Why Does Global Diversification Still Make Sense? A Cross-Firm Analysis of the Risk and Value of Diversified Firms

Transcription

1 Why Does Global Diversification Still Make Sense? A Cross-Firm Analysis of the Risk and Value of Diversified Firms Diego Escobari escobarida@utpa.edu The University of Texas Pan American Mohammad J. Nejad* mjnejad@utpa.edu The University of Texas Pan American Thanh Ngo ngot@ecu.edu East Carolina University * Corresponding author: University of Texas - Pan American, 1201 W. University Dr., Edinburg, TX Phone: (956) ; mjnejad@utpa.edu.

2 Why Does Global Diversification Still Make Sense? A Cross-Firm Analysis of the Risk and Value of Diversified Firms Abstract This paper examines the impact of corporate diversification on firm risk exposure and associate its impact with the excess value of diversified firms for a sample of firms from We document that global diversifications enhances firm value, while industrial diversification destroys firm value. Global diversification adds value by mitigating firm idiosyncratic risk. More importantly, while industrial diversification reduces firm systematic risk, global diversification increases systematic risk. Consistent with the CAPM theory that only non-diversifiable systematic risk should be rewarded with higher expected returns, this effect can explain why industrially diversified firms experience a valuation discount and globally diversified firms experience a valuation premium. Our results remain robust after controlling for the potential endogeneity of the diversification decision through various self-selection models and internal instruments in a dynamic panel setting. JEL Classification: G11; G12; G30; G32 Keywords: Corporate diversification; Firm Value; Idiosyncratic risk; Systematic risk 1

3 1. Introduction Investors who wish to diversify their portfolios across the markets or within a market can choose to invest in firms of different markets/industries themselves (home-made diversified portfolios) or invest in diversified firms instead. Mathur and Hanagan (1983) suggest that barriers to investment including incompletely-integrated capital markets, transaction costs and information access limitation can impose impediments on homemade diversification, and thus make it more efficient for investors to invest directly in diversified firms to achieve diversification. Empirically, researchers have attempted to document the benefits of investing in diversified firms by looking at the valuation or the risks of these firms. Evidence on the impact of diversification on firm risk and firm valuation, however, is in discordant. Studies by Comment and Jarrell (1995), Lang and Stulz (1993), Berger and Ofek (1995), Servaes (1996), Denis et al. (2002), Kim and Mathur (2008) and Hoechle et al. (2012) document strong evidence of diversified firms being traded at a discount, hence the coined term diversification discount. Campa and Kedia (2002), Villalonga (2004b), Villalonga (2004a), Gande et al. (2009) and He (2012), on the other hand, find evidence of diversification premium once they control for the endogeneity of firms decision to engage in diverse operations. Similarly, while most studies agree on the evidence of lower stock return variance among diversified firms in comparison to non-diversified firms, findings on the impact of diversification on firm risk exposure are puzzling. Earlier studies by Fatemi (1984), Lubatkin and Chatterjee (1994), Shaked (1986), and Stulz (1999) show that diversification lowers firm systematic risks. Later studies by Reeb et al. (1998) and Olibe et al. (2008), on the contrary, document higher systematic risks for the diversified firms. To shed more light into the puzzling evidence of diversification effects, we investigate the impact of diversification on firm risk exposure and firm valuation simultaneously. First, we examine firms risk exposure including idiosyncratic risk, U.S. market risk and world market risk, in conjunction with their excess value. Earlier studies by Rugman (1976), Brewer (1981), Fatemi (1984), Thompson (1984), Shaked (1986), and Lubatkin and Chatterjee (1994) examine the impact of diversification on the stock 2

4 return variance and U.S. market beta of firms. While later studies by Mitton and Vorkink (2010), Hund et al. (2010) and Lee and Li (2012) examine the impact of diversification on stock return skewness, the volatility of firm profitability and ROE, respectively, the impact of diversification on firms risk exposure, especially idiosyncratic risk, has been largely ignored in the literature review. Our focus on idiosyncratic risk in this paper further differentiates our study from prior studies. Merton (1987), Campbell et al. (2001) and Malkiel and Xu (2002) suggest that not all investors can have access to diversified portfolios, which is different from the CAPM assumption, and thus investors command a return premium for bearing idiosyncratic risks. Second, to estimate the idiosyncratic risks as well as the systematic risks of the sample firms, we employ a modified version of Fama-French 3-factor model in which we incorporate the world returns factor (MSCI World excluding U.S. index) to circumvent the issue of incomplete modeling. Stulz (1999) points out that applying one-factor local CAPM might be misleading in the context of increased integration of global markets. Indeed, Aggarwal and Harper (2010) suggest that purely domestic firms are not immune to global market factors. They empirically show that domestic firms are exposed to foreign exchange rate changes since they have to face the competition with global firms in terms of cost structure changes driven by foreign exchange rate changes. The use of Fama-French 3-factor model instead of onefactor market model (as employed in prior studies on risk reduction impact of diversification) also helps yield better measures of risk since the model explains stock expected returns better (Fama and French, 1996). Third, we control for the endogeneity of the decision to diversify, as suggested by Campa and Kedia (2002), by employing internal instruments and dynamic panels methods. Arellano and Bond (1991) explain that our dynamic panel estimator is consistent with rational expectation models and allows firms to behave dynamically. This is important because firms are forward looking when deciding to diversify. Our dynamic approach provides consistent estimates and allows us to assess the impact of diversification on risk and excess value simultaneously instead of treating risk and excess value as two separate issues as 3

5 in prior studies. We posit that diversification affects firm risk, which in turn channels through firm excess value. As far as we are aware of, the study by Amit and Livnat (1988) is the only one that examines the impact of diversification on risk and returns simultaneously. They show that related diversification is associated with high risk-high return combination while unrelated diversification is associated with low risk-low return combination for firms. The study, however, is limited to a small sample from 1977 to 1984 and focuses only on industrial diversification; our study covers over a 13-year period from 1998 to 2011 and investigates both industrial diversification and global diversification. In addition, we examine excess value (instead of stock returns), as developed by Berger and Ofek (1995), to capture how investors value the whole firm compare to its segments. Additionally, while most existing studies on this topic in the literature examine data prior to the year 1997 (some studies like the one by Kim and Mathur (2008) extend the sample period till 2000), we investigate a sample of diversified and non-diversified firms from There are issues about the pre-1997 segment data reported by COMPUSTAT, as pointed out by Campa and Kedia (2002) and Villalonga (2004a). We document that industrial diversification and global diversification are inherently different as to their impacts on firm risk and excess value. While global diversification enhances firm value (+8.2%), industrial diversification destroys firm value (-13.1%). We attribute this result to the differential impacts of the two types of diversification on firm risk; global diversification reduces firm idiosyncratic risks significantly while industrial diversification brings about negligible idiosyncratic risk reduction. Furthermore, global diversification augments firm systematic risk while industrial diversification mitigates it. Since systematic risk is nondiversifiable, modern portfolio theory predicts higher returns for firms with higher systematic risks. In this way, higher systematic risks among globally diversified firms might command higher excess value accrued to these firms. Overall, our study contributes to detangle the discrepancies in the literature and to provide new insights into the relation between diversification, firm risk and firm valuation. 4

6 The remainder of the paper is organized as follows. In Section 2, we summarize the existing literature on the diversification effects on firm value and risk and develop our hypotheses. In Section 3, we describe the data and elaborate the methodology. We report the empirical findings and their implications in Section 4. We conclude the paper in Section Literature Review and Hypotheses Development 2.1. Diversification and Risk Following modern portfolio theory, diversified firms can lower their risk exposure by diversifying in industries or countries that have low correlation with their operation. In this case, firms should benefit from a negative relation between diversification and risk. Lamont and Polk (2002) suggest risk has limited explanation power to explain expected returns of firms with diversification discount. Grass (2010) reveals that the conglomeration impact on firm risk is heavily conditioned on firm size; thus, diversification reduces small firm s risk exposure while it does not significantly affect large firms. Mitton and Vorkink (2010) investigate the association between diversification discount and expected returns. They use the skewness of return distribution to proxy for risk to find focused (single-segment) firms have greater skewness exposure as compared to diversified firms while making no distinction between industrial diversification and global diversification. Consequently, they attribute the discount observed among diversified firms to the lower skewness of their returns. Hund et al. (2010) and Lee and Li (2012) use the volatility of profitability and return on equity, respectively, as proxies for risk and suggest that operational diversification reduces firm risk. While various risk measures have been examined in prior studies on diversification impact on firm risk, the impact of diversification on firm idiosyncratic risk has been largely ignored in the literature review. We fill this gap in this literature. The study by Douglas (1967) is perhaps the first empirical study that looks into idiosyncratic risk and finds that residual variance is priced in the cross section of stocks. In a modified CAPM framework, Levy (1978) reveals that the market beta estimator, which measures the 5

7 systematic risk, is biased and there is a possible role for idiosyncratic risk. Merton (1987), Campbell et al. (2001) and Malkiel and Xu (2002) suggest that not all investors can have access to diversified portfolios, which is different from the CAPM assumption, and thus investors command a return premium for bearing idiosyncratic risks; idiosyncratic risks should matter as a relevant pricing factor. We posit that diversification reduces firm idiosyncratic risks since diversification brings about imperfect correlated income streams. Modern portfolio theory suggests an inverse relation between portfolio diversification and firm-specific risk. Given that diversified firms are in fact portfolios of singlesegment firms, they should accordingly have lower risk as compared to their non-diversified counterparts. H1: Diversified firms have lower idiosyncratic risk than domestic single-segment firms. While diversification reduces idiosyncratic risk in general, we believe the magnitude of risk reduction differs between industrial diversification and global diversification. Industrial diversification involves spreading investments and thus income streams among different industries that are all exposed to the same U.S. market risk. As such, income streams for industrially diversified firms are correlated to some degree. Income streams for globally diversified firms, on the other hand, come from different geographic markets that are not fully integrated and thus are imperfectly correlated. H2: Globally diversified firms have lower idiosyncratic risk than industrially diversified firms. While most studies agree on the evidence of lower stock return variance and firm-specific risk among diversified firms in comparison to non-diversified firms, conflicting findings on the impact of diversification on firm systematic risk are puzzling. Examining a sample of 100 large UK firms listed in The Times 300 from , Thompson (1984) reports no significant relationship between firm diversification and systematic risk, confirming the CAPM theory that systematic risk is nondiversifiable. Earlier studies by Fatemi (1984), Shaked (1986) and Lubatkin and Chatterjee (1994) show that diversification lowers the systematic risks of the firms. Later studies by Reeb et al. (1998) and Olibe et al. 6

8 (2008), on the contrary, document higher systematic risks for the diversified firms. Reeb et al. (1998) argue that globalization exposes firms to various economic risks including political risks, foreign exchange risks, agency problems, information asymmetry and management self-fulfilling prophecies. As such, they suggest that diversified firms should experience higher systematic risk. Olibe et al. (2008) examine a sample of 594 firms from and document evidence in support of Reeb et al. (1998). We argue that industrial diversification and global diversification should have differential impact on firms domestic systematic risk. In industrial diversification, firms diversify into various industries of the U.S. market, which in a sense resembles the market portfolio; the more industrial segments the firm has, the more it is similar to the market portfolio. In that way, industrial diversification enables firms to track the market portfolio closely and thus reduces firm systematic risk. In global diversification, firms basically represent only a segment of the U.S. market portfolio, resulting in higher systematic risks. More importantly, global diversification exposes the firms to various economic risks including political risks, foreign exchange risks, agency problems, information asymmetry and management self-fulfilling prophecies (Reeb et al. 1998). H3: Globally diversified firms bear higher U.S. systematic risk and world market risk than industrially diversified firms Diversification and Excess Value In a round table discussion, Villalonga (2003) describes the documented diversification discount as manifestations of strong, semi-strong, and weak market efficiency forms. In the weak form, diversified firms trade at a discount relative to single-segment counterparts (Lang and Stulz, 1993) and (Berger and Ofek, 1995). In the semi-strong form, diversified firms trade at a discount relative to what those firms would be worth if they were split into pieces (Comment and Jarrell, 1995). In the strong form, diversified firms trade at a discount relative to what those firms would be worth if they had not diversified. The focus of this study, however, is mainly on the weak form. 7

9 Empirical studies on diversification effects on firm value report inconclusive results. Comment and Jarrell (1995) present evidence of negative relation between diversification and abnormal stock returns during Similarly, Lang and Stulz (1993) document a negative relation between the same diversification measures and Tobin s q. Berger and Ofek (1995) find similar results by estimating the value of a diversified firm s segments as if they were separate entities. Denis et al. (2002) extend Berger and Ofek (1995) s study to globally diversified firms and find that globally diversified firms, on average, experience the same valuation discount as industrially diversified firms. Several studies, however, have questioned previous findings about diversification discount. Hyland and Diltz (2002) argue that the decision to diversify is not random; firms that choose to diversify and firms that choose not to diversify are different in nature. Campa and Kedia (2002) suggest that diversification discount does not necessarily mean that diversification destroys value. After controlling for the endogeneity of diversification decision, they show diversification discount drops, and sometimes turns into premium (p.1731). Graham et al. (2002) also claim that diversification itself does not reduce firm value because diversifying firms acquire segments that were discounted prior to their acquisition. Villalonga (2004a) argues that the results of prior studies on diversification discount are subject to data snooping bias. Gande et al. (2009) argue that global diversification should enhance firm value since it brings about real and financial effects. We posit that global diversification is positively related to firm excess value while industrial diversification has a negative impact on firm excess value. In addition to the reasons provided by Gande et al. (2009), we argue that the differential impacts of the two forms of diversification on firm valuation are driven by the their differential impacts on firm risk. According to Hypothesis 2, global diversification brings about bigger reduction in idiosyncratic risks as compared to industrial diversification. Investors, however, might not price idiosyncratic risks in their investment if they have access to diversified portfolios as assumed in the CAPM. However, systematic risk is non-diversifiable. As such, investors should command higher excess value for bearing higher systematic risks of globally diversified firms (as 8

10 formulated in Hypothesis 3). Thus, we should observe higher excess value among globally diversified firms. H4: Global diversification enhances firm value by mitigating idiosyncratic risk and augmenting market risk. 3. Data and Methodology 3.1. Data Although prior studies mainly focus on the pre-1998 data, we compile our sample from the Industrial and Geographic Segment Tapes in COMPUSTAT database over the period Villalonga (2004a) suggests that the diversification discount documented in prior studies, which focus on the pre data, might be driven by the various data reporting issues with COMPUSTAT database. She uses Business Information Tracking Series (BITS) 1 database to report that diversification discount disappears or even turns into a premium. Prior to 1998, firms reported their business segment information according to the guidelines of Statement of Financial Accounting Standards (SFAS) No.131 in which business segments were classified by industry codes. Such classification might result in a firm with multiple related business lines being classified as a single-segment firm, which in turn distorts the resulting impact of diversification on firm valuation documented in previous studies (He, 2009). SFAS 14 was introduced to overcome the weaknesses for SFAS 131 in that business segments are now classified based upon their contributions to the firm revenues and expenses. He (2009) compares and contrasts the pre-1998 and post periods to show that classification biases in the pre-1998 period data might be accountable for the documented diversification discount in previous studies. He suggests that the post-1998 data capture better the degree of firm diversity and thus the impact of diversification on firm valuation. This study is different from He (2009) in that we examine the impact of diversification on firm risk and excess value simultaneously. Moreover, we propose a more comprehensive treatment of the 1 A census database that covers the whole U.S. economy at the establishment level. 9

11 endogeneity of the diversification decision. Similar to previous studies, we exclude financial and utility firms (primary SIC codes and ) due to their highly regulated status. Foreign incorporated firms are also eliminated. Following Berger and Ofek (1995) and Denis et al. (2002), we remove firm-year observations where the difference between the sum of the segment sales and total firm sales is greater than 1% and/or the total sales are less than $20 million Measure of Firm Diversification We employ three alternative measures of the diversification degree of the firms. First, following Denis et al. (2002), we classify firms with more than one business segment as industrially diversified and firms with more than one geographic segment as globally diversified. We report the sample distribution in Table 1. The sample includes 7,425 firms and 27,906 firm-year observations. Firms that are only industrially diversified are denoted domestic multi-segment domestic (DM), which represent 19.33% of the full sample (1,435 firms and 4,581 firm-year observations). Firms that are only globally diversified are designated global single-segment global (GS), which account for 24.89% of the whole sample (1,855 firms and 7,038 firm-year observations). Firms that are both industrially and globally diversified are labeled as global multi-segment (GM). There are a total of 1,514 GM firms with 5,484 firm-year observations, representing 20.39% of the sample. Domestic firms with only one business segment are designated as domestic single-segment (DS), which serve as the benchmark. Our sample consists of a total of 2,621 DS firms, representing 10,803 firm-year observations, or 35.3% of our sample. Secondly, we use the number of business and geographic segments to quantify the diversification degree of the firms. We also calculate the sales-based Herfindahl index used as the third proxy for firm diversification. The Herfindahl Index for the i th firm in year t is computed as: ( ) where SSales it denote the segment sales (which can be sales generated from an industrial segment or from a geographic segment) for firm i in year t. FSales it is firm i's total sales across all reported segments in 10

12 that year. Accordingly, we report industrial-segment sales based Herfindahl Index (I-HERF) and geographic-segment sales based Herfindahl Index (G-HERF) separately. For domestic single-segment firms (DS), the Herfindahl Index is equal to 1 and for multiple-segment firms (DM, GS and GM) is less than 1; more diversified firms have smaller index (closer to 0). We report the descriptive statistics of the last 2 measures of firm diversification and other firm characteristics in Table 2. On average, DM firms have 2.67 business segments while GS firms 3.35 geographic segments. GM firms have business segments and geographic segments. Singlesegment domestic and global firms have business-sales based Herfindahl index equal to one since that they are not industrially diversified. Similarly, geographic-sales based Herfindahl index is equal to one for single-segment and multi-segment domestic firms, which are not operating in foreign countries. DM and GM firms have the business-sales based Herfindahl index of and 0.552, respectively. GS and GM firms have the geographic-sales based Herfindahl index of and 0.593, respectively. GM firms have more business segments than DM firms (2.855 vs segments) and more geographic segments than GS firms (3.586 vs segments). Diversified firms tend to be larger in terms of assets and market capitalization than non-diversified firms. Globally diversified firms are larger in size than industrially diversified firms. They also have lower debt levels, lower capital expenditures, higher profitability, and higher advertising expense than industrially diversified firms Measure of Risk The three measures of risks of the sample firms that we are interested in are idiosyncratic risks, exposure to U.S. market performance (or U.S. market risk) and exposure to world market performance (or world market risk). We employ the following modified Fama-French 3-factor model to obtain the idiosyncratic risk, U.S. market risk and world market risk of each firm in the sample as follows: ( ) ( ) where is the excess return of firm i on day t; ( ) is the market excess return on day 11

13 t; is the excess return of the small-stock portfolio over the big-stock portfolio on day t; is the excess return of the high-book-to-market portfolio over the low book-to-market portfolio on day t. is the difference between the returns on the MSCI World Excluding U.S. index on day t and the risk-free rate in the U.S. market. The factors are obtained from Professor Kenneth French website. The above model is estimated cross-sectionally by firm and year. captures the exposure of the firm to the U.S. market. β 4 captures the exposure of the firm to the world markets outside of the U.S. We calculate the standard deviation of residuals for each firm in each year and use it as the idiosyncratic risk of the firm. We multiply the daily idiosyncratic risk by the square root of the number of trading days in a 1 year ( ) to obtain the annualized idiosyncratic risk. While prior studies including the ones by Brewer (1981), Shaked (1986), Amit and Livnat (1988), Kwok and Reeb (2000), Best et al. (2004), Reeb et al. (1998) estimate diversified firms risks using the one-factor market model (e.g. U.S. market return is the only explanatory variable), Stulz (1999) suggests that the global market factor should be controlled for in estimating the expected returns of globallydiversified firms whereby the global market factor is measured as the excess returns of the world index over the U.S. domestic risk-free rate. As the global markets have been more integrated over time, even purely domestic firms are not immune to changes in the global markets. Aggarwal and Harper (2010), for example, document significant exchange rate exposure born by domestic firms; changes in exchange rate affect the cost structure of global firms which in turn affect the competition between these firms and domestic firms. In this way, we incorporate the excess returns on the MSCI World Excluding U.S. index in the model to capture the trend of increased integration of global markets Measure of Excess Value Excess value is measured as the natural logarithm of the firm s actual market value (market value of equity plus book value of debt) divided by its hypothetical imputed value. Following Berger and Ofek (1995), we calculate the imputed segment values as product of segment sales by the median market value 12

14 to sales ratio of single-segment domestic firms in the same industry. To eliminate extreme observations, the calculated excess values are winsorized at the 5% level. 2 ( ) ( ) where, I(V) is the imputed value of the sum of segments as if they were separate entities. S i is the segment i s total sales, ( ) is the ratio of total capital to total sales for the median single-segment firm in segment i Analyses of the Relationship between Diversification, Risk and Excess Value To test for the relationship between diversification, firms excess value and firms risks, we compare and contrast the excess value and risk levels between different types of firms (e.g. DS, DM, GS or GM firms) (in Table 3), between the groups of firms with higher vs. lower number of business/geographic segments than the sample median, and between the groups of firms with higher vs. lower Herfindahl indices than the sample median (in Table 4). To test for the relationship between excess value and firm risk, we break down the sample into quartiles based upon each of the three measures of risks (idiosyncratic risk, U.S. market risk and World market risk) and compare and contrast the excess values between the quartile (in Table 5). We employ the traditional t-test and the non-parametric Wilcoxon rank sum test to test for the significance of the differences between the groups. We extend the univariate analyses and test if the results hold in multivariate framework that accounts for other factors (including unobservables) plus the potential endogeneity of diversification decision. We regress the calculated measures of firms risk and excess value on the firm diversification profile and other control variables as follows: 2 Before the winsorization, observations with extreme excess values are deleted, whereby actual firm value is either more than four times or less than one-fourth the imputed value (Denis et al., 2002). 13

15 where DM i,t, GS i,t and GM i,t are dummy variables that take the value equal to 1 if the firm is a domestic multi-segment firm, global single-segment firm or global multi-segment firm, respectively. In addition, we also replace these 3 dummy variables with (i) a dummy variable for globally diversified firms (which can have one or many business segments), (ii) number of business/geographic segments and (iii) business/geographic-sales based Herfindahl indices. The remaining control variables are selected based on prior studies. The natural logarithm of total assets (or market capitalization) is used to control for firm size. Industry-adjusted total debt to total assets ratio (DEBT) is included to control for relative financial leverage. Following Denis et al. (2002) and Campa and Kedia (2002), we include industry-adjusted capital expenditure-to-sales ratio (CAPX) to control for relative growth opportunities and industry-adjusted earnings before interest and taxes relative to sales (EBIT) to control for firm profitability. We include industry-adjusted ratios of advertising expenses to sales and R&D expenses to sales to control for information asymmetry in the firms. In addition, we control for year-fixed effects and firm-fixed effects. Prior studies by Campa and Kedia (2002), Villalonga (2004b), Ammann et al. (2012) and He (2012) document strong evidence that failure to address for the endogeneity of diversification decision can distort the empirical results on the relationship between diversification and the excess value of firms. Accordingly, we address this issue with two different approaches, alternatively. In the first approach, we 14

16 estimate a two-step Heckman self-selection model. In the second approach, we allow for dynamics and estimate the model using dynamic panels and a rich set of instruments. For the two-step Heckman self-selection procedure we first estimate the predicted probability of a firm decision to diversify, transform it into the Inverse Mills ratio as formulated by Heckman (1979) and include this Inverse Mills ratio (INVERSEMILL i,t ) in the regressions of Excess Value and Risk above. The predicted probability of a firm decision to diversify is estimated using the following regression: (9) We estimate Equation 5 using a probit regression and an ordinal probit regression, alternatively. In the probit regression, DIVERSIFIED is the dummy variable set equal to 1 for industrially and/or globally diversified firms and 0 for domestic single segment firms as in Campa and Kedia (2002). In the ordinal probit regression, DIVERSIFIED is coded to be 0 for domestic single-segment firms, 1 for domestic multi-segment firms, 2 for global single-segment firms and 3 for global multi-segment firms. We follow Campa and Kedia (2002) to select the variables and model the decision to diversify. LNASSET is the natural logarithm of the firm total assets. LAG1EBIT is the one-lagged ratio of EBIT-tosale ratio. LAG1CAPX is the one-lagged ratio of capital expenditure-to-sale ratio. SP is the dummy variable for firms included in the S&P indices. NUMDIVFIRMS is the number of diversified firms in the industry. SALEDIVFIRMS is the percentage of sales in the industry generated by diversified firms. MAVOL is the natural log of the values of all mergers and acquisitions in the industry. MANUM is the natural log of total number mergers and acquisitions in the industry. GDP is the real GDP growth. MAJOREX is the dummy variable for firms listed on major exchanges including NYSE, NASDAQ and AMEX. DIVPAID is the dummy variable for firms that pay dividend in the preceding year; we include this variable as suggested by Villalonga (2004b). 15

17 We estimate equation (5) using firm fixed-effect model since idiosyncratic risk is firm-specific. We conduct the Hausman test to make sure the fixed effect model is more appropriate for our data as compared to random effect model. 3 We estimate equations (6) and (7) using pooled OLS regressions with Driscoll-Kraay standard errors (Driscoll and Kraay, 1998), which allows for heteroskedasticity, autocorrelation and correlation between the panels (e.g. the firms in this case). We apply Driscoll-Kraay regressions since systematic risk is nondiversifiable and all firms are subject to U.S. systematic risk. In addition, given the trend of increased integration of global markets, even domestic firms are not immune to world market risk. Our second approach to account for the endogenous decision to diversify follows the dynamic panels proposed by Arellano and Bond (1991) and Blundell and Bond (1998). Initially these methods employ internal instruments, but we extend the set of instruments to include variables that explain the decision to diversify. In addition to accounting for endogeneity this approach allows agents to behave dynamically and it is consistent with rational expectations. This is important because when firms decide to diversify they are forward looking and form expectation about the future value of the firm. Starting with equation (5) we simply include a first order autoregressive term as a right-hand side regressor to allow for dynamics. Then we stack the set of right-hand side variables into a single matrix. Taking first differences of the resulting equation we construct the moment conditions, where is the set of instruments and denotes the first differences of the error term in equation (5) using matrix notation. Arellano and Bond (1991) propose using the lags of the levels of the variables on the right-hand side of (4). This estimator is known in the literature as the difference GMM estimator. We follow this approach but augment the matrix with the right-hand side variables in equation (9). We expect these to be valid instruments because as equation (9) suggest, they are correlated with the diversification decision. The validity of the moment conditions and the instrument list is tested using the Sargan test of over-identifying restrictions. 3 Results from Hausman test are not reported here to conserve space. 16

18 Blundell and Bond (1998) argue that if the series are persistent, the lags of the levels on the righthand side of equation (5) might be weak instruments. Hence, we combine the moments from the difference estimator with the moments that come from the equation in levels. This is known as the system GMM estimator. Because we allow for the existence of firm fixed effects, for firm i at time t the disturbance term can be written as, where denotes the firm observed and unobserved specific characteristics. For the instrument set we use lagged differences of as well as lagged differences of the right-hand side variables in equation (9). Both, the difference and the system estimators assume that the error term in equation (5) is not serially correlated. We test this assumption by looking the second order serial correlation in. 4. Results 4.1. Univariate Results on the Relationship between Diversification, Risk and Excess Value We report the mean (median) of the excess value, idiosyncratic risk, U.S. market risk and world market risk for each of the 4 groups of firms in Panel A of Table 3. We compare and contrast the statistics between the groups in Panel B. The mean (median) excess value of domestic single-segment firms is % (0.0%). While global single-segment firms enjoy an average valuation premium of 8.2%, domestic multi-segment firms and global multi-segment firms experience an average valuation discount of 13.1% and 3%, respectively. The results suggest that global diversification adds value and industrial diversification destroys value, which is consistent with findings by Kim and Lyn (1986), Campa and Kedia (2002), Villalonga (2004b, a), Doukas and Kan (2006), Dos Santos et al. (2008) and Gande et al. (2009). The t-test and Wilcoxon test in Panel B show statistically significant differences in the excess values between the 4 groups of firms. With regard to risks, global diversification reduces firms idiosyncratic risk significantly. While domestic single-segment firms and domestic multi-segment firms pose similar levels of idiosyncratic risks (0.653 and 0.654, respectively), global single-segment firms and global multi-segment firms 17

19 experience average idiosyncratic risk levels of and 0.539, respectively. The t-test and Wilcoxon test in Panel B show significant differences in idiosyncratic risk between globally diversified firms and domestic firms. The results are consistent with findings by Best et al. (2004). In contrast to industrial diversification, global diversification increases firms risk exposure to U.S. market as well as world market. The average U.S. market beta of domestic multi-segment firms is 0.775, less than that of domestic single-segment firms (0.804). Global single-segment firms and global multisegment firms experience higher levels of U.S. market risks of and 0.959, respectively. The t-test and Wilcoxon test in Panel B show significant differences in U.S. market risks between the 4 groups of firms. Our results are consistent with the findings by Reeb et al. (1998) and Olibe et al. (2008) who document that systematic risk is higher for multinationals. Global single-segment firms experience the highest level of exposure to world market risk (0.091) while the remaining 3 groups of firms have similar world market risks of 0.07, and 0.074, correspondingly; the difference is marginally significant (in Panel B). The results in this Table 3 seem to suggest that global diversification by itself brings about lower idiosyncratic risk, higher U.S. market risk, higher world market risk and subsequently higher excess value in comparison to industrial diversification. According to modern portfolio theory, only exposure to market risk is rewarded and firm-specific risk is not appreciated since diversified portfolios are accessible to all investors. In Table 4, to make sure that our results are not sensitive to how diversification degree of the firm is measured, we employ alternative proxies of firm diversification degree including the number of business segments (in Panel A), the number of geographic segments (in Panel B), business-sales based Herfindahl index (in Panel C) and geographic-sales based Herfindahl index (in Panel D). We break down the sample into two groups based upon whether the firm has higher or lower values of each of the four alternative measures of diversification than the median firm in the sample. We compare and contrast the excess value and risks between the two groups. 18

20 The results in Table 4 show lower excess value associated with higher number of business segments and lower business-sales based Herfindahl index (the lower the index, the more diversified the firm is). Higher excess value, on the other hand, is associated with higher number of geographic segments and lower geographic-sales based Herfindahl index. These findings together with the results in Table 3 provide evidence of global diversification premium and industrial diversification discount. Diversification brings about the expected risk reduction effects; idiosyncratic risk drops significantly, as the number of business/geographic segments increases and the business/geographic-sales based Herfindahl index decreases. Market risk, on the other hand, is unavoidable except for risk-free investment. U.S. market risk is associated positively with the number of business/geographic segments and negatively with the business/geographic-sales based Herfindahl. World market risk is not statistically different between the two groups classified by the alternative measures of diversification Univariate Results on the Relationship between Risk and Excess Value In Table 5, we focus on the association between risk and excess value. We compare and contrast the excess value for each quartile of idiosyncratic risk, U.S. market risk or world market risk, alternatively. We report the analyses for the whole sample in Panel A. The analyses are repeated for each of the four types of firms (DS, DM, GS and GM firms) in Panels B, C, D and E, respectively. Firms in the highest quartile of idiosyncratic risk experience lowest excess value and vice versa. Firms in the highest quartile of U.S. market risk, on the other hand, experience highest excess value and vice versa. Firms in the highest quartile of world market risk experience lowest excess value and vice versa; this applies for all firms except for domestic multi-segment firms. The results in this table confirm the long-standing tenet of modern portfolio theory that only systematic risk is rewarded while firm-specific risk is not appreciated. In short, the univariate results in Tables 3 through 5 suggest that globally diversified firms have higher excess value, lower idiosyncratic risk, and higher U.S. market risk. Industrially diversified firms have lower excess value, higher idiosyncratic risk, and lower U.S. market risk. We conjecture that the higher excess value associated with global diversification might be driven by the lower idiosyncratic risks 19

21 and/or higher market exposure. Investors can diversify their portfolios by constructing their home-made diversified portfolios or by investing directly in diversified firms. They would only value diversified firms if the diversified firms have lower idiosyncratic risks; otherwise, they could have diversified their portfolios themselves by stocking up non-diversified firms to achieve the same effects. Market risk, on the other hand, is undiversifiable and thus should be compensated with higher valuation Multivariate Results on the Relationship between Diversification and Idiosyncratic Risk In Table 6, we report the correlation matrix between the diversification measures, excess value, idiosyncratic risk, U.S. market risk and world market risk. The correlations among the variables are consistent with the univariate results in Tables 3, 4 and 5. We report the results from the probit regressions of the diversification decision from which we obtain the Inverse Mills ratio in Table 7. The results are consistent between the probit regression and the ordinal probit regression, and qualitatively similar to the results reported by Campa and Kedia (2002). We incorporate the Inverse Mills ratio calculated from the ordinal probit regression into later regressions of risks and excess value. In Table 8, we report the results from the fixed-effect regressions of idiosyncratic risks on firm diversification profile. The Hausman test (not reported here) shows that the fixed-effect regression is more appropriate for our data than random-effect regression. In Model 1, the coefficient on the dummy variable for global single-segment firms is negative and significant at the 5% level, suggesting that global diversification reduces firm idiosyncratic risk. The coefficients on the remaining three types of firms are insignificant. In Model 2, we also include a dummy variable for firms engaging in global diversification (which might also engage in industrial diversification at the same time), which is negative and significant at the 1% level. In Model 3, we use the Herfindahl index to capture the diversification degree of the firm. The coefficient on the business-sales based Herfindahl index variable is negative and significant; the higher the index, the lower the degree of industrial diversification and thus, the lower the idiosyncratic risk. The 20

22 coefficient on the geographical-sales based Herfindahl index variable, on the other hand, is positive and significant; the lower the index, the higher the degree of geographical diversification and thus, the lower the idiosyncratic risk. Consequently, while geographic diversification mitigates firm idiosyncratic risk, industrial diversification augments it. Globally diversified firms generate uncorrelated cash flows from different markets of different systematic risks. Industrially diversified firms, however, receive cash flows from different industries in the same market, which are exposed to the same systematic risk. As such, the reduction of idiosyncratic risk should be stronger in global diversification than in industrial diversification. In Model 4, we use the number of business/geographical segments to proxy for firm diversification. The coefficient on the number of business segments variable is positive and insignificant. The coefficient on the number of geographic segments variable is negative and marginally significant. This result is resonant of the results in Models 1 through 3, in which global diversification exerts stronger impact on firm-specific risk reduction as compared to industrial diversification Multivariate Results on the Relationship between Diversification and U.S. Market Risk In Table 9, we report the results from the cross-sectional analyses of the U.S. market betas of firms on their diversification profile. The coefficients on the dummy variables for global single-segment firms and global multi-segment firms (in Model 1) are positive and significant at the 1% level, suggesting that global diversification increases firms exposure to U.S. market risk. The coefficient on the dummy variable for domestic multi-segment firms is negative and significant, implying that industrial diversification reduces firms exposure to U.S. market risk. The results are consistent when we use the dummy variable for globally diversified firms (in Model 2), the Herfindahl indices (in Model 3) and the number of diversified segments (in Model 4) as alternative proxies for firm diversification Multivariate Results on the Relationship between Diversification and World Market Risk In Table 10, we report the results from the cross-sectional analyses of the world market betas of firms on their diversification profile. The coefficients on the dummy variables for global single-segment firms 21

23 and global multi-segment firms (in Model 1) are positive and significant, suggesting that global diversification increases firms exposure to world market risk. The result is consistent when we use the dummy variable for globally diversified firms (in Model 2). In Model 3, the coefficient on the geographic-sales based Herfindahl index variable is negative and significant at the 5% level, suggesting that the more diversified the firm is globally (the lower the index), then more exposed the firm is to world market risk. In Model 4, the coefficient on the number of geographic segments variable is positive and significant, confirming the evidence that global diversification increases firms exposure to world market risk Multivariate Results on the Relationship between Diversification, Risk and Excess Value In this section, we evaluate the simultaneous impact of corporate diversification and risk on excess value. Table 11 reports the fixed-effect regression results from the estimation of equation 8. All models control for year-fixed effects. More importantly, it is reasonable to expect firm-year observations to be highly correlated across years for any particular firm; therefore, we report t-statistics calculated from the standard errors corrected for firm clustering effects. Consistent with the univariate results, the coefficients on the dummy variables for domestic singlesegment firms (-0.056) and global multi-segment firms (-0.028) are negative and statistically significant (in Model 1), suggesting that industrial diversification destroys firm value. The coefficients on the dummy variables for global single-segment firms (0.079 in Model 1) and for global firms (0.061 in Model 2) are positive and significant, implying the value-enhancing role of global diversification. Results in Models 3 and 4 in which Herfindahl indices and number of diversified segments are used as proxies for firm diversification confirm the negative impact of industrial diversification and positive impact of global diversification on firm value. In Models 1 through 4, the coefficient on the idiosyncratic risk variable is negative and statistically significant, providing strong evidence that investors value idiosyncratic risk reduction effect of investing in globally diversified firms. In Model 5, the coefficient on the U.S. market risk is positive and 22

24 significant, supporting our hypothesis that bearing higher systematic risk is associated with higher excess value. In Model 6, the coefficient on the world market risk is negative but not significant, implying that investors do not appreciate bearing additional systematic risk beyond the U.S. market risk. Table 12 reports the dynamic panel regression results from the estimation of equation 8. These estimates control for the firm s potentially endogenous decision to diversify. The reported figures are system GMM two-step estimates with the numbers in parentheses being the t-statistics based on Windmaijer finite-sample corrected standard errors. All models control for year and firm fixed effects. Moreover, all models pass both specification tests. The second-order serial correlation tests strongly support the assumption of no serial correlation. Furthermore, the Sargan tests, which look at the sample analogs of the system GMM moment conditions, validate all the instrument sets. Consistent with the estimated results in Table 11, the coefficients on the dummy variables for domestic single-segment firms is negative and statistically significant while the coefficients of globalsingle segment firms (in Model 1) and for global firms (in Model 2) are positive and significant at the 1% level, confirming the value-reducing effect of industrial diversification and value-enhancing effect of global diversification. Interestingly, the coefficient of global multi-segment firms, which is negative and significant in Table 11 (when fixed effect estimation is employed), turns positive and significant (when GMM estimation is employed). In evaluating the performances of different estimation techniques including ordinary least squares (OLS), fixed effect (FE), long differencing (LD), generalized method of moments (GMM) and corrected least squares dependent variable (LSDVC) specifically in large corporate finance datasets such as unbalanced panels and endogenous regressors, Flannery and Hankins (2013) demonstrate that FE estimation, together with OLS estimation, performs the worst on average, and FE is affected particularly by shorter panels. As such, we take the results in Table 12 (when GMM estimation is employed) as more reliable and conclusive. The positive and significant coefficient of global multisegment firms suggests that global diversification enhances firm value even when it is combined with industrial diversification. In other words, the detrimental effect of industrial diversification on firm value 23