A Cross-Firm Analysis of the Impact of Corporate Governance on the East Asian Financial Crisis. Todd Mitton *

Transcription

1 A Cross-Firm Analysis of the Impact of Corporate Governance on the East Asian Financial Crisis Todd Mitton * Marriott School Brigham Young University Abstract In a sample of 398 firms from Indonesia, Korea, Malaysia, the Philippines, and Thailand, firm-level differences in variables related to corporate governance had a strong impact on firm performance during the East Asian financial crisis of Significantly better stock price performance is associated with firms that had indicators of higher disclosure quality (ADRs and auditors from Big Six accounting firms), with firms that had higher outside ownership concentration, and with firms that were focused rather than diversified. The results suggest that individual firms have some power to preclude expropriation of minority shareholders if legal protection is inadequate. JEL classification: G15, G32, G34 Key words: Financial crises, corporate governance, disclosure, ownership structure, diversification * I am grateful to Simon Johnson, Sendhil Mullainathan, David Scharfstein, and Jeremy Stein for advice and encouragement, and to Simeon Djankov, Kristin Forbes, Ken French, Kathy Kahle, S.P. Kothari, Grant McQueen, Andrei Shleifer, Keith Vorkink, Marc Zenner, an anonymous referee, and seminar participants at MIT, Brigham Young University, the University of Illinois at Urbana-Champaign, the University of Pittsburgh, and Texas A&M University for helpful comments. I thank Simeon Djankov for making data available that is used in Panel C of Table 3. This is a revised version of a chapter of my MIT dissertation. All errors are mine. Todd Mitton, BYU Marriott School, 673 TNRB, Provo, UT 84602, Phone: (801) , Fax: (801) , todd.mitton@byu.edu.

2 1. Introduction Weak corporate governance has frequently been cited as one of the causes of the East Asian financial crisis of While weak corporate governance may not have triggered the East Asian crisis, the corporate governance practices in East Asia could have made countries more vulnerable to a financial crisis and could have exacerbated the crisis once it began. Recent research has highlighted the importance of corporate governance in emerging markets. La Porta, Lopez-de-Silanes, Shleifer, and Vishny (LLSV 1997, 1998, 1999, 2000) demonstrate that, across countries, corporate governance is an important factor in financial market development and firm value. Regarding the East Asian crisis, Johnson, Boone, Breach, and Friedman (JBBF 2000) show that country-specific measures of corporate governance perform better than standard macroeconomic variables at explaining the extent of currency depreciation and stock market decline of emerging markets during the crisis. If corporate governance was a significant factor in the crisis, then corporate governance should explain not just cross-country differences in performance during the crisis, but also crossfirm differences in performance within countries. In this paper, I use firm-level data from the five East Asian crisis economies of Indonesia, Korea, Malaysia, the Philippines, and Thailand to study the impact of corporate governance on firm performance during the crisis. Because the measures of legal protection emphasized in LLSV (1997, 1998, 1999) and JBBF (2000) are country-specific, I examine other aspects of corporate governance that vary at the firm level. I show that the three aspects I examine, disclosure quality, ownership structure, and corporate diversification, all had a significant impact on the stock price performance of firms during the crisis. Because the crisis was, by all accounts, an unexpected event, it presents an interesting 1 Examples include Stiglitz (1998), Harvey and Roper (1999), and Greenspan (1999). 1

3 opportunity to study the proximate effect of corporate governance on firm performance during a period of extreme distress. Corporate governance is the means by which minority shareholders are protected from expropriation by managers or controlling shareholders. Corporate governance could become more critical in a financial crisis for two reasons. First, expropriation of minority shareholders could become more severe during a crisis. JBBF (2000) argue that a crisis can lead to greater expropriation because managers are led to expropriate more as the expected return on investment falls. Second, a crisis could force investors to recognize and take account of weaknesses in corporate governance that existed all along. Rajan and Zingales (1998) argue that investors ignored weaknesses of East Asian firms while the region was doing well economically, but quickly pulled out once the crisis began because they believed the region lacked adequate institutional protection for their investments. For both of these reasons, firms with weaker corporate governance could have lost relatively more value during the crisis. Anecdotal evidence from the East Asian crisis suggests that expropriation of minority shareholders was prevalent. One example occurred in November 1997 when United Engineers Malaysia (UEM) acquired 32.6% of its financially troubled parent, Renong. UEM minority shareholders interpreted this as a bailout of Renong at an inflated price, and UEM s stock price fell 38% the day the transaction was announced (Straits Times 11/19/97, p. 62). Another example comes from Korea where minority shareholders of Samsung Electronics protested that the firm had been providing debt guarantees to less-successful Samsung group companies and that these guarantees often were not disclosed (The Economist 3/27/99, p. 68). JBBF (2000) document other instances of expropriation of minority shareholders during the crisis, and Johnson, La Porta, Lopez-de-Silanes, and Shleifer (2000) describe different forms that expropriation can take. This paper considers whether the presence of firm-level characteristics 2

4 related to corporate governance can help prevent such instances of expropriation and, in turn, preserve firm value during a crisis. The first firm-level characteristic that I study, disclosure quality, is an important element of corporate governance. LLSV (1998) argue that accounting standards play a critical role in corporate governance by informing investors and by making contracts more verifiable. While LLSV (1998) and JBBF (2000) employ country-specific measures of accounting standards, I propose two firm-specific ways in which disclosure quality can be measured. First, a firm may have higher disclosure quality if it has a listed American depository receipt (ADR). This higher disclosure quality can emerge formally, through mandated disclosure requirements of the listing exchange (for level II and III ADRs), or informally, through a larger pool of investors spurring increased demand for disclosure and increased scrutiny of the firm s reports (see Coffee, 1999). Reese and Weisbach (2001) argue that increased protection of minority shareholders is a primary motivation for non-u.s. firms to cross-list in the U.S. (see also Stulz, 1999). Second, a firm may have higher disclosure quality if its auditor is one of the Big Six 2 international accounting firms. Previous research (e.g. Reed, Trombley, and Dhaliwal, 2000; Titman and Trueman, 1986) has associated Big Six auditors (or Big Eight auditors, for earlier years) with higher audit quality. The Big Six firms may be more likely to ensure transparency and eliminate mistakes in a firm s financial statements because they have a greater reputation to uphold (Michaely and Shaw, 1995), because they may be more independent than local firms, and because they face greater legal liability for making errors (Dye, 1993). Additionally, even in cases in which actual disclosure quality is not higher, Big Six auditors may offer higher 2 Six major accounting firms remained at the outset of the crisis as the Price Waterhouse/Coopers & Lybrand merger did not occur until late

5 perceived disclosure quality and allay investors fears because of their prominent, recognizable names (see Rahman, 1998). I find that these proxies for higher disclosure quality are associated with significantly better stock price performance during the crisis period (July 1997 to August 1998). Regression analysis shows that having an ADR is associated with a higher return of 10.8% over the crisis period, and having a Big Six auditor is associated with an additional higher return of 8.1% over the crisis period (after controlling for size, leverage, country, and industry). While alternative interpretations (discussed later) are possible, this finding is consistent with the view that higher disclosure quality benefits minority shareholders by increasing transparency and mitigating expropriation during a period of distress. The second aspect of corporate governance that I study is ownership structure. I first consider levels of ownership concentration. Shleifer and Vishny (1997) argue that ownership concentration is, along with legal protection, one of two key determinants of corporate governance. Large shareholders can benefit minority shareholders because they have the power and incentive to prevent expropriation. On the other hand, large shareholders can themselves engage in expropriation. La Porta, Lopez-de-Silanes, and Shleifer (1999) find high degrees of ownership concentration in firms from countries with relatively poor shareholder protection, and argue that the conflict between large shareholders and minority shareholders is the primary corporate governance problem in such countries. Morck, Strangeland, and Yeung (2000) and Bebchuk, Kraakman, and Triantis (2000) discuss how controlling shareholders may pursue objectives that are at odds with those of minority shareholders. Consistent with the view that large shareholders can prevent expropriation, I find that higher ownership concentration is associated with significantly better stock price performance during the crisis. Regressions show a higher return of 2.6%, on average, for every increase of 4

6 10% in the ownership of the largest shareholder (after controlling for size, leverage, country, and industry). This result suggests that the crisis amplified the pre-crisis valuation premium for emerging market firms with large blockholders reported by Lins (2000). Still, large shareholders could be more likely to pursue objectives that are inconsistent with those of minority shareholders if they are involved with management of the firm or if their voting rights exceed their cash flow rights (Claessens et al., 2000). I find that the return premium associated with higher ownership concentration is largely attributable to large blockholders that are not involved with management. I also find significantly lower returns during the crisis for firms in which the largest shareholders voting rights exceed their cash flow rights and for firms with pyramidal ownership structures, although the significance disappears after controlling for other factors. The third aspect I study, corporate diversification, is not a corporate governance mechanism per se, but previous research has suggested that agency problems are different within diversified firms. The lower transparency of diversified firms in emerging markets results in a higher level of asymmetric information that may allow managers or controlling shareholders to more easily take advantage of minority shareholders (see Lins and Servaes, 2000; Lins, 2000). If expropriation of minority shareholders increases during a crisis period, then the associated loss in firm value could be particularly pronounced for diversified firms. While diversification can also offer the benefit of improving capital allocation (Stein, 1997), particularly in emerging markets (Khanna and Palepu, 2000), this benefit could virtually disappear in a time of crisis as investment opportunities diminish. I find that corporate diversification is associated with significantly worse stock price performance during the crisis. Regressions show that, on average, diversified firms had lower returns of 7.6% over the crisis period (after controlling for size, leverage, country, and industry). This result builds on the finding of a pre-crisis diversification discount in Asian emerging 5

7 markets by Lins and Servaes (2000) and by Claessens et al. (1999a), who also find that this discount widened during the crisis. One way in which diversified firms could dissipate value during the crisis is by inefficiently supporting distressed industries with resources from relatively stable industries. That is, inefficient transfer of resources across divisions (Scharfstein and Stein, 2000; Rajan, Servaes, and Zingales, 2000) could become severe if some divisions are hit harder by the crisis and are inefficiently propped up to survive. Consistent with this possibility, I find that the loss in value for diversified firms is almost entirely attributable to diversified firms that have a high variation in investment opportunities across divisions. Taken together, my results reinforce the claim that corporate governance had a significant effect on firm performance during the East Asian crisis. The results are important because they add to our understanding of the causes of the crisis and demonstrate a link between corporate finance and macroeconomic events. But perhaps more importantly, the results suggest that individual firms, and not just countries, have some control over the level of protection offered to minority shareholders. La Porta, Lopez-de-Silanes, and Shleifer (1999) suggest that if a country s legal environment fails to prevent expropriation of minority shareholders, then firms may opt into legal regimes that are more protective of minority shareholder rights. They cite ADR issuance as an example of this phenomenon. The results in this paper support the viability of opting for better protection of minority shareholders. Whether through higher disclosure quality, improved transparency, a more focused corporate organization, or more favorable ownership structure, minority shareholders can be offered protection beyond their legal rights. To some degree, firms are not hostages to the legal regime of their home country. The next section describes my data and methodology. Section 3 reports the main results. Section 4 presents the results of robustness tests and analyzes alternative interpretations. Section 5 reports the evidence on firm performance following the crisis. Section 6 concludes. 6

8 2. Data and methodology 2.1. Sample selection The countries studied in this paper are Indonesia, Korea, Malaysia, the Philippines, and Thailand, the five countries that were most involved in the East Asian financial crisis. Although other East Asian countries (and other emerging markets outside of Asia) were affected by the crisis, the five considered here suffered disproportionately in terms of stock market decline and currency depreciation (see Table 1). All firms from these five countries are included in the sample provided that they meet three criteria. First, each firm must have financial data reported in the Worldscope database, which is the primary data source used in this study. Second, the primary business segment of each firm must not be in financial services, that is, not in standard industrial classification (SIC) Finally, each firm must be identified in Worldscope as being included in the International Finance Corporation (IFC) global index. 3 The IFC includes firms in the global index only if they are among the largest and most liquid firms in a given market. This criterion reduces the sample size, but it is imposed for three reasons. First, the quality of data available in Worldscope is higher among the firms followed by the IFC. For example, non-ifc firms would be three to four times as likely to be excluded from my regressions because of missing data points. Second, stock prices are essential to this study, and the IFC liquidity requirement reduces the probability that a firm s quoted stock price will be uninformative. A review of the data confirms that stocks of non-ifc firms are more likely to be illiquid (as evidenced by repeated 3 Firms are included if they are added to the IFC global index on or before the IFC s 1997 review. Although this review occurs in October 1997, a firm s inclusion is based on performance during the prior year, so I assume that firms added in 1997 met the standards for inclusion prior to the beginning of the crisis. 7

9 monthly prices). Third, in some cases the corporate governance decisions considered in this paper may be more relevant for larger firms. The clearest example is that the decision of whether to issue an ADR is not as relevant for smaller firms because the cost of doing so is often prohibitive (La Porta, Lopez-de-Silanes, and Shleifer, 1999). The sample selection process is outlined in Table 1. The final sample consists of 398 firms from the five crisis countries. In general, the sample is representative of larger firms that trade on the major stock exchange of each country. Small listed firms and other unlisted firms, including large multinationals with no local listing (which can make significant contributions to GDP) are not represented in the sample. Table 1 shows that Korea has the most firms in the sample, with 144, and the Philippines has the fewest, with 29. The median size of firms, in terms of total assets, also varies, with Korea having the largest (a median size of over $1.45 billion) and the Philippines the smallest (a median size of over $316 million). Table 1 presents other summary statistics of firms by country and correlation coefficients of key variables Definition of crisis period Fig. 1 shows the movement of composite stock indexes for all five countries from 1995 through Lines on the chart delineate the crisis period as defined in this paper. The beginning of the crisis period corresponds to the devaluation of the Thai baht on July 2, 1997, a date generally considered to be the starting point of the crisis. The July beginning point also corresponds to the date when all five indexes began moving downward together. As Fig. 1 shows, before this time some, but not all, of the indexes had been trending downward (see Section 4.1 where I analyze an earlier starting point). The ending point of the crisis period, August 1998, corresponds with the date on which the indexes began a sustained upward trend. 8

10 2.3. Variable descriptions Dependent variable To measure firm performance during the crisis I use stock returns over the crisis period, from July 1997 through August The returns are dividend inclusive and are expressed in local currencies adjusted for local price index changes. I do not calculate abnormal returns using historical betas because data limitations prevent the calculation of pre-crisis betas for many firms. As an alternative, I use measures of leverage and size, industry dummies, and country dummies in the regressions to control for factors that could affect expected returns. 4 Table 1 shows the average return by country for the crisis period Explanatory variables To measure disclosure quality I use two variables. The first is a dummy variable that is set to one if the firm had an ADR listed in the U.S. at the beginning of the crisis and zero otherwise. Firms with ADRs are identified using a comprehensive listing of ADRs from the Bank of New York. Firms with all types of ADRs are included. 5 The second variable is a dummy variable that is set to one if the firm is audited by one of the Big Six international accounting firms and zero otherwise. I identify the names of auditors using data from Worldscope. Because I hypothesize that name recognition of Big Six auditors by investors is 4 Pre-crisis betas can be calculated for about 80% of the firms if a minimal requirement of 24 monthly pre-crisis observations is imposed. In regressions using this subsample of 80% of the firms, beta has no significant explanatory power for returns once size, leverage, and industry are included as control variables. 5 The results are even stronger for firms with level II and III ADRs, but these types of ADRs are rare in these countries (only five in this sample). 9

11 essential, I do not include auditors that do not carry a Big Six name, even if the local firm has an affiliation with a Big Six firm. 6 To measure ownership concentration, I use data reported by Worldscope, which identifies all parties that own 5% or more of each firm. This data set has limitations in that it does not incorporate indirect shareholdings, does not indicate divergence between cash flow rights and voting rights, and does not indicate if a listed shareholding is jointly owned by separate parties. I alleviate these problems by matching my data set, where possible, with data compiled by Claessens et al. (2000), which separately indicates cash flow rights and voting rights. The ownership data I use are pre-crisis data, which means the last reported data from each firm prior to July Data are missing for some firms in Worldscope, in which cases I supplement the data with information from the Asian Company Handbook (1998) and the Corporate Handbook: KLSE Main Board (1998) where possible. Given the data limitations, I identify ownership concentration for 301 of the 398 firms in the sample (75.6%). I consider two measures of ownership concentration. The first is the ownership percentage (in terms of cash flow rights) of the largest shareholder in the firm, which I refer to as largest blockholder concentration. The second is the total holding of all shareholders that own 5% or more of the stock, which I refer to as summed ownership concentration. To determine which blockholdings are held by individuals involved with management, I compare a list of officers and directors in each firm (compiled from Worldscope and the above- 6 The names of the Big Six are Arthur Andersen, Coopers & Lybrand, Deloitte & Touche, Ernst & Young, KPMG Peat Marwick, and Price Waterhouse. None of the Korean firms in my sample have Big Six auditors (see Table 1). Clearly the Big Six have a presence in Korea, but the major Korean accounting firms have Korean names, even if they have some affiliation with a Big Six firm. The results for the Big Six variable are virtually unchanged if Korea is excluded from the regressions. 10

12 mentioned handbooks) with the list of significant owners in each company. If the name of an officer matches the name of an owner, this ownership block is classified as managerial ownership. (Thus the term managerial here implies that an individual is involved with decision making within the firm, and not necessarily that the individual is hired as an outside professional.) This name matching procedure is not exhaustive, but it identifies a subset of managerial blockholdings that are the most transparent. In some cases the true owner of a particular block could be obscured if the owner places the block under the name of another individual or company. I also draw on the data compiled by Claessens, et al. (2000) to evaluate the impact of voting rights of the largest shareholder as compared to cash flow rights. I match their data set with mine, and rely on their measures of cash flow rights and voting rights as well as a dummy variable indicating whether firms are controlled through a pyramidal ownership structure. To measure corporate diversification, I determine the number of industries in which each firm operates, with industries being defined at the two-digit SIC level. The SIC codes are reported by Worldscope, generally from pre-crisis data, but using later data if pre-crisis data are unavailable. I use product segment data from Worldscope and other sources to determine what percentage of each firm s sales corresponds to each two-digit SIC code. The first diversification variable is a multiple-segment indicator that is set to zero if 90% or more of a firm s sales come from one two-digit SIC, and one otherwise. The second variable is the number of industries in which the firm operates. Worldscope reports a maximum of five industries per firm, so this variable could be truncated for some firms. Because many firms from these countries are affiliated with corporate groups, the question arises as to whether firms that are reported as diversified are stand-alone firms with multiple business segments or group-affiliated firms with consolidated balance sheets reflecting 11

13 the activities of a number of different firms. A review of the types of firms in the sample and their accounting practices suggests three reasons why firms that are reported as diversified should generally be interpreted as being diversified and not just group affiliated. First, in this sample, diversified firms are no more likely to be affiliated with groups (as defined by Claessens et al., 2000) than are single-segment firms. 7 Second, while most firms in the sample do report consolidated balance sheets (at least for significant subsidiaries), the percentage of firms with consolidated balance sheets is almost as high among single-segment firms (76%) as among diversified firms (81%). Third, as discussed in Section 4.1, despite some overlap between diversification and group affiliation, a diversification indicator has strong explanatory power for firm performance during the crisis, whereas a group-affiliation indicator has very little explanatory power Control variables I use other variables to control for factors that could affect firm performance. The first is firm size, measured by the logarithm of total assets. Using total assets as a measure of firm size could be problematic if different countries in the sample have varying standards for reporting the cost basis of investments on their balance sheet. I find that each country has some firms that uses strictly historical cost basis and some firms that use some type of market revaluation. The exception is Korea, where all sample firms use historical cost. To address this potential bias, I also use net sales as an alternative size measure. An additional control variable is the firm s debt ratio, measured as the book value of total debt divided by the book value of total capital. These data are reported by Worldscope. I 7 Using a different sample of firms, Claessens et al. (1999b) find a significantly larger fraction of diversified firms among group firms in two of the five countries studied here. 12

14 include dummy variables for four of the five countries included in the regressions to control for country fixed effects. I also include dummy variables for ten of eleven industries, where industries are defined broadly, as in Campbell (1996). By including leverage as a control variable, I am potentially making it more difficult to detect the effects of weak governance. Specifically, weak corporate governance could have been correlated with higher debt levels prior to the crisis (see Friedman and Johnson, 2000), so poor stock price performance attributed to leverage could also be partially caused, indirectly, by weak corporate governance. Still, leverage is included as a control variable because higher debt naturally leads to lower stock returns in a downturn, although Forbes (2000) does not find strong evidence of this during the crisis Econometric issues A number of econometric issues in my regression analysis need to be addressed. Multicollinearity does not appear to be a problem in the model. With all key variables included in the model, the average variance inflation factor is 2.6 (with a maximum of 5.8), which is not unreasonably high. I correct for heteroskedasticity using robust standard errors. I test for omitted variables using two versions of the Ramsey test, one using powers of the fitted values of the dependent variable and one using powers of the independent variables. With all key variables included in the model, I fail to reject the null hypothesis that the model has no omitted variables at the 95% confidence level using both tests. Nevertheless, even though formal tests detect no omitted variables, visual inspection of residual plots suggests some remaining evidence of omitted variables in the model. The remaining pattern in the residuals disappears if returns are converted to logarithmic returns. When I repeat the regressions using logarithmic returns (results not reported), the coefficients on all key variables increase in 13

15 magnitude and retain their significance. This robustness check suggests that my reported results are conservative and not driven by omitted variables. I also consider the potential influence of errors in variables. The variables for which measurement reliability might be a concern would be the ownership variables, the diversification variables, and the size and leverage variables. I perform sensitivity tests to evaluate how low the measurement reliability of these variables could fall before the regression results on the full model would change materially. I find that the results are not particularly sensitive to measurement error. Any of the variables in question can fall to measurement reliability of 0.85 (indicating a 15% measurement noise to total variance ratio) before the results are materially affected (and some reliabilities can fall much lower). The results hold even if all variables in question have reliability of 0.85 simultaneously. Another issue is potential endogeneity of the regressors in the model. If the corporate governance variables I study are not exogenous, then their estimated coefficients may not be consistent, due to simultaneity bias, and inferences about the direction of causality of the variables may not be clear. The exogeneity of ownership variables, in particular, could be in question, as others (e.g. Demsetz and Lehn, 1985) have shown that ownership and firm value may be jointly determined. To some extent, the exogeneity of the disclosure quality variables could be in question as well. I address the issue of endogeneity in three ways. First, I note that concerns about endogeneity should be reduced because the East Asian crisis was an unexpected event, and (with few exceptions) I measure all variables in the model on a pre-crisis basis. Second, for the ownership variables, I am able to check my results with an instrumental variables approach. This approach is discussed in Section 4.1. Third, lacking a suitable instrument for the disclosure quality variables, I examine whether firms that opted for better disclosure prior to the crisis would have expected to have had more stable stock prices in the future (based on past 14

16 experience). I find that on average, firms with Big Six auditors had higher betas than non-big Six firms in the two years preceding the crisis (among firms for which data availability permits calculation of beta). Firms with ADRs had lower betas, but no lower than expected for firms of comparable size. These results are not entirely conclusive, but they suggest that firms did not elect to have ADRs or Big Six auditors based on expectations of having more stable stock prices. A final econometric issue is that errors across firms may not be independent because returns are correlated in calendar time. As a diagnostic measure to address this issue, I run simulated regressions of the actual return data on a wide variety of randomly generated hypothetical variables. In 10,000 repetitions, I find that the coefficients on the hypothetical variables are significant at the 1% level 1.1% of the time, at the 5% level 5.3% of the time, and at the 10% level 10.3% of the time. The lack of spuriously significant coefficients suggests that correlation of errors is not a serious problem in the data. 3. Results To assess the impact of corporate governance variables on firm stock price performance during the crisis, I estimate the following model: Crisis Period Return = a + b 1 (Corporate Governance Variables) + b 2 (Size) + b 3 (Leverage) + b 4 (Country Dummies) + b 5 (Industry Dummies) + e, (1) in which the corporate governance variables included change according to the specification, and the other variables are as defined previously Disclosure quality and firm performance Table 2 presents the results of regressions of crisis-period stock returns on measures of disclosure quality. The first two columns include the ADR indicator (with and without controls 15

17 for size and leverage), the second two columns include the Big Six auditor indicator, and the final two columns include both variables. All columns include country and industry fixed effects. The final column of Table 2 shows that the coefficient on ADR is after all controls are included. The magnitude of the coefficient indicates that firms with ADRs had, on average, a higher return of 10.8% over the crisis period. The coefficient on ADR is significant at the 5% level. The coefficient on Big Six auditor is with all controls included. The magnitude of the coefficient indicates that firms with Big Six auditors had, on average, an additional higher return of 8.1% over the crisis period. The coefficient on Big Six auditor is also significant at the 5% level. The results are economically significant as well. The higher returns attributed to firms with higher disclosure quality seem even larger in light of the fact that firm values declined, on average, almost 70% over the crisis period. For example, if a non-adr firm declined 70% over the period and an ADR firm declined 59.2% (10.8% higher), then by the end of the crisis period, these firms would be valued at 30% and 40.8% of pre-crisis values respectively. This amounts to a 36% post-crisis premium for ADR firms relative to non-adr firms when compared to their pre-crisis valuations. In other words, retaining an additional 10.8% of pre-crisis value amounts to retaining 36% of post-crisis value. This measurement is important because it reflects the valuations of investors at the bottom of the stock market decline. At this point, investors placed a very high premium on firms that had opted for higher disclosure quality. The disclosure quality results should be interpreted cautiously for two reasons. First, firms with ADRs and Big Six auditors could have unmodeled characteristics other than higher disclosure quality that affect their returns. These potential alternative explanations are considered in Section 4.2. Second, the lack of a valid instrument for these variables could leave some question about the direction of causality. Nevertheless, as noted previously, the crisis was 16

18 an unanticipated event, and firms opted for ADRs and Big Six auditors before (sometimes many years before) the crisis began. And when these choices were made it doesn t appear that firms opting for higher disclosure quality would have expected to have more stable stock prices (based on past data) Ownership structure and firm performance Table 3 presents the results of regressions of crisis-period stock returns on ownership structure variables. As noted in Section 2.3.2, these regressions are based on a subsample of 301 firms because of data limitations. Panel A presents the results for levels of ownership concentration (measured as cash flow rights). The first two columns analyze the largest blockholder concentration. With all control variables included, the coefficient on largest blockholder concentration is This indicates that each increase of 10% in ownership concentration is associated with a higher return of 2.6% during the crisis. The coefficient on largest blockholder concentration is significant at the 1% level. The second two columns of Panel A analyze summed ownership concentration. With all control variables included, the coefficient on summed ownership concentration is 0.174, indicating a higher return of 1.7% for each increase of 10% in ownership concentration. This coefficient is significant at the 5% level. These results indicate that the presence of a strong blockholder was beneficial during the crisis, consistent with the hypothesis that a strong blockholder has the incentive and power to prevent expropriation of minority shareholders. In Panel B of Table 3, I differentiate between ownership blocks held by individuals involved with management and blocks held by others. The first two columns include blockholdings of those involved with management (with and without controls for size and leverage), the second two columns include nonmanagement blockholdings, and the last two 17

19 columns include both types of blockholdings. The coefficient on management blockholdings is positive when all controls are included, but it is insignificant in all specifications. The coefficient on nonmanagement blockholdings is slightly higher than the coefficient on general blockholdings (in Panel A) and is significant at the 1% level. The difference in coefficients for management and nonmanagement holdings indicates that the value of a large blockholder is greater during a crisis when that blockholder is not involved with management. This result is consistent with the idea that if blockholders are involved with management they could have more opportunity or incentive for expropriation of minority shareholders. In Panel C of Table 3, I differentiate between cash flow rights and voting rights of the largest shareholders. As mentioned previously, I draw on data from Claessens et al. (2000), which is available for 311 of my sample firms. The first two columns show the difference in coefficients when large blockholdings are measured as cash flow rights or voting rights. The coefficient is slightly higher when measured as voting rights, but the two coefficients are very similar in magnitude and significance. The second two columns analyze a dummy variable that is set to one if a firm has a divergence between the cash flow rights and voting rights of the largest owner. Consistent with the idea that cash flow/voting rights divergence increases the incentive for expropriation, the coefficient on this variable is negative. However, it is not significant once all control variables are included. Claessens et al. (1999b) find a significant negative effect on firm value prior to the crisis associated with cash flow/voting rights divergence. The lack of significance of this variable in my regressions suggests that there was no incremental loss of value during the crisis for firms with this divergence. The difference in explanatory power attributed to this variable could also be due to sample differences (Claessens et al, 1999b, cover nine countries) or methodological differences (Claessens et al, 1999b do not use country fixed effects; the third column of Panel C shows that the variable is significant in my 18

20 regressions if country fixed effects are omitted). The final two columns of Panel C analyze a dummy variable that is set to one if the firm has a pyramidal ownership structure. Consistent with the idea that such structures increase the likelihood of expropriation of minority shareholders, the coefficient on this variable is negative, but it is significant only when size and leverage controls are not included Corporate diversification and firm performance Table 4 presents the results of regressions of crisis-period stock returns on diversification variables. 8 The first two columns include the diversified indicator (with and without controls for size and leverage). With all controls included, the coefficient on diversified is 0.076, which indicates that diversified firms, on average, had a lower return of 7.6% over the crisis period. The coefficient on diversified is significant at the 1% level. The second two columns include diversification measured as the number of industries per firm. The coefficient on this alternative diversification measure is also negative and significant at the 1% level. These results are consistent with the hypothesis that the reduced transparency of diversified firms offers greater opportunities for expropriation of minority shareholders. Valuations declined much less for firms that had opted for a focused structure prior to the crisis. In the final two columns of Table 4, I distinguish between diversified firms that have high and low degrees of variation in investment opportunities across operating segments. I create a measure of the diversity of investment opportunities similar to that used in Rajan, Servaes, and 8 I also calculate the diversification discount for firms in the sample using methodology similar to Berger and Ofek (1995). Consistent with my regression results and with Claessens et al. (1999a), I find that the diversification discount widened significantly during the crisis period. However, I do not report these results because reliable estimates of the discount require a larger number of sample firms. 19

21 Zingales (1999). I use the market-to-book ratio of each firm as a proxy for Tobin s q to indicate the level of investment opportunity. I match each segment of each diversified firm to the industry median market-to-book ratio for single-segment firms in each industry in each country. Industries are defined at the two-digit SIC level, but if no single-segment firms are available for a particular two-digit SIC in a particular country, I use broader industry classifications as defined in Campbell (1996). If no match is available at all, I use the countrywide median market-to-book ratio as a fill-in. To measure variation in investment opportunities I take the standard deviation of the matched market-to-book ratios for all segments in the firm. Diversified firms with variations above (below) the median for all diversified firms are designated as having high (low) variation in investment opportunities. In Table 4, I then interact this measure with the diversification indicator. The final two columns of Table 4 show that the negative coefficient on diversified firms is much stronger among those firms that have high variation in investment opportunities. This result is consistent with the idea that diversified firms lose value if segments that are relatively stable are used to inefficiently support segments that are hit particularly hard by the crisis. 4. Robustness and alternative interpretations 4.1. Robustness checks Table 5 presents the results from additional regressions to test the robustness of the results presented in Section 3. I present each robustness check twice (except in Panel D), without ownership variables and with ownership variables (which reduces the sample size). All columns include both disclosure quality variables, diversification interacted with variation in investment opportunities, and all control variables used previously. In Panel A, I leave all variables unchanged from the previous tables to assess whether results presented separately in 20

22 previous tables are manifestations of the same effect. Panel A shows little collinearity among the results, as the disclosure quality, ownership, and diversification variables change very little in magnitude and retain their significance when included together. 9 The one exception is the coefficient on Big Six auditor, which, while still significant in the full sample, is not significant in the reduced sample. In Panel B of Table 5, I change the definition of the crisis period to begin on May The motivation for this robustness check can be seen in Fig. 1, which shows that the stock markets of Thailand and Korea were declining steadily well before July 2, 1997, a date usually considered to be the start of the crisis. Panel B shows that the results are robust to measuring returns over this longer period. Although the magnitude of some coefficients declines somewhat, all key variables retain significance. In Panel C of Table 6, I measure returns in U.S. dollars. The results are robust to this change as well. The magnitude of the coefficients declines, but again all key variables retain significance. In Panel D of Table 6, I address the potential endogeneity of ownership structure with an instrumental variables approach. To create an instrument for the percentage holdings of the largest nonmanagement blockholder, I use 1994 data on the percentage holdings of the largest blockholders from Worldscope. I find the holdings of the largest blockholder for only a subset of the firms in my sample, because Worldscope coverage was not as extensive in Although the coefficients should be interpreted cautiously because of the relatively small number of firms per country, I repeat the model on each individual country and find that the coefficient on ADR ranges from (Philippines) to (Indonesia; this is the only instance where a coefficient is of the unexpected sign, Malaysia is the next lowest at 0.042), Big Six Auditor ranges from (Indonesia) to (Malaysia), Diversified*High Variation ranges from (Thailand) to (Korea), and Largest Nonmanagement Blockholdings ranges from (Indonesia) to (Korea). 21

23 Ownership percentages from 1994 are highly correlated with pre-crisis percentages. This approach assumes that past values of the explanatory variables are uncorrelated with the error term in the crisis-period regressions. Other examples of this approach include Schmukler and Vesperoni (2000) and Lins (2000). Panel D shows that the results from the instrumental variables regression are similar to the results from the ordinary least squares regressions. The coefficient on largest nonmanagement blockholder increases to 0.341, and is significant at the 5% level. In other tests not reported, I find that the results are robust to using alternative measures of firm size, including the square and cube of total assets and the log values of those measures. The results are also robust to measuring size as net sales, the square and cube of net sales, and the log values of those measures. In an analysis of outliers, I find that the results are robust to truncating the data at the 1 st and 99 th percentiles of the return, size, leverage, and ownership variables. In a test of the robustness of the diversified variable, I find that the magnitude and significance of the coefficient on diversified remain the same if I also include a dummy variable indicating group affiliation. This suggests that the diversification of operating segments drives the results on diversified firms, not their affiliation with corporate groups Alternative interpretations One alternative interpretation of the results is that disclosure quality, ownership structure, and diversification are proxies for other characteristics that affect firm returns. Firm size, leverage, and industry are three possibilities that have been controlled for in the regressions, but other possibilities remain. One firm characteristic that my variables could potentially be proxies for is the degree to which firms conduct business internationally. Firms with a higher proportion of sales to foreign countries would be insulated from the crisis to the extent that sales are to 22

24 countries that did not experience relative currency depreciation. To test this possibility I use a variable constructed as the percentage of a firm s foreign sales divided by total sales. I am able to test this variable only on a subset of firms, because Worldscope reports this information for only 230 of the firms in my sample. In regressions using this subsample (not reported), I find that the percentage of foreign sales has no explanatory power for returns during the crisis, whether the variable is used alone or with control variables. The lack of significance of this variable indicates that firms with a high percentage of international business are not driving the results. In addition, disclosure quality, ownership structure, and diversification could be correlated with other variables that have been shown to be correlated with firm returns. One possibility is a firm s book-to-market ratio (Fama and French, 1992). This variable is available for about 95% of the firms in my sample. The book-to-market ratio is insignificant in all regressions, and is not reported. As discussed previously, a firm s beta also has no explanatory power for returns in the subset of firms for which a pre-crisis beta can be calculated (once firm size, leverage, and industry are included as control variables). Another alternative interpretation of the results is that disclosure quality, ownership structure, and diversification always affect stock returns in these countries, and that their significance is not specific to the crisis period. If the importance of these variables during the crisis is due to an increased risk (or awareness) of expropriation of minority shareholders, then the variables should not have as great an impact in pre-crisis periods. To address this issue, I repeat the regressions of corporate governance variables on returns for two pre-crisis years. (I do not look at earlier periods because limitations on Worldscope s historical data begin to greatly reduce the available sample of firms prior to 1995.) Panel A of Table 6 shows the results of regressions of returns from the pre-crisis period of July 1995 to June Panel B shows the 23

25 results for the pre-crisis period of July 1996 to June The number of observations declines in each earlier year due to data limitations. Very few strong patterns are evident in the coefficients in Panels A and B. In some cases the sign of the coefficients is opposite of the sign during the crisis period, but in other cases the sign is the same. None of the corporate governance variables in the pre-crisis regressions is significant at standard levels. The results in Panels A and B suggest that the impact of these variables during the crisis was not the continuation of effects that existed prior to the crisis. Nor does the performance of these variables during the crisis appear to be a reversal of abnormal returns prior to the crisis. A final important alternative interpretation to consider is that firms with ADRs (or perhaps, those with Big Six auditors) perform better during the crisis because these firms have better access to capital. Lins, Strickland, and Zenner (2000) argue that access to external capital markets is an important benefit for non-u.s. firms listing their stock in the U.S. One indicator of whether firms with ADRs benefited from better access to capital is to look at how investment levels changed for firms during the crisis. As would be expected, investment levels fell for most firms during the crisis. Between the last pre-crisis year (fiscal year-ends before July 1997) and the mid-crisis year (fiscal year-ends between July 1997 and June 1998) the median drop in capital expenditures for all firms in the sample is 39.3%. The declines for firms with ADRs and without ADRs were fairly similar. The median decline in capital expenditures for firms with ADRs was 36.1%, and the median decline for firms without ADRs was 39.5%. The difference is even smaller between Big Six and non-big Six firms. The decline in capital expenditures divided by sales was 18.0% for firms with ADRs and 20.0% for firms without ADRs. The median decline for Big Six firms was greater than for non-big Six firms. The relatively small differences in investment declines between these groups (none of which are statistically significant) suggest that differences in access to capital were not a large factor in performance 24