The Dynamics of Diversification Discount SEOUNGPIL AHN*

Transcription

1 The Dynamics of Diversification Discount SEOUNGPIL AHN* NUS Business School National University of Singapore Singapore Tel: (65) Current version: June 2007 Preliminary and Incomplete Please do not quote * Author acknowledges generous financial support for the project from Ministry of Education, Singapore and Academic Research Council of National University of Singapore.

2 Abstract Using a sample of diversified firms over the period of , we investigate changes in the diversification discount during the two recent merger waves. The time-series patterns of the diversification discount coincide with the merger waves. The average discount is higher when many firms diversify during the waves and it is eliminated as firms in deep discount exit. We document some evidence that is consistent with the endogeneous self-selection hypothesis in the estimation of the average excess value. Nonetheless, we find that the distribution of excess value is meaningful. Deep discounted firms are more likely to reverse their diversification within shorter time periods while the survival of diversification strategies among premium firms and moderate discount firms is unrelated to their excess values. After accounting for value effects, premium firms perform better than focused firms as well as discount firms. We interpret the results as evidence that (1) diversification is good for some firms, but it is not for other firms, and (2) excess value correctly identify these firms successful and unsuccessful in their diversification.

3 1. Introduction The effect of corporate industrial diversification on value remains controversial and is still not well understood despite numerous studies. Previous research has indicated the existence of a diversification discount that conglomerates are valued at a discount as compared to a portfolio of single segment firms. 1 Two broad views have emerged to explain the diversification discount phenomenon. One view is that industrial diversification per se is value decreasing. The diversification discount hypothesis implies that the measured discounts represent inequilibrium which will be eventually resolved by dismantling diversification. If external and internal disciplinary mechanism works properly, any true value loss associated with diversification will be subsequently eliminated by these disciplinary forces. Berger and Ofek (1996) document that diversification discount increases the possibility of bust-up takeovers. Comment and Jarrell (1995) document a trend toward focus during late 80 s. Kaplan and Weisbach (1992) find that, following large acquisitions, almost 44% of target companies are divested in the later period. Mitchell and Lehn (1990) also find that poorly performing firms are more likely to be the target of acquisitions. Lang and Stulz (1994) notice that the diversification discount is gradually decreased starting from year 1986 through year Nonetheless, they left us with a puzzle why there are still many firms that diversify and remain diversified when diversification leads to lower value. The other view is that diversification does not destroy value and an observed diversification discount due to the mechanics of the calculation of the discount. Whited 1 For examples, see Berger and Ofek (1995), Lang and Stulz (1994), and Rajan, Servaes, Zingales (2001) among others. 1

4 (2001) and Villalonga (2003, 2004) argue that once the measurement error in calculation of excess value is corrected, diversification discount disappears. 2 Campa and Kedia (2002) and Graham, Lemmon, and Wolf (2002) find that diversified firms are not in discount after controlling for a firm s endogenous decision to diversify. This evidence is consistent with the persistence of diversification strategy even today. However, it contradicts the evidence from corporate refocusing literature. 3 Another set of literature indicates that mergers occur in waves in response to industry, macroeconomic, stock market conditional and regulatory shocks. Mitchell and Mulherin (1996) and Andrade, Mitchell, and Stafford (2001) report that merger waves occur in the early 80 s. Moeller, Schlingemann and Stulz (2005) document merger wave in the late 90 s. Shleifer and Vishny (2002) propose a model that the merger waves are driven by stock market run-ups which lead groups of bidders with overvalued stock to purchase target stock, which may be overvalued as well. Merger waves are followed by restructuring waves and a vast amount of assets are reallocated. Is a firm s diversification strategy related to merger waves? Exogenous shocks may drive many firms to diversify. Average diversification discount is high when there are more firms that are unsuccessful in their diversification. Disciplines from internal and external control force these unsuccessful firms to undo their diversification and, once the restructuring is completed, average diversification discount disappears. Firms that are successful in their diversification remain diversified through the restructuring wave. We observe 2 Whited (2001) reports that diversified firms are not in discount under measurement error consistent GMM estimation. Villalonga (2003) sorts out comparison stand-alone firms most closely resembled the segments of diversified firms and find that excess value is not negative when segments are compared with the propensity score matched stand-alone firms. Villalonga (2004) uses establishment data of Business Information Tracking Series (BITS) database and finds that diversified firms are in premium. 3 Refocusing literature including Ahn and Denis (2004), Gertner, Powers, Scharfstein (2002), and Dittmar and Shivdasani (2000) also document that diversified firms are valued less than focused firms and refocusing is value enhancing through the elimination of inefficiency present in diversification. 2

5 diversification discount during merger waves when many firms diversify. The subsequent disappearance of the discount can be understood as a winner-picking process, in which successfully diversified firms remain diversified and failed diversification is disciplined. Therefore, diversification per se does not necessarily destroy firm value. For some diversified firms, the benefits of diversification outweigh the costs of diversification and, for others, it is not. If excess value correctly measures value loss from diversification, excess value can predict a firm s survival as a diversified firm and indicates firms successful and firms failed in their diversification. Using a sample of firms over the period , we look at the explanatory power of the competing hypotheses by investigating changes in diversification discount during two recent merger waves. We find that time series pattern of diversification discount is broadly matched with merger waves. The time series pattern of the diversification discount suggests that refocusing activities in the mid to late 1980s eliminated the aggregate diversification discount existed in the early 80 s. The average value effect of diversification is not different than zero during The disappearance of the diversification discount is consistent with an effective market for corporate control that punishes inefficient diversification. To test whether excess value correctly measures the inefficiency of diversification, we examine the relation between the length of diversification and excess value. If excess value represents value loss from diversification, more discounted diversified firms will remain diversified for shorter periods. We expect that value-losing diversification cannot be sustained under the presence of the internal and external pressure. If discounted firms are forced to refocus, we should expect the disappearance 3

6 of discounted firms and diversification discount after the restructuring waves during late 80 s. We find that more discounted firms tend to reverse their diversification within shorter time period. Using Hazard model, we find that 1% increase in diversification discount is associated with around 20% increase in the probability of refocusing. The result is robust after controlling for leverage, operating performance, investment rate, and firm size. We find even stronger result after removing firms financially constrained and firms having poor performance. In addition, we find that excess value does not predict the survival of focused firms, confirming that excess value measures the value loss from diversification. Why diversification discount reappears in year 1999 and diversification strategies thrive again during late 90 s through early 2000 s? This can be explained in connection with merger waves. Moeller, Schlingemann and Stulz (2005) find that the period from saw shareholders of acquiring firms make losses that are probably higher than in earlier periods. Consistent with their results, we find that the diversification discount, after being eliminated in the period from returns in We think the return of diversification is caused by another set of diversified firms responding to different shocks. The level of diversification discount implies subsequent restructuring wave in the forthcoming years. Finally, we investigate value changes when discount firm remain diversified. We find that after adjusting for the value effect (Lamont and Polk, 2002), premium firms outperform discount firms. The result confirms that excess value correctly distinguishes firms successful in their diversification and firms failed in the diversification. 4

7 The remainder of the paper is organized as follows. In Section 2, we describe our sample selection procedure and explain our measure of diversification discount. Section 4 examines the relation between the length of diversification and excess value. Section 5 examines performance of firms that remain diversified. Section 6 provides some discussion on our findings. Section 7 concludes. 2. Sample selection and estimation of diversification discount 2.1. Sample selection Our initial sample consists of the universe of companies covered by Compustat (including the research file) in both firm level and segment level at any time over the period 1980 to We matched the firm data with the segment data. We obtain segment data from the Compustat Industry Segment (CIS) files over the period of 1981 to According to the change in the segment reporting standard from SFAS No. 14 to SFAS No. 131 in the end of year 1997, we use segment data from the Compustat Operating Segment (COS) files over the period of 1998 to To avoid any distortion effect of really small companies in calculating excess value, we exclude firms with sales revenue less than $20 million. We also exclude firms with any segment operating in the financial industry (SIC code ), those in other regulated utilities (SIC code ), those that are listed as American Depository Receipts (ADRs), and those lacking the required data at the firm level or segment level. 4 Segment reporting in Compustat segment files is subject to managerial discretion in reporting the number of segments. As a result, the number of segments reported could 4 Alternatively, we exclude all firms with segment sales less than $20 million. This reduces our sample size from 8,674 firm-years and 24,400 segment-year to 5,600 firm-years and 15,983 segment-years. Nonetheless, with one exception (noted later) our results are virtually unchanged. 5

8 fluctuate over time even if there is no real change in a firm s organization. To minimize the effect of the managerial discretion in segment reporting, we also require that firms remain as single segment firms or diversified firms for at least two consecutive years. We define diversified firms as those firms reporting at least two segments operating in different three-digit, standard industrial classification (SIC) codes. We define focused firms as those firms reporting only one segment. We exclude any multiple segment firms reported segments operating in the same three-digit SIC codes. 5 Therefore, by definition, diversified firms in our sample consist of pure conglomerates operating in unrelated industries. For those diversified firms, we eliminate segments entitled corporate, allocation, inter-segment allocation, and others. These segments are non-economic activities representing unallocated amount in segment sales, segment assets and segment operating profits. After elimination of the non-economic segments, we exclude diversified firms in which the sum of segment sales deviates by more than 1% from the total sales of the firm. After the selection process, final sample consists of 16,347 firm-year observations of diversified firms and 24,602 firm-year observations of focused firms over the period 1980 to On average, the diversified firms report 2.9 segments with median of 3 segments Estimation of diversification discount We estimate diversification discount with excess value measure. Similar to Berger and Ofek (1995), Lang and Stulz (1994), and Rajan, Servaes, and Zingales (2000), 5 Our main findings are not sensitive to this definition. We find qualitatively similar results if we define industry at either the two-digit or four-digit SIC level. 6

9 we compute excess value as the natural log of the Tobin s q of the firm divided by its imputed q. Tobin s is proxied by market value to sales ratio. Imputed q is calculated as the sales weighted sum of the ratio of market value to sales for single-segment firms in the same industry. Market value of a firm is the book value of assets plus the market value of common equity minus the sum of the book value of common equity and balance sheet deferred tax. We match each segment of a diversified firm with the industry median value of single segment firms. Industry is defined at the 4-digit SIC level provided that there are at least five single-segment firms in the industry. If there are fewer than five single-segment firms in a 4-digit SIC industry, we define industry at the 3-digit SIC level and then 2-digit SIC level. We winsorize excess values as well as all other final variables at the first and 99th percentiles. We make several choices based in calculating excess value. First, we match segments with industry median instead of sales or asset weighted mean. We prefer to use industry median over weighted mean because median is less subject to the size effect. It is possible that the mean value is driven by large and, presumably, more successful single segment firms resulting in a downward bias in the calculated excess value. Second, our sample of diversified firms confined to conglomerate having segments operating at least two different industries and exclude firms with multiple segments in the same industry at the three-digit SIC level. Third, we prefer to use segment sales instead of segment assets or net income. Segment assets tend to be under-allocated across segments and segment net income is subject to earnings manipulation. 7

10 3. Diversification discount and the length of diversification strategy In this section, we provide evidence that diversification discount changes in a systematic way across two recent merger waves. Diversified firms trade at discount when many firms diversify in the waves and the discount is eliminated during subsequent restructuring wave. This dynamic pattern of diversification discount implies that diversification can be understood as a winner-picking process in which successful diversification survives and unsuccessful diversification is eliminated. We also examine whether excess value is associated with a firm s survival time as a diversified firm. It appears that the measure of excess value can correctly predict the sustainability of diversification strategies over time Change in the diversification discount over time Table 1 reports our measure of excess value in each year from year 1980 to We confirm a significant diversification discount overall. Over the entire sample period, diversified firm, on average, trade at discount compared to focused firms: average excess value is with median excess value of However, the magnitude of this discount is changing over time. Figure 1 shows that diversification discount is substantially higher during and during periods, but it is mostly eliminated during period. Similarly, Servaes (1996) finds significant diversification discount during the 60 s, but not during the early and mid 70 s. A trend toward refocusing during the second half of the 1980 s and corresponding decrease of diversification discount are documented by Comment and Jarrell (1995) and Lang and Stulz (1994). Campa and Kedia (2002) also document similar time patterns over the 8

11 period of We complement their findings by adding the return of diversification discount from 1999 to Interestingly, the dynamic pattern of diversification discount coincides with the periods of merger waves during the early 80 s and the late 90 s. 6 Figure 1 also shows that the diversification discount is positively correlated with the number diversified firms existing in the certain time period. 7 Diversification discount increases as many firms diversify during the waves and the discount is relatively low as many firms undo their diversification subsequently. It is well established that mergers tend to occur in waves and the merger waves are followed by restructuring waves through which a vast amount of assets are reallocated. Mitchell and Mulherin (1996), Andrade, Mitchell, and Stafford (2001), and Shleifer and Vishny (2002) argue that the clustering of merger activities is in part driven by exogenous shocks such as macro-economic, industrial, technological, and regulatory changes and stock market run-ups. The coincidence between the time series pattern of diversification discount and the merger waves suggests that firms diversify in response to exogenous shocks and diversification and subsequent restructuring can be understood as a process of massive asset reallocation. Our view on the time patterns in diversification discount is that many firms diversify in a certain period in response to exogenous shocks, followed by exits of unsuccessful diversified firms. Firms successful in their diversification stay diversified, those unsuccessful in their diversification exit, and this cycle repeats when another shock arrives. Things that cannot last, do not. We expect that the firms experiencing value loss from diversification will eventually refocus. 6 Merger waves in the early 80 s are reported in Mitchell and Mulherin (1996) and Andrade, Mitchell, and Stafford (2001). The recent merger waves in the late 90 s are examined by Moeller, Schlingemann and Stulz (2005). 7 The number of diversified firms increases from 586 firms in 1997 to 878 firms in This jump is in part due to segment reporting change from SFAS 14 to SFAS 131 in

12 However, the cross-sectional average diversification discount is temporarily high during the restructuring period, as the elimination of discount firms is delayed. There are, at least, two reasons that the restructuring process is slow and sometimes costly. First, managers of diversified firm may resist against restructuring when doing so is value-enhancing. At the same time, a slow correction by the market for corporate control and inadequate internal monitoring would result in a temporarily higher discount during the merger waves that is corrected over time. The presence of antitakeover defenses may diminish the effectiveness of the external discipline. Jensen (1993) argues that the restructuring through hostile takeovers could take two or three years while the voluntary restructuring through internal control mechanism could take up to ten years in some cases like General Mills restructuring during the 1980 s. Second, compared to the market for financial assets, the market for physical assets is illiquid. Schlingemann, Stulz, and Walking (2000) document that the liquidity of the market for corporate assets is an important factor in determining which segments of a firm are divested. The liquidity of the market for corporate assets might deter a diversified firm from exiting and sometimes it is costly. Pulvino (1998) argues that financially constrained firms are more likely to sell their assets at discounted prices. The liquidity problem could put binding constraints on diversified firms during the merger waves when many firms restructure their assets at the same time. The survival of successful diversified firms and the refocusing of unsuccessful diversified firms are similar to the argument in the value-maximizing neoclassic model of diversification in Maksimovic and Phillips (2002). They, however, implicitly assume that the entire process of diversification and refocusing is, on average, timely and efficient. 10

13 However, if firms reverse their diversification immediately whenever their diversification turns out to be value-decreasing, there will be no time-series fluctuations in the diversification discount. Therefore, we interpret that the observed dynamic patterns in the diversification discount arises from the delay in eliminating unsuccessful diversification. If excess value is informative, we expect that firms with positive excess values (successful diversification) are more likely to stay diversified than firms with negative excess value (unsuccessful diversification). Our focus in this paper is to study whether the measure of diversification discount (excess value) can correctly identify these firms that are successful and unsuccessful in their diversification. This is an important issue in the following reason. People, in general, agree that there are firms who perform well through diversification and firms who do not. In theory, diversification has both potentials that could increase or destroy the firm value. Then, what make the benefits of diversification outweigh the costs of diversification in some firms, but not in other firms? If excess value can be used to identify these firms that are successful and unsuccessful in their diversification, we can answer the question by comparing characteristics of these two types of firms. This is different question from how to interpret the negative average excess value from panel data set, as it has been examined in the debate on the diversification discount. 8 Villalonga (2004) and Whited (2001) argue that the average excess value is estimated with a downward bias and, after correcting for the estimation error, the average excess value is indifferent from zero. Their findings on the location of the average excess value, however, do not necessarily imply that the distribution of excess value is also meaningless. As long as the behavior of premium firms and discount firms (identified 8 Villalonga (2003) provides an excellent summary on the debate. 11

14 with excess value) is different in a predictable way, the excess value is a meaningful measure. Campa and Kedia (2002) analyze the diversification discount in self-selection models which endogenize a firm s decision to diversify and refocus with instrumental variables representing the firm s characteristics, industry conditions, and macroeconomic indicators. They argue that the excess value measure overestimates the diversification discount among diversifying firms and underestimates the discount among refocusing firms. The result suggests that excess value is a nosy measure so that it might misclassify firms as discount firms when their diversification is not value-destroying and it might misclassify firms as premium firms when their diversification is value-destroying. It is an empirical question whether this bias is large enough to make the distribution of the excess value measure to be completely meaningless. In the following analysis, we show that the distribution of excess value is useful measure in identifying firms that are successful and unsuccessful in their diversification Excess value and the decision to exit What causes the dynamic pattern in excess value? To answer the question, we need to explain how the discount during the early and mid 80 s is eliminated and why the discount returns in the late 90 s. If diversification is the result of value-maximizing firm behavior and, therefore, the distribution of excess value is unrelated to the value loss from diversification, firms decision to exit will be independent of excess value because premium firms and discount firms (identified with excess value) have equal probability to refocus when it is optimal to do so. 9 Alternatively, diversification could be viewed 9 Similarly, Campa and Kedia (2002) argue that the time patterns in the diversification discount can be explained by the changes in industry compositions. As low valued firms diversify, the average value of the 12

15 unfavorably by the market during the early to mid 80 s and during the period of , but not during the period of If diversification is viewed unfavorable by the market during certain periods, firms remaining diversified and firms undoing their diversification during these periods are valued similarly by the market. We compare excess values between diversified firms that remain diversified and diversified firms that disappear. Firms disappear from the sample of diversified firms in various cases including being acquired, bankrupt, refocusing, and privatized. We aggregate all these cases to exiting diversified firms. 10 From Table 2, the remaining diversified firms are, on average, performing better than the exiting diversified firms. From panel A., diversified firms that remain diversified for the entire period of are discounted at , while firms disappeared during the periods are substantially more discounted at Similarly in panel C, exiting firms are more discounted than remaining firms over the periods. The result indicates that the diversification discount in the early and mid 80 s are eliminated as deeper discount firms exit slowly from the sample of diversified firms and moderate discount firms and premium firms stay diversified. Therefore, the time series pattern of excess value is not caused by the changes in market remaining focused firms increases. This exit of poor performers increases the imputed value of diversified firms and leads to higher discount during certain periods. However, their argument is more nuanced instead of directly testing how these changes in the industry composition are related to the year-by-year changes in average excess value. For example, in 1989, the mean excess value of focused firms is highest at while the excess value of diversified firms is highest at 0.00 when it is expected to be lowest under their argument. 10 Managers have discretion in reporting the number of segments. Some firms could report one segment in one year and multiple segments in the next year even if there are no changes in the firm s organization. In the multivariate analysis, we exclude cases that diversified firms disappear due to the managerial discretion in segment reporting. 13

16 perception on diversification. It is also inconsistent with the prediction of the endogenous self-selection story that excess value is unrelated to the firms exit decision. The middle period of our study, , is characterized by a relatively quiet period regarding mergers and refocusing activities. This sub-period fits well with the market perception story and endogenous self-selection story. Excess values appear to be uninformative to the choice to exit during this time period. The excess values of firms that choose to remain diversified are very similar to firms that choose to exit. The mean diversification discount for both the firms that remain diversified and the firms that choose to refocus are around half (around 5% discount) of average discount over the entire period. Note, however, that this quiet period comes only after the restructuring waves in the mid 80 s. It suggests that, as deeper discount firms are eliminated in the previous periods, remaining firms do not experience the value loss from diversification and firms exit decision among these remaining firms are driven by other reasons that are independent from diversification discount. Therefore, when diversified firms trade at relatively moderate discount, excess value is not informative. However, when diversified firms are in deeper discount (greater than 5% discount in our data), excess value appears to be an important measure associated with the exits of diversified firms Excess value and the length of diversification Firm age is widely used in the finance literature to measure the quality of a firm. 11 Intuitively, a long existing strategy is synonymous to the viability of that strategy. If everything else is equal, firms successful in their diversification will maintain their diversification strategies for longer periods. In this section, we investigate the 11 For examples, see Ritter (1984), Michaely and Shaw (1994), and Chemmanur and Paeglis (2005). 14

17 relationship between excess value and the length of diversification. On average, diversification is short lived: average length of diversification is 4.7 years. We find that there are relatively few firms that maintain diversification strategies for long periods. For example, from , there are only 198 firms that are diversified over the entire period. In contrast, there are 1,032 firms that were diversified at some point during These firms that remain diversified for at least 10 years, have a relatively small diversification discount of less than 4%. In table 3, excess values of the remaining firms and the exiting firms are compared according to the years of firms diversification. Firms having lower discount tend to remain diversified for longer time periods. Firms remaining diversified at least for the two consecutive years have average discount of (median of -0.09) while firms remaining diversified at least for the six consecutive years have average discount of (median of -0.05). It appears that the discount below 4% does not have a marginal impact on the survival time as a diversified firm. Figure 2 shows diversification discount and the length of diversification are negatively related. The relationship becomes flat at around 4% of diversification discount. It seems at the odds that remaining firms are also significantly discounted at around 4% (around half of average discount over the entire period). If negative excess value implies the value loss of diversification, should we expect remaining firms have diversification discount of 0%? Graham, Lemmon, and Wolf (2004) argue that this magnitude of diversification discount can be explained by biases in the excess value measure. Lamont and Polk (2004) argue that about half of the diversification discount is due to the difference in expected returns between diversified firms and focused firms. 15

18 Our findings complement this view and suggest that the location of average excess value could be biased downward. Nevertheless, our result indicates that the distribution of excess value is informative: diversification discount greater than 4% is a relevant number in predicting the survival of a firm s diversification strategy. Why excess value is associated with the survival of diversification strategies? As we argued previously, the relationship can not be fully explained by measurement errors in calculating excess value. It is, however, possible that the correlation between excess value and firm characteristics may affect our findings. To address the concern, in the following sections, we provide multivariate tests of the relationship between diversification discount and the length of diversification strategies controlling for endogeniety issue as well as firm s financial characteristics. 4. Duration analysis of diversification 4.1. Excess value and Hazard rate What we are interested in is cases of exits that are driven by the loss from diversification and whether these cases are related to excess value. Some diversified firms may exit with reasons other than diversification discount. For example, poor performance that is not directly related to the value loss from diversification may affect the firm s decision to exit. If such firm characteristics are correlated with excess value, our finding that excess value is related to the firm s exit decision is spurious. It is, however, practically hard to isolate cases that are driven by diversification discount. In the reason, we aggregate all exit cases in univariate tests, but in the multivariate analysis, we control for firm characteristics that could affect firms decision to exit. 16

19 Previous descriptive analysis of the relationship between the length of diversification and excess value is formally tested with a duration analysis, where the duration of interest is survival time of a firm s diversification strategy. Key element of duration analysis is the estimation of hazard rate. 12 The hazard rate, in our models, is the conditional likelihood that a diversified firm drops out of the sample of diversified firms in time period t, given that the firm has survived through time (t-1). We use Cox (1972) s Proportional Hazard Model, which does not require to estimate baseline hazard function as we do not have prior knowledge on the shape of hazard function that affects all diversified firms in reversing their diversification strategies. Cox Proportional Hazard model analysis can accommodate the impact of time varying explanatory variables in panel data settings. The model can also address the right censoring problem. Right censoring occurs when a firm does not have a completed spell within the data observation window. In our data set the survival or diversified firms is right censored at year We require diversified firms stay in the sample at least for two consecutive years. In this way, we exclude firms changing the number of segments due to the managerial discretion in segment reporting. In table 4, explanatory variables are measured annually and they are time-variant. Dependent variable is survivor time as a diversified firm. DISCOUNT is -1 times excess value. Estimated coefficient represents the risk of failure in the next period. For example, coefficient of 1.2 indicates that 1 unit increase in diversification discount increases the hazard by 20%. Coefficient of 0.7 indicates that 1 unit increase in diversification discount 12 See J. M. Wooldrige, Econometric analysis of cross section and panel data, The MIT Press, 2002, pp for more detailed explanation for the hazard model. 17

20 decreases the hazard by 20%. If diversification discount does not predict the failure of diversification strategy, we expect the coefficient on DISCOUNT equals to 1. We include a set of control variables: leverage, sales margin, investment rate and size (sales revenue) in excess forms. EXLEV is excess leverage and it is measured as the difference between the firm s ratio of total debt to the book value of total assets and the firm s imputed leverage. Imputed leverage for each segment is equal to the leverage of the median focused firm operating in the same 3-digit SIC industry. If there are fewer than five focused firms in the industry, industry matching is performed at the 2-digit level. Imputed leverage for the firm is then equal to the sales weighted sum of leverage for the firm s segments. EXSM is excess sales margin and it is measured as the difference between the firm s ratio of EBIT to sales revenues and the firm s imputed sales margin. EXINV is excess investment rate equals to the difference between the firm s net capital investment divided by sales and the firm s imputed investment rate. EXSales is excess size of a firm measured by the difference between the firm s sales revenues and the firm s imputed sales revenues. From model (1) of table 4, the coefficient of DISCOUNT is 1.11 and it is significant. It indicates that the increase in diversification discount by 1% is associated with 11% increase of the probability to exit in the next year. EXLEV is also an important factor in determining the length of diversification; higher leverage carries higher hazard to terminate diversification. The result is consistent with the hypothesis that debt serves a disciplinary role. We also find that more profitable and larger firms tend to stay diversified for longer periods. Investment rate is not related to the survival time as a diversified firm. 18

21 In model (2) (4), we include lagged variables to further control for any endogenous effects associated with the observed diversification discount. The inclusion of lagged DISCOUNT variable alleviates the concern that diversification discount is lowvalued firm phenomenon. The sample size drops to 10,806 observations with one lag and to 8,433 observations with two lags. If diversification discount is low-valued firm phenomenon and, for unknown reasons, the low valued firms are more likely to flip their diversification, we expect that the decision to exit will be more affected by initial discount rather than current discount. Therefore, coefficient on lagged DISCOUNT is expected to be greater than 1 and coefficient on current DISCOUNT is expected to equal to 1. The relationship between the current excess value and hazard rate is stronger. From model (2), after controlling for the previous year s diversification discount, current year s discount is associated with two times higher probability of exit; hazard rate of 22%. Coefficient on lagged DISCOUNT is insignificant 0.91 (hazard rate of -9%). Given that a firm s initial discount, further increase in diversification discount is associated with higher hazard to exit. Although firms might add segments in the current year, it is not clear whether the addition increases diversification discount. Lang and Stulz (1994) and Denis, Denis, and Sarin (1997) argue that the most of diversification discount occurs when a firm diversifies from one segment to two segments and further increase in diversity does not exacerbate diversification discount. The hazard rate associated with diversification discount increases to around 26% in model (3) with second lagged control variables. Parametric estimation of hazard function in model (4) provides similar result. The result confirms our previous findings in the descriptive analysis that excess value correctly predicts which diversified firms 19

22 undo their diversification. As excess value decreases, the probability that a diversified firm reverses its diversification strategy increases. Our result is consistent with the findings of Berger and Ofek (1996) who examine 334 takeovers occurred during period and find that the probability of being targets of hostile takeovers is positively related to diversification discount. We analyze more extended time period ( ) and include all possible reasons for disappearance of diversified firms including being targets of hostile takeovers. In addition, by relating diversification to the merger waves, we provide one possible explanation for the persistence of diversification discount until recent years even after the hostile takeover periods during the 80 s Robustness tests In this section, we discuss some of concerns in interpreting the results in table 4. Although we identify that excess value is related to the length of diversification, it is possible that excess value is correlated with poor performance that is unrelated to the value loss from diversification per se. If poor performing firms are more likely to be disciplined and these firms tend to have lower excess value, the observed relationship between excess value and the length of diversification is spurious. Similarly, financially constrained firms are more likely to restructure their organization and, therefore, could affect our result. To examine whether our result is driven by poor performing firms and financially constrained firms, we re-estimate Cox model after removing firms having (1) interest coverage ratio below 1, (2) Return on Assets lower than bottom 30% in each year, and (3) Kaplan-Zingales Index within top 30% in each year. Kaplan-Zingales Index is a measure of a firm s financial constraints. We follow the specification in Lamont, Polk, 20

23 and Saa-Requejo (2001) to compute the index. Financially constrained firms are designed to have higher scores in the index. From model (1)-(3) in table 5, the coefficients on DISCOUNT are around 1.4, suggesting an increase in diversification discount of 1% increases the possibility of refocusing by amazing 40%. Another test is whether the predictability of excess value is unique to diversified firms. If diversification discount is something to do with diversification, excess value will be useful measure for the survival of diversified firms, but not for focused firms. In model (4), we calculate excess value for focused firms. Because there is only one segment in a focused firm, excess value of focused firms is basically industry-adjusted q. Excess value does not work for focused firms. In model (4), coefficient on DISCOUNT is close to 1, indicating that excess value is not related to the exits of focused firms from the sample of focused firms. Leverage and sales margin are significantly related to the survival of focused firms. The result confirms that excess value captures the value loss associated with diversification itself. Finally, we divided sample of diversified firms into two sub-groups based on the magnitude of discount and re-estimate the Cox models. Descriptive analysis in table 2 and table 3 show that excess value is related to the length of diversification only for deeper discount firms having discount of greater than 5%. This result holds in multivariate analysis settings in model (5) and (6). Coefficient on DISCOUNT is 1.7 among deeper discount firms, but it is insignificant 1.0 among firms with lower than 5% discount. It is possible that low valued firms are more likely to diversify and easy to flip their diversification than premium firms. Matsusaka (2001) argues that low valued firms are in search of a match for the firm s organizational capabilities and abandon their core 21

24 business when they find good matches. The argument, however, is inconsistent with the negative coefficient on DISCOUNT among low-valued firms (discount firms) in model (5); if the diversification by discount firms is equally optimal, why lowest-valued firms are more likely to undo their diversification than less low-valued firms. The result is more consistent with the argument that deep discount represents value loss from diversification and, therefore, deeper discount firms are more likely to reverse their diversification. It is also consistent with agency view that managers tend to delay restructuring unless there is a crisis. In model (6), coefficient on DISCOUNT is insignificantly different from 1. It suggests that the magnitude of gains from diversification does not affect firms exit decision, as long as premium firms have gains from diversification. In sum, tests in table 5 confirm that the relationship between excess value and the length of diversification is not driven by spurious correlation between excess value and poor performance or financial constraints, and it is unique to diversified firms having deeper discount. 5. Performance of premium and discounted firms In the previous sections, we show that excess value is useful measure in predicting the survival of diversification strategy. As we discuss in the previous section, this result, however, does not directly imply that diversification itself destroys value. We rather interpret the result as evidence that (1) diversification is good for some firms, but it is not for other firms; (2) excess value correctly distinguishes successful and unsuccessful diversified firms. 22

25 If premium firms are successful diversified firms and discount firms are failed diversified firms, we expect that premium firms perform better than discount firms when they remain diversified. We compare the performance of these two types of firms when they stay diversified. Lamont and Polk (2002) show that at least a part of the diversification discount is attributable to higher expected returns earned by discount firms. Therefore, it is important to adjust for the differences in expected returns before we compare the performance of these two types of diversified firms. We confirm the existence of the value effect in excess value. In an unreported test, we find that our measure of excess value is negatively related to future changes in excess value among diversified firms. Table 6 reports that this value effect of excess value holds for singlesegment firms as well. We sub-grouped focused firms into 20 portfolios based on their year-by year excess value. For lowest excess value portfolio, the performance over the subsequent two years is positive 0.31 while highest excess value portfolio performs worst at In our effort to understand the changes in excess value over time, we define a measure of the excess change in excess value (ΔEV EXCESS ) that adjusts for this value effect. The excess change in excess value from period t to period t+2 is defined as the difference between the change in excess value for a diversified firm minus the change in excess value for a matched focused firm over the same period. 13 For a given calendar year, focused firms are grouped into twenty portfolios according to the level of excess value at the base year. Diversified firms are also grouped into twenty portfolios according to the excess value cutoff excess value of focused firm portfolios. The average 13 We also examine Excess Changes in Excess Value over the (t, t+1) and (t, t+3) periods. Our inference remains the same. We find stronger result as the time interval increases; outperformance of premium firms are stronger as the time interval increase. 23

26 changes of excess value for the twenty portfolios are assigned to the corresponding diversified firm. Since the value effect is known to be strongly associated with the level of value on the base year, ΔEV EXCESS can, by design, capture the net change in excess value that can not be explained by the value effect. Table 7 reports the change in excess value over two consecutive years. In the first column, diversified firms experience negative change in excess value of on average, while matched focused firms have positive change in excess value of The excess change in excess value (ΔEV EXCESS ) is for the entire diversified firms. In the next two columns, it turns out that premium firms outperform focused firms and discount firms underperform focused firms. ΔEV EXCESS among premium firms is 0.02 while it is among discount firms. The difference in ΔEV EXCESS between premium firms and discount firms are statistically significant at 1% level. This result confirms that premium firms are those successful in their diversification strategies and continue to perform better than focused firms by remaining diversified. Discount firms, however, continue to perform poorly as they remain diversified. This result is consistent with the findings in Maksimovic and Phillips (2002). They find that the surviving diversified firms grow efficiently across industries in which they operate while the growth of the refocusing firms is inconsistent with the neo-classic model of value-maximizing firms. We complement their findings by relating excess value to the survival of diversified firms: these surviving diversified firms that are successful in their diversification can be identified with excess values. The poor performance of discount firms holds as long as they remain diversified regardless of the changes in their diversity. Table 8 reports ΔEV EXCESS for premium firms 24

27 and discount firms according to the change in diversity during subsequent years. The change in diversity is measured with the change in Herfindahl index over the two consecutive years once matching is done. 14 A firm is classified as focus increasing when Herfindahl index increases and otherwise it is grouped as focus decreasing firm. Premium firms perform as good as focused firms even if they diversify further. Discount firms, however, performs poorly even if they increase focus. The result suggests that poor performance of discount firms cannot be eliminated unless they are completely dismantled. 6. Discussion In previous sections, we document that the cross-sectional average diversification discount shows the dynamic patterns corresponding to the merger/restructuring waves. There are two issues related to the efficiency of the entire process. First issue is whether the initial decision to diversify is ex-ante efficient. As long as we are unable to know the exact value of firms when they did not diversify, our result is not supportive of certain agency view that diversification is ex-ante inefficient as it is often driven by managerial objectives at the costs of shareholders wealth. This agency view, however, contradicts the return of the diversification discount and the re-increase in the number of diversified firms in the late 90 s. If diversification is value-decreasing, diversified firms and the discount should be completely eliminated once and for all. As Gomes and Livan (2004) argue, some agency models fail to explain why diversified firms exist at all when it is exante inefficient. 14 The result is quantitatively the same when we measure the change in focus with the number of segments and standard deviation of segment q. 25

28 This agency view is also in odds with bidder announcement return from diversification programs. 15 The agency view predicts that the diversification discount will be eliminated as all diversified firms are forced to be dismantled. Maksimovic and Phillips (2002), however, argue that the growth rate and investment of the remaining diversified firms are consistent with value-maximizing firm behavior. Although there is evidence that some diversification is driven by managerial objectives and, as a result, it is forced to refocus, the number of these cases seems small relative to the entire universe of diversified firms. Our result does not contradict with the claim that initial decision to diversify is efficient on average. In our view, although the ex-ante decision to diversify is efficient on average, some diversification could be ex-ante inefficient or turn out to be a failure expost. Nevertheless, these unsuccessful firms could hold up to their value-losing diversification strategies unless they are forced to refocus. Intuitively, it is harder to reverse existing strategies once assets are already in place than do not make mistakes in the first place. If the restructuring process on the failed diversification is timely, we would not observe the fluctuation of diversification discount over time. The dynamic pattern of diversification discount in Figure 1 suggests that the correction process is slow and value-losing diversified firms may stay for a while until they are disciplined eventually. Although endogeneous self-selection hypothesis recognizes that agency models works well in some cases, it implicitly assumes that the correction process is 15 Bidder announcement returns for diversifying acquisitions are positive during the 60 s (Matsusaka (1993) and Hubbard and Palia (1998). The evidence on bidder returns is at most mixed for the 80 s (see Morck, Shlefer and Vishny (1990), Kaplain and Weisbach (1992), Hyland (1999), and Graham, Lemmon, and Wolf (2002)). 26