DOES INFORMATION ASYMMETRY EXPLAIN THE DIVERSIFICATION DISCOUNT? Abstract

The Journal of Financial Research Vol. XXVII, No. 2 Pages 235 249 Summer 2004 DOES INFORMATION ASYMMETRY EXPLAIN THE DIVERSIFICATION DISCOUNT? Ronald W. Best and Charles W. Hodges State University of West Georgia Bing-Xuan Lin University of Rhode Island Abstract We examine the diversification discount while controlling for differences in information asymmetry between diversified and nondiversified firms. We show that both diversified and nondiversified firms with higher levels of information asymmetry have discounted firm values relative to firms with lower levels of information asymmetry, although a diversification discount remains at all levels of information asymmetry. Fixed-effect Fama-MacBeth regressions confirm the existence of a statistically significant relation between information asymmetry proxies and excess value, but they also show that a significant diversification discount remains after controlling for differences in information asymmetry and other firm characteristics discussed in earlier studies (e.g., size, profitability, leverage, and capital constraint). JEL Classifications: 630, 614 I. Introduction Myers and Majluf (1984) show that in the presence of information asymmetry between the firm and market, firms will forego value-increasing projects rather than go to the external capital market to obtain financing. Williamson (1975) and Stein (1997) argue that diversification should create internal capital markets with lower levels of information asymmetry that allow firms to channel resources to the correct segments and reduce the underinvestment problem. However, most empirical studies such as Berger and Ofek (1995) and Lang and Stulz (1994) conclude that diversification is on average a value-decreasing activity. Other studies such as Servaes (1996) and Berger and Ofek (1996) suggest that diversification should not have occurred in the first place. Furthermore, Comment and Jarrell (1995) and The authors gratefully acknowledge the contribution of IBES Inc. for providing earnings per share forecast data, available through the Institutional Brokers Estimate System. 235

236 The Journal of Financial Research Lane, Vikas, and Ranjini (1997) find that corporate refocusing increases the value of the firm, and Berger and Ofek (1999) interpret such results as evidence that firms were undoing previous mistakes of merger and diversification. Consequently, a body of literature including studies such as Denis, Denis, and Sarin (1997) has developed that attempts to explain the diversification discount. We extend previous research by examining the relation between the diversification discount and the level of information asymmetry for multisegment and single-segment firms. We further explore issues raised by previous authors such as Nanda and Narayanan (1999), who develop an asymmetric information model for divestitures that assumes that the market can observe the aggregate cash flows of the firm but not the individual divisional cash flows, resulting in misvaluation of the firm s securities. Krishnaswami and Subramaniam (1999) find that firms that engage in spin-offs have higher levels of information asymmetry compared with their industry- and size-matched counterparts, and the information problems decrease significantly after a spin-off. They argue that the size of gains around spin-offs is positively related to the pre-spin-off degree of information asymmetry. Although we do not explicitly examine the efficiency of internal capital markets as discussed in Scharfstein and Stein (1997) and Denis and Thothadri (2000), our study has implications for such studies because higher information asymmetry makes it more difficult for investors to assess the efficiency of internal capital markets. Hadlock, Ryngaert, and Thomas (2001) examine a transparency hypothesis that suggests that diversified firms may be harder to value (less transparent) than focused firms. They suggest that accounting figures reported for focused firms are more informative than those for diversified firms because of the aggregate nature of diversified firms accounting reports. This implies that asymmetric information problems should be more severe for diversified firms than for focused firms. In a similar vein, Dunn and Nathan (1999) show that as the level of a company s total diversification increases, analysts earnings forecasts are less accurate and exhibit more interanalyst disagreement. Similarly, Lamont and Polk (2001) provide evidence that conglomerate firms have higher required returns and that this can account for approximately one-third of the observed diversification discount. However, Thomas (2002) reports that more highly diversified firms do not exhibit increased asymmetric information. In this article we use the standard deviation of the residuals of the market model, the standard deviation of financial analysts earnings forecasts standardized by stock price, and the level of institutional ownership as information asymmetry proxies. We calculate Berger and Ofek (1995) excess values for firms grouped into high-, medium-, and low-information-asymmetry categories. Our results show that both diversified and nondiversified firms with higher levels of information asymmetry have discounted firm values relative to firms with lower levels of information asymmetry, although we also find that a diversification discount remains

Diversification Discount 237 at all levels of information asymmetry. We confirm a significant relation between excess value and information asymmetry after controlling for differences in firm characteristics discussed in earlier studies (e.g., size, profitability, leverage, and capital constraint). However, we find that the inclusion of the information asymmetry proxies does not result in a substantial change in the diversification coefficient estimate or its statistical significance. Our results indicate that excess values are significantly related to information asymmetry but that information asymmetry cannot explain the diversification discount. II. Sample Selection and Data Description Sample Selection The Financial Accounting Standards Board s (FASB) Statement of Financial Accounting Standards (SFAS) No. 14 and the Securities and Exchange Commission s (SEC) Regulation S-K required firms to report segment information for fiscal years ending after December 5, 1977. Firms were required to report audited footnote information for segments whose sales, assets, or profits exceed 10% of consolidated totals. The Compustat Industry Segment (CIS) database reports the number of different business segments, up to a maximum of 10, defined by the firm under SFAS No. 14. Additionally, segment information for all Compustat firms other than utility subsidiaries is recorded. 1 We collect date from 1987 to 1998 for all firms in the CIS database and use the method suggested by Berger and Ofek (1995) to complete the data-selection process. We include only firms with no regulated utility (Standard Industrial Classification (SIC) codes 4900 4999) or financial segments (SIC codes 6000 6999). We require that the sum of segment sales must be within 1% of total sales of the firm. This restriction reduces noise because some multidivisional firms do not fully allocate total firm sales to the segments, which occasionally results in major differences between total firm sales and the sum of segment sales. As in most previous studies, our sample is restricted to firms with total sales greater than $20 million. The restriction is imposed to help assure comparability of our findings with those of other studies. However, we find similar results when no minimum sales level is enforced or when all firms with individual segments whose sales are less than 1 Effective with fiscal years beginning after December 15, 1997, FASB replaced SFAS No. 14 with SFAS No. 131. SFAS No. 131 defines business segments based on management s internal organization of the firm, as opposed to SFAS No. 14 s business segment definition that focused on groups of services and products to external customers. Results of SFAS No. 131 were to increase the number of firms reporting themselves as multisegment firms and to increase the average number of segments being reported by multisegment firms. See Street, Nichols, and Gray (2000) for further information and discussion.

238 The Journal of Financial Research $20 million are included. 2 The final sample includes 33,359 firm observations (48,089 segment observations). Calculation of Excess Value We examine the effects of diversification on corporate value by measuring the percentage difference between a firm s total value and the sum of its imputed segment values as in Berger and Ofek (1995). Excess value = ln(v/i(v)), where V is the firm s total capital defined as market value of common equity plus book value of debt, and I(V) is the imputed value of the sum of a firm s segments. The imputed value is calculated as follows: I (V ) = n AI i (Ind i (V/AI) mf ), (1) i=1 where AI i is segment i s sales in that year, and Ind i (V/AI) mf is the industry median ratio of total capital to sales. It is based on the most narrow SIC grouping that includes at least five single-line businesses with at least $20 million of sales and sufficient data for computing the ratios. The imputed values of 53.8% of segments are based on four-digit SIC codes, 28.6% are based on three-digit SIC codes, 14.7% are based on two-digit SIC codes, and 2.9% are not available. To further reduce noise in the analysis, we exclude observations if the calculated excess value is above 1.386 or below 1.386 (i.e., actual values that are either more than four times the imputed value or less than one-fourth imputed value). This restriction reduces the total number of observations by approximately 7%. However, similar results obtain if the restriction is removed and all observations are retained. We report results based on sales as the accounting multiplier in the excess value calculation because, like Berger and Ofek, we find similar results using either assets or earnings before interest and taxes (EBIT) as the accounting multiplier. Descriptive Statistics The final sample includes all firms for which calculation of at least one information asymmetry proxy is possible. The number of observations used in each test varies according to the information asymmetry proxy used. Table 1 contains descriptive information for the final sample of 33,359 firm observations (48,089 segments). The average number of segments per firm for the overall sample is 1.44, 2 Unless otherwise noted, throughout the article, use of the term similar in reference to alternative test specifications indicates that the alternative tests yield similar coefficients and similar levels of statistical significance.

Diversification Discount 239 TABLE 1. Descriptive Statistics for Single-Segment and Multisegment Firms. Characteristic All Firms Single Segment Multisegment No. of observations 33,359 24,994 8,365 No. of segments 1.44 1.00 2.76 [1.00] [1.00] [2.00] Firm sales 1230.47 915.74 2170.84 [189.75] [155.11] [438.93] Firm total assets 1298.30 938.39 2373.67 [167.32] [133.89] [402.70] Segment sales 865.59 915.74 715.74 [157.87] [155.11] [169.53] Segment total assets 900.45 938.39 766.95 [138.11] [133.89] [158.30] Note: Mean [median] values are shown for the 33,359 observations from 1987 to 1998. Sample firms have total sales greater then $20 million and with no segments in utilities (Standard Industrial Classification (SIC) codes 4000 4999) and financial services (SIC codes 6000 6999). Not covered and covered indicate whether analyst forecasts of each firm s earnings are available in the last month of the fiscal year under consideration. TABLE 2. Excess Value and Diversification. Excess Value Difference in Means Sample of Firms No. of Observations Mean Median [Medians] All firms 33,359 1.47 0.00 Single segment 24,994 0.68 0.00 12.39 Single/multisegment [13.86 ] Multisegment 8,365 7.88 8.70 Note: Excess value is measured using the Berger and Ofek (1995) measure. It is calculated as the logarithm of the ratio of the firm s actual value to its imputed value, and the actual value is defined as the market value of equity plus the book value of total debt. Imputed value is calculated by multiplying the sales and the industry median capital-to-sales ratio. The sample includes 33,359 observations from 1987 to 1998. The significance of mean excess value and the difference in means is measured using t-statistics. The difference in medians is tested using the nonparametric Wilcoxon signed-rank test for medians. Significant at the 1% level. and the average number of segments for multisegment firms is 2.76. Average total assets and sales are substantially higher for multisegment firms than for singlesegment firms. Mean segment total assets and sales are smaller for multisegment firms than for single-segment firms, but median values are slightly larger for multisegment firms. Table 2 contains mean and median excess values for the overall sample and the sample firms categorized by diversification. Firms are considered diversified if they report information for multiple segments. The mean excess value is 1.47%, and the median excess value is 0.00% for all firms. The mean (median) excess

240 The Journal of Financial Research value for single-segment firms is 0.68% (0.00%) and the mean (median) excess value for multisegment firms is 7.88% ( 8.70%). Tests indicate that the mean and median excess values for multisegment firms are significantly different from those of single-segment firms. These results are consistent with previous studies that document a diversification discount. III. Information Asymmetry Measures Krishnaswami and Subramaniam (1999) and Best, Best, and Young (1998) show that many commonly used information asymmetry measures are highly correlated. However, to ensure the robustness of our results, we use several information asymmetry proxies. The results reported are based on using the standard deviation of market model residuals, the dispersion of analyst earnings forecasts, and the level of institutional ownership as proxy for information asymmetry. Standard Deviation of Market Model Residuals We first use the standard deviation of the residuals of the market model (SDRES) as a measure of information asymmetry, as in Bhagat, Marr, and Thompson (1985) and Blackwell, Marr, and Spivey (1990). Krishnaswami and Subramaniam (1999) suggest that information asymmetry about a firm is high when managers have a relatively large amount of value-relevant, firm-specific information that is not shared by the market. They indicate that the residual volatility in a firm s stock returns may be used as a proxy for information asymmetry about firm-specific information if the investors in the market and the firm s managers are equally well informed about the systematic factors influencing firm value. We estimate the market model for each calendar year for all firms with sufficient data available on the Center for Research in Security Prices (CRSP) daily stock return database. We then calculate the standard deviation of the residuals of the market model for each year. Within each year, the various standard deviations of residuals are ranked to form percentile rankings. The use of fiscal year percentile rankings of the variable help remove possible calendar year biases. In reported tests, we use the percentile rankings of the standard deviations of market model residuals calculated concurrently with excess value as the measure of information asymmetry. Results are similar when the information asymmetry proxy is lagged one year. Dispersion of Analyst Earnings Forecasts Following Best and Zhang (1993) and Krishnaswami and Subramaniam (1999), we use financial analysts earnings forecasts to develop additional measures of information asymmetry. Studies such as Gonedes, Dopuch, and Penman

Diversification Discount 241 (1976), Givoly and Lakonishok (1979), and Elton, Gruber, and Gultekin (1984) find financial analysts forecasts convey important information to investors. Given the importance of earnings information in security selection and performance evaluation, it is expected that predictions of corporate earnings should provide significant information to the market. Analyst coverage, the number of analysts following a firm, earnings prediction error, and dispersion of analyst forecasts are several of the earnings-forecast-based information asymmetry proxies used in the literature. We report results based on analyst earnings forecast dispersion (EFD), which is measured as the standard deviation of analyst earnings forecasts for the last month of the current fiscal year divided by the concurrent stock price. 3 EFD is calculated for all firms with sufficient data on the Institutional Brokers Estimate System (IBES) database. Similar to SDRES, all of the firms dispersion measures are ranked from lowest to highest within each year to form percentile rankings. These percentile rankings are used as the measure of information asymmetry. EFD is computed simultaneously with the excess value measure because the excess value measures depend on accounting information that is released concurrently with earnings information. We find qualitatively similar results when EFD is lagged one period (i.e., EFD from the previous fiscal year is matched to the current fiscal year excess value). 4 Level of Institutional Ownership Often in the academic literature, institutional investors are described as informed investors who carefully monitor the firms in which they invest. Brickley, Lease, and Smith (1988) show that institutional investors and other blockholders are more involved in voting on antitakeover amendments than are nonblockholders and that institutional opposition is greater when the proposal seems to harm stockholders. Jiambalvo, Rajgopal, and Venkatachalam (2002) find that the extent to which stock prices lead earnings is positively related to the level of institutional ownership. Szewczyk, Tsetsekos, and Varma (1992) find that the absolute magnitude of the share price reaction to new issues of common stock is negatively related to the level of institutional ownership in the announcing firm. Their results are consistent with the argument that the information-acquisition activities of institutional investors reduce information asymmetries between managers and the capital market. Based on these results, we use the level of institutional ownership (IOWN) as an information 3 Brous (1992), Christie (1987), and Pound (1988) discuss the possible merits of normalizing by price per share instead of earnings per share. However, Best, Best, and Young (1998) show that converting variables to percentile rankings reduces the effect of the choice of standardization variable. 4 Because single-segment firms have a lower average stock price than multisegment firms, the forecast dispersion measure could be biased. However, we obtain similar results using the standard deviation or coefficient of variation of earnings forecasts or the analysts percentage prediction error measured as a percentage of the earnings forecast or actual earnings. We thank an anonymous reviewer for this observation.

242 The Journal of Financial Research asymmetry proxy. Consistent with previous research, we hypothesize that higher IOWN reduces information asymmetry. We gather information from the annual CDA/Spectrum Institutional 13-f Common Stock Holdings and Transactions database. The files contain information about common stockholdings and institutional money managers. All institutional managers filing 13-f reports with the SEC are covered. Institutional investment managers controlling a portfolio with an aggregate fair market value of at least $100 million are required to file Form 13-f to report their holdings to the SEC. IOWN is calculated by aggregating the shares owned by the institutions as of the end of the current fiscal year and dividing by the total number of shares outstanding. The percentage of institutional ownership values are then sorted into percentile rankings by calendar year and the percentile rankings are used as the information asymmetry proxy in all tests. We find similar results if IOWN is measured as aggregate number of shares owned by institutions as of the end of the previous fiscal year. IV. Method and Empirical Results Diversification, Information Asymmetry, and Excess Firm Value In initial tests, firms are placed into three groups based on the percentile rankings of each of the three information asymmetry proxies (SDRES, EFD, and IOWN). Low indicates firms with information asymmetry percentile rankings less than or equal to 33, medium indicates firms with percentile rankings between 34 and 66, and high indicates firms with percentile rankings greater than or equal to 67. Firms with low SDRES, low EFD, or high IOWN are viewed as having less informational asymmetry than firms with high SDRES, high EFD, or low IOWN. Table 3 contains the mean and median excess values for the sample firms categorized by both the diversification and information asymmetry variables. Results based on SDRES as the information asymmetry proxy are shown in Panel A. The mean [median] excess value for single-segment firms is 21.57% [ 22.80%] for high SDRES, 6.25% [1.90%] for medium SDRES, and 11.52% [8.20%] for low SDRES. Mean [median] excess values for multisegment firms exhibit a similar pattern (low SDRES = 32.29% [ 37.40]; medium SDRES = 10.16% [ 13.40%]; high SDRES = 1.04% [1.20%]). In all instances, difference-in-mean [median] tests indicate that the mean [median] excess values are significantly different across SDRES categories. Tests also show that the mean [median] excess values are significantly smaller for multisegment firms than for single-segment firms for all levels of SDRES. Panel B of Table 3 contains excess values based on EFD as the information asymmetry proxy. Although the magnitude of the mean [median] excess values for each grouping varies from that found in Panel A, the general pattern of the excess

Diversification Discount 243 TABLE 3. Excess Value, Diversification, and Information Asymmetry. Single Segment Multisegment Difference Difference Difference in Means Mean in Means Mean in Means [Medians] [Median; Obs] [Medians] [Median; Obs] [Medians] (Single/Multi) Panel A. Standard Deviation of Market Model Residuals as Information Asymmetry Proxy High 21.57 32.29 5.64 [ 22.80; 5,845] [ 37.40; 1,145] [6.02 ] High medium 28.75 10.46 [ 24.30 ] [ 8.15 ] Medium 6.25 10.16 12.12 [1.90; 9,589] [ 13.40; 2,229] [12.62 ] Medium low 5.81 7.72 [ 8.74 ] [ 7.88 ] Low 11.52 1.04 10.03 [8.20; 5,503] [1.20; 3,624] [10.84 ] Panel B. Dispersion of Analyst Earnings Forecasts as Information Asymmetry Proxy High 11.00 14.45 2.29 [ 9.50; 4,825] [ 17.10; 1,521] [0.60] High medium 20.23 7.58 [ 18.57 ] [ 5.48 ] Medium 9.84 2.01 9.68 [7.20; 5,371] [ 3.00; 2,117] [7.74 ] Medium low 21.00 11.12 [ 16.14 ] [ 10.22 ] Low 30.27 14.82 11.72 [27.40; 6,255] [14.90; 1,910] [6.73 ] Panel C. Level of Institutional Ownership as Information Asymmetry Proxy High 19.29 4.39 12.56 [15.50; 6,171] [3.40; 2,594] [11.24 ] High medium 17.42 10.77 [13.47 ] [8.62 ] Medium 2.65 12.04 10.80 [0.00; 6,825] [ 13.35; 1,940] [11.07 ] Medium low 14.64 2.86 [14.27 ] [2.36 ] Low 11.41 17.17 3.74 [ 10.90; 7,045] [ 20.80; 1,720] [4.38 ] Note: Excess value is measured using the Berger and Ofek (1995) measure. It is calculated as the logarithm of the ratio of the firm s actual value to its imputed value, and the actual value is defined as the market value of equity plus the book value of total debt. Imputed value is calculated by multiplying the sales and the industry median capital-to-sales ratio. The sample includes 33,359 observations from 1987 to 1998. The significance of mean excess value and the difference in means is measured using t-statistics. The difference in medians is tested using the nonparametric Wilcoxon signed-rank test for medians. Significant at the 1% level. Significant at the 5% level.

244 The Journal of Financial Research values across categories is similar. The mean [median] excess value for singlesegment firms is 11.00% [ 9.50%] for high EFD, 9.84% [7.20%] for medium EFD, and 30.27% [27.40%] for low EFD. The mean [median] excess value for multisegment firms is 14.45% [ 17.10%] for high EFD, 2.01% [ 3.00%] for medium EFD, and 14.82% [14.90%] for low EFD. Other than the median values for high EFD, mean and median excess values for multisegment firms are statistically significantly lower than those for single-segment firms across information asymmetry levels. Panel C of Table 3 contains excess values based on IOWN as the information proxy. The pattern of excess values is inverted relative to that shown in Panels A and B because higher institutional ownership indicates less information asymmetry. The mean [median] excess value for single-segment firms is 19.29% [15.50%] for high IOWN, 2.65% [0.00%] for medium IOWN, and 11.41% [ 10.90%] for low IOWN. The mean [median] excess value for multisegment firms is 4.39% [3.40%] for high IOWN, 12.04% [ 13.35%] for medium IOWN, and 17.17% [ 20.80%] for low IOWN. The mean [median] excess values are significantly different across ownership levels and diversification categories. Overall, the significant differences in mean [median] excess values across information asymmetry categories for all asymmetry proxies indicates that information asymmetry is related to firm excess value. However, the finding that excess values remain significantly different across single-segment and multisegment firms after controlling for the level of information asymmetry suggests that our measures of information asymmetry do not fully explain the diversification discount. Regression Analysis Studies such as Berger and Ofek (1995) and Denis, Denis, and Sarin (1997) suggest that there are factors that may affect excess value but are not necessarily determined by the level of diversification. To control for influences other than diversification and information asymmetry, we estimate various regression models with excess value as the dependent variable. Similar to Denis, Denis and Sarin, we include several independent variables to control for possible biases due to differences in firm size (SIZE), profitability (PROF), and leverage (LEV). 5 SIZE is calculated as the log of total assets, PROF is calculated as EBIT divided by sales, and LEV is calculated as the ratio of total debt to total assets. As suggested by Lang and Stulz (1994), we control for the level of capital constraint of the firm by 5 If diversified firms are valued at a discount because they are less profitable, including profitability as a control variable is overcontrolling in some sense. Regressions estimated without including profitability yield similar coefficients and levels of statistical significance for the remaining variables. We thank an anonymous reviewer for this observation.

Diversification Discount 245 including a dividend dummy variable (DIV) that is equal to 1 if the firm pays a dividend, and 0 otherwise. The effect of diversification is measured by a multisegment dummy variable (MULTI) that is equal to 1 if the firm has multiple segments, and 0 otherwise. Information asymmetry proxies SDRES, EFD, and IOWN are included in various models. When discussed as a group, the information asymmetry proxies are referenced as INFOASYM. The possibility of a differential information asymmetry effect for single-segment and multisegment firms is addressed through the inclusion of interaction variables denoted as MULTI SDRES, MULTI EFD, and MULTI IOWN. The variables are calculated as the product of the respective information asymmetry proxy and the multisegment dummy variable. As a group, the interaction variables are referenced as MULTI INFOASYM. Because our study uses panel data, the same firm may enter the sample multiple times. Thus, observations are not independent, which introduces the problems of potentially downward-biased standard errors and possible unobserved heterogeneity due to cross-sectional differences in organizational structure or segment reporting. To address these concerns, we use a modification of the Fama-MacBeth (1973) method. Specifically, we compute fixed-effects Fama-MacBeth regressions in the manner described in Rajan, Servaes, and Zingales (2000). 6 Except for dummy variables, we first calculate the time-series average for all variables for each firm and use the resulting values to demean firm variables. We next estimate a series of annual cross-sectional regressions including dummy variables and the demeaned variables for each firm. This yields a set of 12 annual estimated coefficients for each variable. We report the average of the annual estimated coefficients and compute statistical significance using the time-series standard deviation of the annual estimated coefficients. Reported significance levels are based on the t-distribution with 11 degrees of freedom. The regression results are shown in Table 4. The DIV coefficient estimate is not significant in any of the models. The SIZE coefficient estimate is significant in two of the three control variable models but is not significant in any model containing the information asymmetry proxies. This result is not surprising given that some previous studies use size as a simple proxy for information asymmetry. The LEV coefficient estimate is negative and significant in five of the six models. The PROF coefficient estimate is positive and significant in all six models. Overall, the findings for the control variables are similar to those of previous studies; 6 We find statistical significance of the various coefficient estimates to be similar using either: (1) ordinary least squares (OLS) with firm fixed-effects and calendar-year dummies, (2) OLS using mean variable values for each firm, or (3) the standard Fama-MacBeth (1973) method. As discussed in Leamer (1978) and Conolly (1989), large sample corrections to the test statistics are employed in all OLS regressions.

246 The Journal of Financial Research TABLE 4. Regression Tests on Excess Value. Variable Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Intercept 2.31 2.21 2.03 1.97 2.26 2.17 (1.17) (1.17) (1.18) (1.28) (1.15) (1.02) SIZE 3.44 1.36 1.76 2.23 3.90 1.94 (2.42 ) (0.94) (1.19) (1.64) (2.43 ) ( 0.96) PROF 75.15 65.43 84.52 63.23 82.53 76.38 (10.78 ) (10.60 ) (9.47 ) (8.66 ) (10.86 ) (10.17 ) LEV 11.68 3.43 24.08 14.61 15.76 9.82 ( 4.90 ) ( 1.60) ( 6.11 ) ( 3.93 ) ( 5.33 ) ( 3.46 ) DIV 0.32 0.52 0.51 0.53 0.07 0.04 ( 0.33) ( 0.53) ( 0.47) ( 0.56) (0.08) (0.04) MULTI 2.48 2.37 1.83 1.78 2.78 2.57 ( 4.33 ) ( 4.53 ) ( 2.31 ) ( 2.33 ) ( 3.79 ) ( 3.24 ) SDRES 0.54 ( 12.22 ) MULTI SDRES 0.18 (5.16 ) [ 9.00 ] EFD 0.39 ( 15.35 ) MULTI EFD 0.12 (3.92 ) [ 8.84 ] IOWN 0.49 (8.27 ) MULTI IOWN 0.14 ( 2.67 ) [19.08 ] Note: Regressions are estimated with percentage excess value as the dependent variable. Excess value is measured using the Berger and Ofek (1995) measure. It is calculated as the logarithm of the ratio of the firm s actual value to its imputed value, and the actual value is defined as the market value of equity plus the book value of total debt. Imputed value is calculated by multiplying the sales and the industry median capital-to-sales ratio. The sample includes 33,599 observations from 1987 to 1998. Fixed effects Fama-MacBeth (1973) methodology is used. First, we subtract the time-series average for each variable for each firm. Next, we estimate a series of cross-sectional regressions with the demeaned variables. SIZE = firm size, calculated as the log of total assets; PROF = profitability, calculated as earnings before interest and taxes divided by sales; LEV = leverage, calculated as the ratio of total debt to total assets; DIV = dividend dummy variable that is equal to 1 if the firm pays a dividend, and 0 otherwise; MULTI = multisegment dummy variable that is equal to 1 if the firm has multiple segments, and 0 otherwise; SDRES = standard deviation of the residuals of the market model; MULTI SDRES = interaction variable that is the product of MULTI and SDRES; EFD = standard deviation of analyst earnings forecasts for the last month of the current fiscal year divided by the concurrent stock price; MULTI EFD = interaction variable that is the product of MULTI and EFD; IOWN = level of institutional ownership; MULTI IOWN = interaction variable that is the product of MULTI and IOWN. The time-series standard deviation of the estimated coefficients is used to compute statistical significance (t-statistics are in parentheses). The t-statistics for testing the statistical significance of the sum of coefficients for variables with multisegment interaction terms are shown in brackets. Significant at the 1% level. Significant at the 5% level.

Diversification Discount 247 thus, we focus our discussion on the diversification and information asymmetry variables. In models 1, 3, and 5, only MULTI is included with the control variables. The control variable model is estimated three times because the number of observations available for each information asymmetry proxy differs. In the three models, the MULTI coefficient estimates are significant and range from 1.83% to 2.78%. This indicates that differences in firm size, profitability, leverage, and capital constraint cannot alone explain the existence of a diversification discount. In models 2, 4, and 6, SDRES, EFD, and IOWN are separately included in the estimated models with their associated multisegment interaction terms. The INFOASYM coefficient estimates from the three models range from 0.39 to 0.54 and all are significant at the 1% level. The results indicate that a 1 percentile increase in the information asymmetry measure results in an approximately.5% decrease in value. However, in each estimated model, the MULTI coefficient estimate retains similar magnitude and statistical significance as found in the control regressions. In models 2, 4, and 6, the MULTI INFOASYM coefficient estimates are significant and of opposite sign from the INFOASYM coefficients. The MULTI INFOASYM coefficient estimates are approximately one-third the magnitude of the INFOASYM coefficients. In each model the sum of the INFOASYM coefficient estimate and the associated MULTI INFOASYM coefficient estimate is statistically different from zero at the 1% level. This indicates that increased information asymmetry results in a larger valuation discount for single-segment firms than for multisegment firms. This result appears to be inconsistent with the hypothesis that multisegment firms face a discount because of information problems that make them more difficult to value than single-segment firms. An alternative explanation is that investors are less confident about information for multisegment firms. Given studies such as Dunn and Nathan (1999) that show that analyst forecasts are less accurate and that there is more interanalyst disagreement as the level of a company s total diversification increases, market participants may consider information about multisegment firms to be less reliable than that for single-segment firms. Thus, additional information about multisegment firms may yield smaller valuation increases than corresponding information about single-segment firms. Overall, the regressions show that information asymmetry is significantly related to firm excess value. However, the estimated diversification discount remains significant in all models regardless of the information asymmetry proxy used. Furthermore, on average, the inclusion of the information asymmetry proxies reduces the magnitude of the estimated diversification discount coefficient by only 5%. This indicates that information asymmetry is related to excess value but it cannot explain the observed diversification discount.

248 The Journal of Financial Research V. Conclusion Most empirical studies conclude that, on average, diversification is a valuedecreasing activity. Although there is strong empirical evidence that diversified firms sell at a discount to nondiversified firms, there is less empirical evidence as to why this discount exists. Previous studies examine differences in firm characteristics such as size, profitability, leverage, and capital constraint across diversified and nondiversified firms. We examine the diversification discount while controlling for the difference in information asymmetry between diversified and nondiversified firms. Group mean [median] excess values show that both diversified and nondiversified firms with higher levels of information asymmetry have discounted firm values relative to firms with lower levels of information asymmetry. However, mean [median] excess values remain significantly different across diversified and nondiversified firms for all levels of information asymmetry. Fixed-effect Fama-MacBeth regressions confirm the existence of a statistically significant relation between information asymmetry proxies and excess value but also show that a significant diversification discount remains after controlling for differences in information asymmetry and other firm characteristics discussed in earlier studies (e.g., size, profitability, leverage, and capital constraint). References Berger, P. and E. Ofek, 1995, Diversification s effect on firm value, Journal of Financial Economics 37, 39 65. Berger, P. and E. Ofek, 1996, Bustup takeovers of value-destroying diversified firms, Journal of Finance 51, 1175 1200. Berger, P. and E. Ofek, 1999, Causes and effects of corporate refocusing programs, Review of Financial Studies 12, 311 45. Best, R. J., R. W. Best, and A. Young, 1998, An examination of proxies of information asymmetry, Advances in Financial Planning and Forecasting 8, 17 34. Best, R. W. and H. Zhang, 1993, Alternative information sources and the information content of bank loans, Journal of Finance 48, 1507 22. Bhagat, S., M. W. Marr, and R. Thompson, 1985, The Rule 415 experiment: Equity markets, Journal of Finance 40, 1385 1401. Blackwell, D., M. W. Marr, and M. Spivey, 1990, Shelf registration and the reduced due diligence argument: Implications of the underwriter certification and the implicit insurance hypotheses, Journal of Financial and Quantitative Analysis 25, 245 59. Brickley, J. A., R. C. Lease, and C. W. Smith Jr., 1988, Ownership structure and voting on antitakeover amendments, Journal of Financial Economics 20, 267 81. Brous, A., 1992, Common stock offerings and earnings expectations: A test of the release of unfavorable information, Journal of Finance 47, 1517 36. Christie, A., 1987, On cross-sectional analysis in accounting research, Journal of Accounting and Economics 9, 231 58. Comment, R. and G. Jarrell, 1995, Corporate focus and stock returns, Journal of Financial Economics 37, 67 87.

Diversification Discount 249 Connolly, R. A., 1989, An examination of the robustness of the weekend effect, Journal of Financial and Quantitative Analysis 24, 133 69. Denis, D., D. Denis, and A. Sarin, 1997, Agency problems, equity ownership and corporate diversification, Journal of Finance 52, 135 60. Denis, D. and B. Thothadri, 2000, Internal capital markets, growth opportunities, and the valuation effects of corporate diversification, Working paper, Purdue University. Dunn, K. and S. Nathan, 1999, The effect of industry diversification on consensus and individual analysts earnings forecasts, Working paper, Georgia State University. Elton, E., M. Gruber, and M. Gultekin, 1984, Professional expectations, accuracy and diagnosis of errors, Journal of Financial and Quantitative Analysis 19, 351 63. Fama, E. and J. MacBeth, 1973, Risk, returns, and equilibrium: Empirical tests, Journal of Political Economy 81, 607 36. Givoly, D. and J. Lakonishok, 1979, The information content of financial analysts forecasts of earnings, Journal of Accounting and Economics 1, 165 85 Gonedes, N., N. Dopuch, and S. Penman, 1976, Disclosure rules, information-production, and capital market equilibrium: The case of forecast disclosure rules, Journal of Accounting Research 14, 89 137. Hadlock. C., M. Ryngaert, and S. Thomas., 2001, Corporate structure and equity offerings: Are there benefits to diversification? Journal of Business 74, 613 35. Jiambalvo, J., S. Rajgopal, and M. Venkatachalam, 2002, Institutional ownership and the extent to which stock prices reflect future earnings, Contemporary Accounting Research 19, 117 45. Krishnaswami, S. and V. Subramaniam, 1999, Information asymmetry, valuation, and the corporate spin-off decision, Journal of Financial Economics 53, 73 112. Lamont, O. and C. Polk, 2001, The diversification discount: Cash flows vs. returns, Journal of Finance 56, 1693 1721. Lane, D., M. Vikas, and S. Ranjini, 1997, Corporate focus and value creation evidence from spin-offs, Journal of Financial Economics 45, 257 81. Lang, L. and R. Stulz, 1994, Tobin s q, corporate diversification, and firm performance, Journal of Political Economy 102, 1248 80. Leamer, E. E., 1978, Specification Searches: Ad hoc Inferences with Nonexperimental Data (Wiley, New York). Myers, S. and N. Majluf, 1984, Corporate financing and investment decisions when firms have information that investors do not have, Journal of Financial Economics 13, 187 221. Nanda, V. and M. Narayanan, 1999, Disentangling needs: Financing needs, firm scope and divestitures, Journal of Financial Intermediation 8, 174 204. Pound, J., 1988, The information effects of takeover bids and resistance, Journal of Financial Economics 22, 207 27. Rajan, R., H. Servaes, and L. Zingales, 2000, The cost of diversity: The diversification discount and inefficient investment, Journal of Finance 55, 35 80. Scharfstein, D. and J. Stein, 1997, The dark side of internal capital markets: Divisional rent-seeking and inefficient investment, Journal of Finance 55, 2537 64. Servaes, H., 1996, The value of diversification during the conglomerate merger wave, Journal of Finance 51, 1201 25. Stein, J., 1997, Internal capital markets and the competition for corporate resources, Journal of Finance 52,111 33. Street, D., N. Nichols, and S. Gray, 2000, Segment disclosures under SFAS 131: Has business segment reporting improved? Accounting Horizons 14, 259 85. Szewczyk, S. H., G. P. Tsetsekos, and R. Varma, 1992, Institutional ownership and the liquidity of common stock offerings, Financial Review 27, 211 25. Thomas, S., 2002, Firm diversification and asymmetric information: Evidence from analysts forecasts and earnings announcements, Journal of Financial Economics 64, 373 96. Williamson, O. E., 1975, Markets and Hierarchies, Analysis and Antitrust Implications: A Study in the Economies of Internal Organization (Collier Macmillan, New York).