurance distribution environment in Taiwan provides a unique opportunity to explore the CE of distribution channels. This section provides a brief overview of the insurance distribution channels in Taiwan. By law, an agency is authorized to have licenses to sell either life or P-L insurance policies, while a broker is authorized to have licenses to sell both life and P-L insurance policies. Some brokers adopt a diversified strategy to sell both life and P-L insurance, while others follow a focused strategy to sell either life or P-L insurance, even where the joint distribution of insurance products has been legally allowed. Figure 1 shows the industry trends regarding the premiums written and commissions for agencies and brokers over the past decade in Taiwan. The insurance intermediaries over the past decade have experienced rapid growth in terms of both the insurance premium and fee income. The premiums written for all agencies and brokers reached NTD$ million (about US$13.32 million) in The total commissions and agency fee income reached approximately NTD$27 million (about US$0.83 million) in (Insert Figure 1 here) 4. Hypothesis Development We develop two hypotheses in this section. Hypothesis I Insurance agencies are more efficient than insurance brokers. We conjecture that agencies may be more efficient than brokers for the following two main reasons. First, agencies may be more efficient than brokers due to incentive conflicts between agencies and insurance firms. In Taiwan, insurance agencies are authorized by insurers to serve as sales representatives and to act as business agencies on the insurance firms behalf, whereas brokers only distribute insurance products on the basis of the interests of the insured. This arrangement better aligns the interests of agencies and 6

10 insurance firms than those of brokers and insurance firms. Thus, the incentive conflicts between agencies and insurance firms may be weaker than the incentive conflicts between brokers and insurance firms. The relationship between agencies and insurance firms is closer than the relationship between brokers and insurance firms. Accordingly, agencies may have lower costs of search and negotiation for distributing insurance products relative to brokers. Thus, agencies are expected to be more cost efficient than brokers. Second, agencies perform more service functions than do brokers, and thus agencies are expected to have larger outputs than brokers. According to agency management law in Taiwan, an agency may be authorized to collect premiums, handle claims, and engage in policy underwriting. Agencies can receive compensation for performing these services on behalf of the insurance firm on the basis of an agency contract or a letter of authorization. On the other hand, the principal goal of the broker is to sell insurance and the broker is compensated only on a commission basis. Agencies are expected to receive more compensation than brokers to reflect the variety of customer services provided by agencies, all else equal. Thus, agencies are expected to produce more output than brokers. As discussed, we anticipate that agencies will tend to be more efficient than brokers. Hypothesis II Diversified brokers are more efficient than specialized agencies and specialized brokers. The proponents of the strategic focus hypothesis contend that specialized firms outperform diversified firms due to core competency, agency problems, and internal capital. Specialized firms may add value by focusing on core competency and business. If specialized firms can develop superior expertise and provide tailored products for customers, they may reduce costs or may be able to charge more than diversified firms (Berger et al., 2000). In addition, diversified firms may be less efficient than specialized firms due to agency problems, the less efficient use of internal capital, and higher costs of coordination and administration. Diversification may create an internal capital market, and this internal capital market may reduce market discipline (Easterbrook, 1984). In the absence of external capital market discipline, it may be easier for managers to accept negative present value investment for their private benefit (Jensen, 1986) or to add more business to protect their human capital in the company (Amihud and Lev, 1981). Diversified firms may also suffer significant losses of value due to funds being channeled to poorly performing segments (Jensen, 1986; Berger and Ofek, 1995). In addition, if firms become more complex, the monitoring and controlling function may become 7

11 increasingly complex, thus exacerbating the agency problems of diversified firms. By contrast, the proponents of the conglomeration hypothesis argue that the benefits of diversification may be derived from the internal capital market, risk diversification, and economies of scope. First, firms may create a larger internal capital market through diversification and allocate the capital more efficiently, which may be less easy to do in the external capital markets due to the information asymmetry between the firm and outsiders (e.g., Williamson, 1970; Myers and Majluf, 1984; Gertner, Scharfstein, and Stein 1994). Second, diversified firms may be able to benefit from risk diversification. Diversified firms may be able to reduce their income volatility if the covariances between the revenue streams of different segments are low (Lewellen, 1971). This reduction in income volatility risk may allow firms to charge more to risk-sensitive customers who are willing to pay more for products from safer firms. Third, diversified firms may increase their profits due to economies of scope by improving their cost and/or revenue economies of scope. Potential sources of cost economies of scope include sharing fixed inputs in joint production (e.g., Teece, 1980). Providing a variety of products may help firms charge more to their customers who prefer one-stop shopping and are willing to pay more for the convenience. Diversified firms may also raise revenues and/or reduce marketing costs by sharing a brand name among products, thus increasing their cost economies of scope and/or revenue economies of scope (Berger et al., 2000). In terms of the conglomeration hypothesis, we should expect diversified distribution channels that sell both life and P-L insurance to have advantages over specialized distribution channels. Diversified distribution channels may benefit from product diversification by sharing fixed inputs in the joint distribution of insurance products, thereby improving their efficiency. 5. Methodology In this section, we first discuss the data envelopment analysis and Malmquist indices used to measure the relative efficiency and total factor productivity change of insurance agencies and brokers. We then describe the sample data. The section concludes by discussing the characteristics of distribution channels that may influence these agencies and brokers efficient performance and introduce the regression model. 5.1 Data Envelopment Analysis There are two major approaches to estimating frontier efficiency: the parametric 8

12 (econometric frontier) approach and the nonparametric (mathematical programming) approach. The parametric approach uses stochastic frontier technology to estimate the economic efficiency of decision-making units. Two potential drawbacks of the parametric approach are the selection of the functional form and the possible specification error of the error term. 7 We choose the nonparametric approach to estimate the relative efficiency of agencies and brokers, because the nonparametric approach avoids the problem of choosing a functional form and the problem of specification error in the error term. In addition, the efficiency measurement can be decomposed into different components using DEA, while it is more difficult to decompose the efficiency into its components based on the parametric approach. 8 Moreover, the Malmquist indices for the nonparametric approach allow us to examine the productivity evolution over time. This approach decomposes the total factor productivity change into TE change and technological change. The nonparametric approach has been widely adopted in insurance efficiency studies, 9 and the nonparametric approach used in this study is based on the work of Farrell (1957), Färe, Grosskopf, and Lovell (1985), and Färe et al. (1994). For expositional ease and brevity, we do not present the methodology in detail here. 10 In order to show whether agencies are more efficient than brokers, we use the DEA approach to measure CE for each agency and broker. CE is the ratio of the minimum costs the firm could have realized by operating on the efficient cost frontier to its actual costs and can be decomposed into AE and TE. AE measures the firm s ability to use the cost minimizing combination of input vectors, given the input prices. TE presents the firm s ability to generate maximum outputs for a given level of inputs. A firm achieves full TE if it operates on the production frontier. TE can be further classified into two components: pure technical efficiency (PTE) and scale efficiency (SE). PTE reflects the proportion by 7 The parametric approach requires the exact forms of cost, revenue, or profit functions to estimate the frontier for the sample firms being studied. It also requires a distributional assumption regarding the error term in order to correctly separate the random error term from the inefficient error term. 8 See Cummins and Weiss (2000, p. 784) for details. 9 For a survey on efficiency studies in the insurance industry, see Berger and Humphrey (1997), Cummins and Weiss (2000), and Eling and Luhnen (in press). 10 Readers wishing to be more fully informed of the nonparametric approach are referred to Farrell (1957), Färe et al. (1985), Färe et al. (1994), and Cummins and Weiss (2000), which discuss and present frontier efficiency and productivity methods in more detail. 9

13 which the firm could reduce its input usage by using the best technology, and is measured relative to the variable returns to scale frontier. SE reflects whether the firm operates under constant returns to scale (CRS). A firm will have achieved full scale efficiency if it operates at CRS. Therefore, the firm can be technically inefficient because it uses inappropriate technology and/or it operates at the wrong scale of operations. All efficiency scores range between zero and one, with a fully efficient firm having an efficiency score equal to 1 and an inefficient one having an efficiency score above zero and below 1. In order to examine whether agencies constantly outperform brokers over the sample period, we use Malmquist indices to estimate the evolution of the change in the total factor productivity of agencies and brokers over time. The productivity change mainly comes from the TE change, which measures the change in a firm s distance from the production frontier, and technological change, which measures shifts in the frontier over time. If the firm improves its TE from period t to period t +1, then the firm will be closer to the production frontier in period t +1 than it was in period t. The score for TE change >1 is usually interpreted as catching-up to the production frontier, while the score for TE change <1 is usually interpreted as falling behind over time. On the other hand, if favorable technological change occurs, the production frontier will shift in a favorable direction. Technological change >1 implies technical progress, while technological change <1 refers to technical regress. Outputs, Inputs, and Input Price Following Fields and Murphy (1989), we use commission dollars as a proxy for the outputs of insurance agencies and brokers for two reasons. First, the primary function of agencies and brokers is distributing insurance products. The commission is the compensation received by agencies and brokers for distributing insurance products. Second, commissions are important for agencies and brokers and are commonly available across all agencies and brokers. We also use the total premium volumes as a proxy for another output of agencies and brokers. Total premium volume is useful to represent the product distribution function. If agencies and brokers sell more insurance policies, they could collect more premiums for insurance firms. The commission dollars and total premium volume are deflated to 2006 New Taiwan (NT) dollars using Taiwan s consumer price index (CPI). We follow the recent insurance efficiency literature (e.g., Berger et al., 2000; Jeng, Lai, and McNamara, 2007) in defining inputs as equity, labor, and business services. The 10

14 equity is measured by the book value of the equity capital of agencies and brokers. The price of equity is based on the liability-to-capital ratios. 11 The quantity of labor equals the number of salesmen from the agencies and brokers at the end of the year. The price of labor is the national average wage rate for life industry and insurance intermediaries from the Directorate-General of Budget, Accounting and Statistics (DGBAS) of the Executive Yuan in Taiwan. Business services are measured as the total business service expenses divided by the price of business services. The price of business services is given by the national wage rate for the business services sector from the DGBAS. The equity and business services are deflated to 2006 NT dollars using the CPI. 5.2 Sample Selection Our primary data source consists of the annual financial statements filed by agencies and brokers for We include only agencies and brokers that remained either diversified or specialized for the whole sample period since agencies and brokers that change their strategy may not be perfectly representative of the benefits or costs of maintaining one of the distribution channel strategies. We also exclude agencies and brokers owned by banks, motor companies, and business offices owned by individuals. The agencies and brokers owned by banks may have problems of interpretation, because banks may use the same labor input to produce insurance products as well as bank products but the outputs in this paper only include insurance outputs. The DEA approach does not allow for the exclusion of the irrelevant input part, such as the labor input for producing bank outputs. Therefore, we exclude agencies and brokers that are subsidiaries of banks. Likewise, for agencies and brokers that are subsidiaries of motor companies, if the same labor input produces insurance products as well as non-insurance products, the estimated efficiency of agencies and brokers may have problems of interpretation. Thus, we exclude agencies and brokers that are subsidiaries of banks and motor companies. We 11 The ratio of an insurer s net income to capital is not used as the price of equity because insurers with poor performance tend to have negative net incomes and the price of equity cannot be negative. Following Jeng and Lai (2005), we use the debt-equity ratio of the insurer as the price of equity, because the price of equity should be a function of a firm s debt-equity ratio. In the theory of corporate finance, ROE=ROA+D/E(ROA-ROD), where ROE indicates the return on equity; ROA is the return on assets; D/E is the ratio of debt to equity; and ROD denotes the return on debt. We often assume that the ROA of different firms in the same industry is approximately equal. If we further assume that the ROD for different firms in the same industry is approximately the same, then the ROE mainly depends on the ratio of debt to equity. The explanation is provided in detail in Footnote 12 of Jeng and Lai (2005). 11

15 also exclude agencies and brokers with zero or negative values for total output, input, input prices, or equity. These agencies and brokers are deleted because they cannot be considered to be viable operating firms. The final sample consists of 83 agencies and 197 brokers, together accounting for about 73.89% of industry insurance premiums in All 83 agencies sell only life insurance and do not sell P-L insurance. Among the 197 brokers, 80 brokers also sell P-L insurance. These unique panel data sets enable us to gain a greater insight into the CE of the distribution channels. The means of output, inputs, input prices, and firm characteristics are shown in Table 1. The summary statistics for the agency and specialized broker samples are shown in the first and second columns, respectively. These columns show that agencies are on average significantly larger than specialized brokers for all output and input quantities, implying that agencies use more input quantities and also produce more output quantities than specialized brokers. Thus, we cannot determine whether agencies are more efficient than brokers by simply examining the output and input quantities of these two groups of firms. In addition, these columns show that agencies on average achieve better profitability than specialized brokers in terms of their returns on assets. The summary statistics for the diversified brokers are shown in the third column of Table 1, which indicates that diversified brokers on average are significantly larger than specialized agencies and specialized brokers for all output quantities, inputs, and total asset size. These findings imply that we cannot determine whether diversified brokers are more efficient than specialized agencies and specialized brokers by simply comparing the output and input quantities of these firms. (Insert Table 1 here) 5.3 Regression Analysis Following Wang, Jeng, and Peng (2007), we first measure efficiency scores and then regress the efficiency scores on firm characteristics to analyze differences in the efficiency of distribution channels. We estimate the regression model using pooled ordinary least squares (OLS) regressions 12 and Tobit regression models 13. The dependent 12 We also estimated the model, using random effects models. The results of the random effects models (not reported) are consistent with those of pooled OLS models with year dummies. The reason for using random effects models is that our key dummy variables (Agency and Broker) are entirely time invariant during the 12

16 variables are three major DEA efficiency measures used to proxy for the performance of agencies and brokers CE, AE, and TE. To further analyze the source of TE, we further treat the PTE and SE as dependent variables. In order to examine the productivity growth of agencies and brokers, we also perform regressions where the dependent variable comprises the Malmquist indices, which include TE change, technological change, and total factor productivity change. Test of Distribution Channels The most important independent variables in the regression model are the distribution channel dummies. The effects of the distribution channels can be captured by two dummy variables, namely, Agency (Agency=1 for agencies, 0 otherwise) and Specialized Broker (Specialized Broker =1 for specialized brokers, 0 otherwise). The base of these two dummy variables is diversified brokers. We expect a positive coefficient for Agency if the results of the analysis support our first hypothesis that agencies have greater efficiency than brokers. We expect a negative coefficient for Specialized Broker if our second hypothesis holds. The control variables in the regressions include firm size, the number of insurance firms represented, market share, financial leverage, and net income. We discuss the details as follows: Firm Size Fields and Murphy (1989) find a positive relationship between agency size and cost efficiency. Some of the evidence on the relationship between insurer size and performance has yielded mixed results. Some studies find that larger insurers outperform smaller ones (e.g., Gardner and Grace, 1993; Sommer, 1996; Cummins and Nini, 2002; Liebenberg and Sommer, 2007; Wang et al., 2007; Elango et al. 2008). Other studies find evidence of a negative effect of size on performance (e.g., Lai and Limpaphayom, 2003; Jeng et al., 2007). We use the natural log of assets as a proxy for firm size. As discussed, we do not have a clear prediction of the direction of the relationship between efficiency and size. sample period. Given the data structure, we use the random effects models and do not use the fixed effects models. 13 The previous efficiency literature uses Tobit regression models to analyze the efficiency because the range of the efficiency score is from zero to one, which violates the assumptions of the OLS regressions regarding the distribution of the dependent variable. 13

17 Insurer Partner We expect that the number of insurer partners has an impact on the efficiency of agencies and brokers, but we are not able to clearly predict the relationship between efficiency and the number of insurance partners. The insurance firms mainly perform the function of underwriting, investing premium funds, and evaluating claims, while agencies and brokers are responsible for distributing insurance products. Agencies and brokers must engage in negotiations with insurers regarding product distribution to ensure that the partnership works well. When agencies and brokers cooperate with more insurance firms, they may incur more costs of coordination, training, and administration, which may reduce their CE. If agencies and brokers cooperate with few insurance firms, they may avoid multiple negotiations, and reduce the costs of transactions and coordination. On the other hand, it may be advantageous for agencies and brokers to cooperate with more insurance firms because cooperating with more insurance firms may provide agencies and brokers with more product flexibility and may allow them to offer insurance products at more competitive prices. In addition, agencies and brokers cooperating with more insurance firms may more easily take advantage of the promotional efforts of the advertising insurance firms. For instance, if clients are attracted to the distribution channels by the promotional efforts of insurers, the distribution channels may have an incentive to encourage their clients to switch to buying the products of non-advertising insurers to reduce the shared advertising costs imposed by the advertising insurers, and thereby increasing the agencies and brokers cost efficiency. Agencies and brokers may more easily be able to take advantage of such opportunities if they partner with more insurance firms. To control for the impact of the number of insurer partners on efficiency, we include the number of insurer partners of agencies and brokers in the regression analysis. Market Share The relationship between market share and performance is addressed by Rhoades (1985). According to the relative market power hypothesis (RMP), firms with greater market shares that produce well-differentiated outputs are able to increase prices and earn higher profit, implying that the relationship between market shares and performance is positive. Berger (1995) provides some empirical support for the RMP in banking, while Choi and Weiss (2005) find mixed results regarding the RMP in the U.S. P-L insurance industry. A related theory is the efficient structure (ES) hypothesis, which holds that more efficient firms can charge lower prices, and therefore gain larger market share and 14

18 economic rents, resulting in a higher level of market concentration (e.g., Demsetz, 1973, 1974; Peltzman, 1977; and Brozen, 1982). The ES hypothesis predicts a positive relationship between efficiency and market share. To capture the impact of market share on efficiency, we use the ratio of the premiums of agencies and brokers to total industry premiums as the proxy for the market share. Leverage The relationship between an agency s financial leverage and performance is ambiguous. Financial leverage may be positively related to performance, because financial leverage can be used as a mechanism to bond managerial actions (Jensen, 1986). Agency cost theory predicts that leverage affects agency costs and thus influences firm performance. Berger and Bonaccorsi di Patti (2006) find evidence consistent with the agency costs hypothesis and suggest that higher leverage is associated with higher profit efficiency in the U.S. banking industry. On the other hand, high financial leverage may be negatively related to performance due to the increased probability of financial distress. High financial leverage may increase the probability of insolvency and subject firms to higher levels of expected costs of financial distress or bankruptcy. The higher level of financial risk may be reflected in the lower market value of firms. We use the debt-to-equity ratio to proxy for financial leverage in the regression analysis, but we are unable to clearly predict the variable. Net Income Prior studies find that a more efficient firm tends to be more profitable (e.g., Cummins and Zi, 1998; Cummins and Nini, 2002). In addition, the net income should be highly correlated with the efficiency of agencies and brokers because commissions, an important output for agencies and brokers, is subtracted from revenues to obtain net income. To control for this relationship, we include the net income variable in the regression analysis and anticipate that net income is positively correlated with the efficiency of agencies and brokers. 6. Empirical Results 6.1 Average Efficiencies Table 2 presents the results for DEA efficiency and the Malmquist index, and includes the average efficiencies for the agencies, specialized brokers, and diversified brokers in 15

19 the first, second, and third columns, respectively 14. The relative efficiency of each group of firms is based on the pooled frontier consisting of all agencies and brokers 15. (Insert Table 2 here) Our results show that the CE for agencies (specialized brokers) averaged 52.2 (40.9) percent, indicating that agencies (specialized brokers) could have reduced their cost by 47.8 (59.1) percent on average if they had operated with full CE. This implies that agencies significantly outperform specialized brokers in terms of CE. The CE for diversified brokers is, on average, 51.1 percent and is significantly higher than that for specialized brokers, implying that diversified brokers outperform specialized brokers in the CE sense. The values of AE for agencies, specialized brokers, and diversified brokers are 68.2, 65.0, and 67.7 percent, respectively. It implies that agencies are more successful in choosing the cost-minimizing combination of inputs. There are no significant differences in AE among these groups of insurance distribution channels. The values of TE show that agencies (specialized brokers) could have produced their outputs using 61.1 (50.1) percent of the inputs that were actually consumed. The TE for agencies is significantly higher than that for specialized brokers, implying that agencies are significantly more efficient than specialized brokers in terms of TE. It also suggests that the larger cost advantage of agencies over specialized brokers is attributable to TE, rather than AE. This means that, given a fixed amount of outputs, agencies could use smaller amount of inputs than specialized brokers. The TE results also show that diversified brokers statistically and significantly outperform specialized brokers in terms of TE. We further examine the components of TE, namely, PTE and SE. They show that agencies have significantly higher PTE and SE than specialized brokers. Thus, the source of the TE advantage of agencies over specialized brokers is derived from PTE and SE, implying that agencies are more technically efficient than specialized brokers because 14 The asterisks in the last three columns give the results of the t-statistics for differences between the efficiency results in the corresponding cells of the two columns. 15 We follow Cummins et al. (1999) and Jeng and Lai (2005) and test the null hypothesis of whether the agencies and brokers operate on the same frontier, i.e., that a pooled efficiency frontier can be used to examine the difference in efficiencies between two groups. According to the results of the non-parametric tests, including ANOVA, Wilcoxon Z, Median Z, Van Der Waerden Z, and Savage Z, we cannot reject the null hypothesis that the pooled frontier and separate frontiers are identical. Thus, we compare the efficiency scores based on the pooled efficiency frontier. 16

20 agencies choose better technology and produce at output levels closer to the minimum average cost point than specialized brokers. The PTE for diversified brokers is significantly higher than that of specialized brokers (72.8 percent versus 61.8 percent), suggesting that diversified brokers are more efficient than specialized brokers in terms of PTE. Thus, the larger TE of diversified brokers is attributable to PTE, rather than to SE. This implies that diversified brokers are more technically efficient than specialized brokers because they choose the better technology. Overall, the results of CE and its components suggest that agencies outperform specialized brokers and that diversified brokers are more cost efficient than specialized brokers. Next, Malmquist indices are used to examine the evolution of productivity change over time. This shows that the average technological change for agencies, specialized brokers, and diversified brokers is 1.203, 1.418, and 1.248, respectively. This implies that specialized brokers experience more technical progress relative to agencies and diversified brokers over the sample period. 6.2 Regression Analysis of Efficiency To further analyze the difference in efficiency between agencies and brokers, we perform regression analyses with efficiency as a dependent variable and firm characteristics as the independent variables. Table 3 shows estimates of the impact of parameters from the OLS and Tobit models on performance measures. Panel A of Table 3 reports the results of CE, AE, and TE, while Panel B reports the results of PTE and SE, and Panel C reports the results of the Malmquist indices. We test for multicollinearity using a variance inflation factor (VIF) for OLS regression models. The VIF are all below the benchmark of 10 as suggested by Belsley, Kuh, and Welch (1980). Thus, the multicollinearity problem does not tend to be an empirical issue in the sample. The adjusted R-squares of the OLS regressions indicate that the goodness of fit of the estimated OLS regression model is moderate. (Insert Table 3 here) We first compare the efficiency between the diversified and specialized distribution channels, before comparing the efficiency between the agencies and specialized brokers. When comparing the efficiency between diversified brokers and specialized agencies, it 17

21 can be seen that the coefficients of Agency are positive but insignificant in all equations, implying that there is no evidence that diversified brokers outperform specialized agencies. It should be noted that specialized agencies have two effects: the agency effect and the specialization effect. 16 As already discussed, we expect that agencies may outperform brokers and that diversified firms may be more efficient than specialized firms. The insignificant coefficient of Agency may reflect the net effect of these two influences on the marginal difference in efficiency. When comparing the efficiency between diversified brokers and specialized brokers, it is shown that the coefficients of Specialized Broker are found to be negative and statistically significant in the CE, TE, and ROA equations, indicating that specialized brokers are less efficient than diversified brokers, which is consistent with our second hypothesis. The CE advantage of diversified brokers over specialized brokers is attributable to differences in TE, and not in AE. This means that diversified brokers use smaller quantities of inputs to produce a given level of output than specialized brokers. The efficiency benefits of diversified brokers may be realized through shared salesmen, information systems, service centers or other fixed inputs. Our results also agree with Kellner and Mathewson (1983), Meador, Ryan, and Shellhorn (1998), Cummins, Weiss, and Zi (2003) and Hirao and Inoue (2004), who find evidence of a positive relationship between diversification and performance in the case of insurance firms. Among other control variables, the coefficients of Size are significantly positive in the AE model, suggesting that larger firms are more successful in choosing cost-minimizing combinations of inputs. This result supports the finding of Fields and Murphy (1989) with respect to agency size. However, the coefficients of Size are significantly negative in the TE model, indicating that smaller firms are more technically efficient than larger firms. The Market Share variable has significantly positive coefficients in all performance models. The positive relationship between market share and ROA is statistically significant, agreeing with the prediction of the relative market share hypothesis. This suggests that the agents are more likely to increase their profitability when their market share increases. The coefficient of Leverage is positively significant in all efficiency models while it is negatively significant in the ROA models. The efficiency results are consistent with the prediction of the agency costs hypothesis that higher financial 16 Because agencies in Taiwan can only distribute either life insurance products or property liability insurance products, agencies are also characterized by a specialization effect. 18

22 leverage is associated with improved efficiency performance due to the effect of monitoring by the debt holder. The coefficient for Net Income in the CE model is positively significant, which is consistent with expectations. This result suggests that a more cost efficient distribution channel tends to result in more net income, all else equal. Panel B reports the results of the components of TE. The coefficient for Specialized Broker is negatively significant in the PTE models, suggesting that diversified brokers are more efficient than specialized brokers in terms of PTE. Since PTE is one component of TE, this result also suggests that the superior TE of diversified brokers is due to PTE gains rather than the improvement in SE. The coefficients of Size are negative and significant in the PTE and SE models, a finding that is consistent with the results for the TE models. This indicates that a small distribution channel tends to be more efficient than a larger one in terms of PTE and SE. The positive coefficients of Market Share in the PTE and SE models suggest that the relationship shared between efficiency and market share is positive, which is consistent with the TE results. This suggests that the distribution channels with greater market shares tend to have higher performance in terms of PTE and SE. The results for Leverage in the PTE and SE models are consistent with the TE results and suggest that leverage affects agency costs and thus positively influences firm efficiency. The coefficient of Net Income is significantly positive in the PTE equations but is significantly negative in the SE equations. It shows that the relationship between PTE and net income is positive. Panel C shows the results of the Malmquist indices. The Agency and Specialized Broker variables are insignificant in all regressions. Thus, we cannot draw conclusions regarding whether agencies outperform brokers and whether diversified channels outperform specialized channels in terms of the Malmquist indices. The coefficients of Market Share and Leverage are significantly negative in the technological change model, suggesting that firms with a higher market share and/or more leveraged experience slow down the evolution of technological change. Next, we compare the efficiency between agencies and specialized brokers. It should be noted that there are three groups of insurance distribution channels and we include two dummy variables for the insurance distribution channels, namely, Agency and Specialized Broker, and the omitted category is diversified brokers. The coefficient of Agency only reflects the differences in the coefficients between agencies and diversified brokers. The coefficient of Specialized Broker only reflects the differences in the coefficients between specialized brokers and diversified brokers. Next, we test whether agencies are more 19