CORPORATE FRAUD AND LIQUIDITY

Transcription

1 CORPORATE FRAUD AND LIQUIDITY Emre Kuvvet Nova Southeastern University H. Wayne Huizenga School of Business and Entrepreneurship 3301 College Avenue Fort Lauderdale-Davie, Florida Phone: Abstract We examine the implications of corporate fraud on stock liquidity. We find that the revelation of corporate fraud has an adverse effect on liquidity not only in the short term but also in the long term. We also find that the complexity and severity of fraudulent activities do not further reduce liquidity; rather, the existence of fraud is what matters to investors. Our results also suggest that liquidity is more adversely affected for fraudulent firms with weak governance than for those with better governance. JEL Codes: G10; G12; G340 1

2 I. Introduction Corporate fraud reduces shareholders' wealth for the fraudulent firm (Karpoff, Lee, and Martin 2008a) and its rival firms (Goldman, Peyer, and Stefanescu 2012) while increasing the cost of capital (Lin, Song, and Sun 2012). In this study, we identify a new channel through which fraud harms shareholders' wealth: reduced liquidity. Former Securities and Exchange Commission (SEC) Chairman Arthur Levitt commented that "Quality information is the lifeblood of strong, vibrant markets. Without it, investor confidence erodes. Liquidity dries up. Fair and efficient markets simply cease to exist." (Remarks to The Economic Club of New York, New York City, October 18, 1999.) It is possible that fraud can increase the transaction costs associated with trading a stock in the stock market. The revelation of fraud can decrease the credibility of firms' financial reporting and increase informational asymmetry. But evidence of the effect of fraud on liquidity is scarce. We fill this gap by analyzing the implications of corporate fraud for both short-term and long-term liquidity. One of the ways to align the financial incentives of executives with those of shareholders is through executive compensation programs. However, the prior literature extensively shows that executive compensation provides managers the incentive to commit fraud. Johnson, Ryan and Tian (2008) found that firms are more likely to commit fraud if the size of managers equity ownership is high. Beneish (1999) showed that managers manipulate earnings in order to sell their stocks and exercise option holdings. Burns and Kedia (2006) suggested that the sensitivity of the CEO's option portfolio to stock price is positively related to the propensity to misreport. Similarly, Bergstresser and Philippon (2006) found that firms are more likely to manipulate their earnings when the CEO compensation is more closely related to the value of stock and option holdings. During the commission of fraud, the fraudulent activities are perpetuated and hidden so well that there is little information asymmetry among traders. At the trigger event, it is first revealed to the public that fraud 2

3 may have been perpetuated. During this period, the magnitude of fraudulent activity is still unknown. It is also unclear when the truth behind the fraud will be determined. Since investors are trying to value the implications of fraud, the impending penalties, and its impact on reputation, there is an increase in uncertainty about the firm s value. The revelation of fraud suggests that information previously disclosed by companies to investors is wrong; therefore, prior beliefs about firms' riskiness have to be reassessed. Kim and Verrecchia (1994) suggested that less-accurate disclosures provide better opportunity for certain traders to process public information into private information, thus increasing information asymmetry among traders and thereby decreasing liquidity. This increase in uncertainty would lead to an increase in the cost of liquidity. In addition, there will be a decrease in management's credibility and a loss of trust by investors. Jarrell and Peltzman (1985) and Klein and Leffler (1981) showed that corporate fraud harms firms reputations by changing the conditions of trade for investors and thus causes higher uncertainty about companies' future prospects. Milgrom and Roberts (1982) and Kreps and Wilson (1982) suggested that corporate fraud can increase the information asymmetry identified by investors due to loss of trust. As a result, investors might not participate in equity markets for fear that they would lose out to better informed traders and insiders, thus reducing liquidity (Velikonja 2013). Together, the studies cited above suggest that firms convicted of corporate fraud may have higher spreads. To the best of our knowledge, this is the first study to examine the effect of corporate fraud on liquidity. We compare fraudulent firms' liquidity before and after fraud incidents. We use the Karpoff et al. (2008a, b) database to classify corporate fraud from 1993 through This database identifies the first public revelation dates for frauds. The database allows us to focus on liquidity during the relevant time periods. Our liquidity measures come from Vanderbilt's Market Microstructure database. We use daily observations of liquidity measures from 1993 until

4 We find that the revelation of corporate fraud has an adverse effect on liquidity not only in the short term but also in the long term. In the short run, the quoted spread (effective spread) increases by 0.138% (0.108%) after the revelation of fraud. In the long term, the quoted spread (effective spread) increases by 0.152% (0.098%) after the revelation of fraud. The fraudulent activities differ in terms of complexity and the amount of harm caused to shareholders. Thus, we also examine the impact of the severity and complexity of fraud on liquidity. The results suggest that increased complexity and severity of fraudulent activities does not further reduce liquidity; rather, the existence of fraud is what matters to investors. Apart from analyzing the effect of fraud on liquidity, we also study the interaction effect of corporate fraud and governance on liquidity. We examine whether firms with weak corporate governance have more severe liquidity problems after the revelation of fraud than those with better corporate governance. Our results suggest that the liquidity of fraudulent firms with weak governance is more adversely affected than the liquidity of those with better governance. These results highlight the importance of corporate governance as a channel through which the adverse effect of fraud on liquidity can be moderated. II. Literature Review Several studies in the corporate fraud literature demonstrate that corporate fraud has a significant negative effect on firms' financing costs, investment decisions, and shareholder wealth. For example, Lin, Song, and Sun (2012) studied the impact of corporate fraud on fraudulent firms external financing costs. They compared fraudulent firms cost of debt financing before and after fraud exposure. They found that firms cost of debt significantly increased after the revelation of fraud. The 4

5 log value of firms loan spread increased by 29.6% after the revelation of corporate fraud. Lin and Paravisini (2011) found that banks have a weakened ability to fund syndicated loans externally after a firm they monitor commits fraud. Tian, Udell, and Yu (2012) found that a venture capital firm s reputation is negatively affected when it fails to prevent fraud in its portfolio companies. They found that venture capital firms that failed to prevent fraud had greater trouble in taking future portfolio companies public. Wang, Winton, and Yu (2008) suggested that a firm is more likely to commit fraud when investors are more optimistic about the industry s prospects. Wang and Winton (2012) found that the propensity for fraud is higher in competitive industries. They also found that poor performance in competitive industries following booms is largely concentrated in firms that are likely to have committed fraud during the booms. They suggested that fraud can intensify cyclical fluctuations in the real economy. Goldman, Peyer, and Stefanescu (2012) looked at how the announcement of an accusation of fraud affected industry rivals of the accused firm. They found that, in competitive industries, the announcements resulted in negative returns to rival shareholders. Velikonja (2013) also suggested that false information associated with hiding fraud and avoiding detection impairs risk evaluation by investors, the firms suppliers, and customers. It also alters the business decisions of fraudulent firms and their rivals. Dyck, Morse, and Zingales (2013) estimated that, on average, corporate fraud costs investors 22% of enterprise value in fraud-committing firms and 3% of enterprise value across all firms. Karpoff et al. (2008a) found that the reputational penalty associated with financial misrepresentation is more than 7.5 times the sum of all penalties imposed through the legal and regulatory system. They found that, on average, for every dollar a firm deceitfully inflates its market value, that dollar, plus an additional $3.08, is lost when the firm s misconduct is revealed. Thus, the 5

6 evidence in the fraud literature shows that fraud adversely affects a firm's financing costs, its investment decisions, and its shareholders' wealth. This study is also related to the literature examining the impact of firm-level governance (Chung, Elder and Kim 2010) and country-level governance (Eleswarapu and Venkataraman 2006) on stock liquidity. This literature found that poor governance lowers liquidity in the cross-section of stocks and that improvements in governance are associated with improvements in liquidity. However, companies with good corporate governance still commit fraud. The literature review shows that there is a lack of direct evidence of the impact of fraud on liquidity in the literature. In this study, we examine the effect of fraud on liquidity. We look at whether and how fraud affects the short-term and long-term liquidity of a stock. We are the first to document the increased transaction costs associated with trading a firm s stock after the revelation of fraud. We also suggest that good corporate governance is still beneficial to the firm even though good governance might not prevent the firm from committing fraud. Our results suggest the importance of good governance for mitigating the effect of fraud on liquidity. III. Data, Descriptive Statistics, and Methodology Data and Sample We obtained a sample of frauds detected between 1976 and 2005 from Karpoff et al. (2008a, b). Following the example of Karpoff, Koester, Lee, and Martin (2013), we identified an enforcement action as fraud if the SEC or Department of Justice (DOJ) filed charges alleging the violation of: (i) Section 17(a) of the Securities Act of 1933 for fraudulent interstate transactions related to the issuance of a security or (ii) Section 10(b) of the Securities Exchange Act of 1934 for manipulative and deceptive devices related to the trading of an already issued security. For a detailed description of the 6

7 data, please see Karpoff et al. (2008a, b). We collected our liquidity variables from Vanderbilt's Market Microstructure Database (Financial Markets Research Center). The Market Microstructure Database uses the Trade and Quote (TAQ) data distributed by the NYSE to calculate the variables for each stock-day. The data set starts on 4 January 1993 and is stored as a permanent SAS data set. It contains daily observations of variables measuring trade and quote properties for more than 8,000 common domestic and foreign stocks. Filters are applied to the trade and quote data before the variables are calculated. The stock-day begins at 9:30:00 hours and ends at 16:00:00 hours. Stocks whose monthly average price is less than three dollars are excluded from the file in that particular month. Infrequently traded stocks and foreign stocks are included. The data set contains 81 variables. Some of those variables include: daily share volume, number of trades, opening quote, average ask and bid size, mean half spread, mean trade-weighted effective half spread, average trade-weighted ask and bid price, share volume at the bid and ask side, and trading halt. We use Quoted Spread and Effective Spread in percent from the database as inverse measures of liquidity. Quoted Spread (%) is the average trade-weighted relative half-spread. Quoted Spread in percent is computed as follows: Quoted Spread (%) = 100*(A - B) / (A + B) / 2, where A is the ask and B is the bid price of the quote. Effective Spread (%) is the average relative effective half-spread of the stock-day. Effective Spread in percent is computed as follows: The numerator is the mean of the trade-weighted effective half-spread. The effective halfspread is the absolute difference between the trade price and the quote midpoint of the associated quote. The relative effective half-spread is the effective half-spread divided by the trade price. We excluded frauds detected before 1993 due to the lack of liquidity data. The Financial Markets Research 7

8 Center also limited our access to liquidity variables to data pertaining to dates through Thus, our final fraud sample contains 202 enforcement actions dating from 1993 through Table 1 shows the description of our fraud sample. The sample consists of 202 SEC enforcement actions from 1993 through The violation period extends from the beginning of the fraud until the end of the fraudulent activity. The median duration of the violation period is 729 days. The first public revelation date is the earliest date that information about the fraud became public. First public revelation dates are from the Karpoff et al. (2008a, b) database. First public revelation events include firm disclosures, internal investigations, restatements, auditor changes, management changes, third-party actions, SEC filing delays, whistle-blower charges, related investigations, related litigation, class action lawsuit filings, informal inquiries, formal investigations, and Wells Notices. The median duration from the beginning of the violation until the first public announcement is 760 days. The median durations of the violation period and the period from the beginning of the violation to the first public revelation increased between 1993 and The number of days from the end of the violation to the first public announcement can be interpreted as the difference between "Violation Beginning to First Public Announcement" and "Violation Period". The median duration from the ending of the violation until the first public announcement is 47 days and seems to decline between 1993 and Panel B of Table 1 also reports the number of events each year by first public revelation event types. [INSERT TABLE 1 HERE] Table 2 shows the distribution of the fraud sample by industry using two-digit Standard Industrial Classification (SIC) codes. While the fraudulent firms are broadly spread across 39 different industries, there are some concentrations of firms in the business services, electronic, and industrial machinery industries. 8

9 [INSERT TABLE 2 HERE] Figure I, replicated from Karpoff et al. (2008a), illustrates the sequence of an SEC enforcement action. An enforcement action starts with a trigger event that exposes the possible fraud and draws the SEC s attention. There are different types of trigger events (e.g., self-disclosure of malfeasance, a restatement, a change in management or in the audit firm, a shareholder lawsuit, a complaint by an investor or a whistle-blower, delayed SEC filings, an article in the press, investigations by other federal agencies). Following the trigger event, the SEC collects information through an informal inquiry that can turn into a formal investigation. The SEC then sends a Wells Notice to potential defendants, informing them that the SEC intends to file charges and offering them an opportunity to reply as to why civil charges should not be filed. The SEC then imposes administrative sanctions. The SEC can also refer the case to the DOJ, which may file criminal charges (Karpoff et al. 2008a, p. 588). [INSERT FIGURE I HERE] Table 3 shows summary statistics for the 1-day quoted spread (%), effective spread (%), and trading halt for important dates in our fraud sample from 1993 to Panel A presents quoted spread (%), effective spread (%), and trading halt for the first public announcement date, which is the earliest date that information about the fraud was available to the public. Quoted and effective spread data are available for 181 firms at the time of first public announcement. For all the first public announcements (181 events), the mean 1-day quoted spread is 1.205%. The mean 1-day effective spread is 0.913%. Of those firms, 29% experienced a trading halt during those days. The SEC identified the first public announcements for 170 firms. SEC-identified events consist of firm disclosures (50 events), auditor changes (8 events), internal investigations (47 events), management changes (5 events), registrations (9 events), related investigations (2 events), related 9

10 litigation (6 events), restatements (33 events), and third-party actions (9 events). The mean 1-day quoted spread on the SEC-identified trigger date is 1.223%. The mean 1-day effective spread on the SEC-identified trigger date for those firms is 0.924%, and 29.3% experienced a trading halt during those days. In 11 cases, the first public revelation date is identified in the Karpoff et al. (2008a, 2008b) database. Other first revelation events consist of announcements of SEC informal inquiries (5 events), formal investigations (3 events), Wells Notices (1 event), and class action lawsuits (2 events). For those 11 events, the mean quoted spread (effective spread) is 0.935% (0.744%). Panel B of Table 3 shows important follow-up announcements about each fraud that were made after the first public announcement date. The follow-up announcements consist of notices of informal SEC inquiries, formal SEC investigations, Wells Notices, initiation of regulatory proceedings, initiation of class action lawsuits, and bankruptcy. We only use those firms that have at least three announcements in Panel B. The mean quoted spread (effective spread) for the firms with second announcements is 0.755% (0.604%). The mean quoted spread (effective spread) for the firms with third announcements is 0.738% (0.576%). In Panel C, we divide our sample into low, medium, and high levels based on market cap, volatility, and volume and show quoted spread (%), effective spread (%), and trading halt for the first public announcement date. [INSERT TABLE 3 HERE] Methodology We employed the following regression to study the effect of fraud on firms short-term and long-term liquidity while controlling for firm-specific variables noted in the literature (Huang and Stoll 1996). We use Quoted spread (%) and Effective spread (%) as dependent variables (inverse liquidity measures). 10

11 (1) where Post fraud is a dummy variable equal to 1 after the revelation of fraud and 0 otherwise, Price volatility is the standard deviation of returns, Price is the average trade price, Volume is the trading volume, and Firm size is the dollar market capitalization of the stock. There are two observations for every firm. Day 0 is the trading day on which the fraud was publicly revealed. For the pre-fraud period (Before [-5,-1]), the short-term liquidity is calculated as the mean of the liquidity on days -5 through - 1. For the post-fraud period (After [0,+5]), the short-term liquidity is calculated as the mean of the liquidity on days 0 through +5. We removed firms that did not have liquidity data for both periods. For the pre-fraud period (Before [-200,-6]), the long-term liquidity is calculated as the mean of the liquidity on days -200 through -6. For the post-fraud period (After [+6,+200]), the long-term liquidity is calculated as the mean of the liquidity on days +6 through Pre and post averages for the control variables are calculated the same way as the liquidity measures. Standard errors are estimated, allowing for clustering at the two-digit SIC industry level because the fraudulent activities are correlated with economic downturns for the industries (e.g., Wang, Winton, and Yu 2010; Wang and Winton 2012). IV. Empirical Results Liquidity around the Revelation of Fraud In Panel A of Table 4, we show the mean quoted spread (%) and effective spread (%) for our fraud sample during the 401-trading-day window around the revelation of fraud. Number of events indicates 11

12 the number of firms used in calculating the mean for the quoted spread and effective spread for each trading day. Number of events changes due to the limited availability of data on liquidity for some trading days. Quoted spread (%) and effective spread (%) increase gradually from day -200 through day 0, hitting the highest point on day 0. Although quoted spread (%) and effective spread (%) steadily decrease through day +200, they remain high. Panel B shows the difference in spread across time. The notations ***, **, and * in Panel B represent 1%, 5%, and 10% significance levels, respectively. [INSERT TABLE 4 HERE] Figure II plots the mean quoted spread (%) and effective spread (%) for our fraud sample during the 401-trading-day window around the revelation of fraud. Day 0 is the trading day on which the fraud was publicly revealed. Quoted spread (%) and effective spread (%) increase over the 200 days before the revelation of fraud. Quoted spread (%) and effective spread (%) remain high in the days after the revelation of fraud. [INSERT FIGURE II HERE] Table 5 demonstrates changes in liquidity around the revelation of fraud. In Panel A, we show short-term changes in liquidity. Difference (After - Before) indicates the changes in liquidity derived by subtracting the pre-fraud levels from the post-fraud levels. The difference for quoted spread (%) is 0.306% and is statistically significant at 1%. The difference for effective spread (%) is 0.236% and is statistically significant at 1%. In other words, quoted spread (%) and effective spread (%) increase by 0.306% and 0.236%, respectively, in the post-fraud window compared with the pre-fraud window. The difference for trading halt (%) is 9.2% and is statistically significant at 1%. Panel B of Table 5 presents long-term changes in liquidity. The difference for the quoted spread (%) is 0.314% and is statistically significant at 1%. The difference for effective spread (%) is 12

13 0.232% and is statistically significant at 1%. The decreased liquidity continues over the long run. The difference for trading halt (%) is 0.5% and is statistically significant at 1%. Thus, the univariate results show that liquidity deteriorates after the revelation of fraud for both the short-term and long-term windows. [INSERT TABLE 5 HERE] Table 6 explains the impact of severity of fraud on liquidity for both short-term and long-term windows. We employ several indicator variables to measure the severity of fraud (Karpoff et al. 2007). Insider trading dummy is an indicator variable taking the value of 1 for enforcement actions that include allegations of insider trading. Department of Justice dummy is an indicator variable taking the value of 1 for enforcement actions that include one or more civil or criminal proceedings by the DOJ. Cooperated in investigation dummy is an indicator variable taking the value of 1 for firms identified by regulators in the enforcement action as having self-reported, cooperated, or taken remedial actions. Initial public offering dummy is an indicator variable taking the value of 1 for enforcement actions associated with a firm s initial public offering. Foreign company dummy is an indicator variable taking the value of 1 for foreign firms. Bankruptcy dummy is an indicator variable taking the value of 1 for firms that filed for bankruptcy. Panel A of Table 6 presents the effect of severity of fraud on short-term changes in liquidity. The results show that fraudulent firms with allegations of insider trading have higher quoted and effective spreads during the short-term window than fraudulent firms without insider trading allegations. We can also see that a firm s liquidity suffers more severely in the short-term window if the enforcement action is associated with the firm's initial public offering. Fraudulent firms that 13

14 eventually filed for bankruptcy also have higher quoted and effective spreads during the short-term window than those that did not file for bankruptcy. Panel B of Table 6 shows the effect of fraud severity on long-term changes in liquidity. We still see a negative effect on long-term liquidity for firms that filed for bankruptcy and firms whose enforcement actions were associated with their initial public offerings. Insider trading accusations do not seem to affect firms' long-term liquidity. We do not see any difference in short-term and long-term liquidity for foreign firms, non-cooperating firms, or firms that faced additional enforcement actions by the DOJ. [INSERT TABLE 6 HERE] The Effect of Corporate Fraud on Short-Term Liquidity In this section, we examine the effect of fraud on a firm s short-term liquidity. In Panel A of Table 7, the dependent variable is Quoted spread (%), which is the average trade-weighted half-spread of a stock. We removed firms that did not have liquidity data for both periods, leaving 185 firms in our sample. In Model 1 of Panel A, Post fraud dummy is and is statistically significant at 1%. In other words, after the revelation of fraud, the quoted spread increased by 0.138% in the short run. The results show that the revelation of fraud adversely affects the short-term liquidity of fraudulent firms by a considerable amount. The coefficients for post-fraud variables are statistically significant in all regressions. The fraudulent activities differ in terms of complexity and the degree of harm caused to shareholders. Thus, we also examine the impact of the severity of fraud on short-term liquidity by using several variables to measure the complexity and severity of the fraud (Karpoff et al. 2007). Regulatory penalties are the total fines assessed against the firm and its agents by the SEC, DOJ, and 14

15 state attorneys general. The variable # Respondents represents the number of individuals named in the enforcement proceedings. The variable # Proceedings represents the total number of administrative, civil, and criminal proceedings in the enforcement action. We examine the interaction of the Post fraud variable with the variables measuring the complexity and severity of the fraud. None of the interactions between those variables and Post fraud dummy is statistically significant. This suggests that the complexity and severity of fraudulent activities does not further reduce a firm s short-term liquidity; rather, the existence of fraud is what matters to investors. 1 In Panel B of Table 7, the dependent variable is Effective spread (%), which is the average trade-weighted effective half-spread of the stock. In Model 1 of Panel B, Post fraud dummy is and is statistically significant at 1%. The overall results suggest that short-term market liquidity deteriorates after the revelation of corporate fraud. [INSERT TABLE 7 HERE] The Effect of Corporate Fraud on Long-Term Liquidity Table 8 looks at the effect of fraud on long-term liquidity. We removed firms that did not have liquidity data for both periods, leaving a sample of 188 firms. In Panel A of Table 8, the dependent variable is Quoted spread (%). In Model 1 of Panel A, Post fraud dummy is and is statistically significant at 5%. In the long term, the quoted spread increased by 0.152% after the revelation of fraud. These findings validate the hypothesis that the revelation of fraud negatively affects long-term liquidity. The coefficients for the Post fraud variable are statistically significant in all regressions. We look at the effect of the severity and complexity of fraud on long-term liquidity by using several 1 Our results do not change when we use the abnormal stock returns around the announcement dates as a proxy for the severity of the fraud. 15

16 variables (Karpoff et al. 2007), and we examine the interaction of our Post fraud variable with those variables. None of the interactions of those variables with Post fraud dummy is statistically significant. This suggests that the complexity and severity of fraudulent activities does not further reduce longterm liquidity. In Panel B of Table 8, the dependent variable is Effective spread (%). In Model 1 of Panel B, Post fraud dummy is and is statistically significant at 5%. The results in Table 7 and Table 8 suggest that the revelation of corporate fraud has an adverse effect on liquidity not only in the short term but also in the long term. [INSERT TABLE 8 HERE] The Interaction Effect of Corporate Fraud and Governance on Liquidity The results above suggest that the revelation of fraud decreases liquidity both in the short term and in the long term. In this section, we examine the interaction effect of corporate fraud and governance on liquidity. It is possible that firms with weak corporate governance have more severe liquidity problems after the revelation of fraud than those with better corporate governance. Strong corporate governance can assure investors that there will be adequate internal and external mechanisms through which fraudulent activities can be prevented in the future. We examine the governance aspects of outside director percentage, the percentage of institutional ownership, and the quality of the external audit firm (Farber 2005). We show the interactions of Post fraud and our corporate governance proxies in Table 9. We use Quoted spread (%) and Effective spread (%) as dependent variables (inverse liquidity measures). In Panel A of Table 9, we look at the interaction effect of corporate fraud and governance on short-term liquidity. The sample in Panel A contains 185 firms. Our first proxy for corporate 16

17 governance is Outside directors %. Rosenstein and Wyatt (1990), Weisbach (1988), and Brickley and James (1987) found that boards consisting of outside directors are more effective than boards composed of insiders. Outside director % is the percentage of independent directors on a firm's board. In Models 1 and 2 of Panel A, we include the interaction term between Post fraud and Outside director % in the regression. In Model 1, the dependent variable is Quoted spread (%); the coefficient for the interaction term is and is statistically significant at 5%. In other words, the revelation of fraud affects the short-term liquidity of firms with good governance less adversely than it does firms with weak governance. In Model 2 of Panel A, the dependent variable is Effective spread (%). The coefficient for the interaction term is and remains significant at 5%. Our second proxy for corporate governance is Institutional holdings %. Institutional investors can provide external monitoring of firms (Jensen 1993). Institutional holdings % is the percentage of ownership by institutions. In Models 3 and 4 of Panel A, we include the interaction term between Post fraud and Institutional holdings % in the regression. In Model 3, the dependent variable is Quoted spread (%). The coefficient for the interaction term is , and it is statistically significant at 5%. In Model 4, the dependent variable is Effective spread (%). The coefficient for the interaction term is and remains significant at 1%. Our third proxy for corporate governance is Major accounting firm dummy. DeFond (1992) and Palmrose (1988) suggested that a major audit firm is likely to provide better monitoring than a smaller audit firm. Major accounting firm dummy is an indicator variable taking the value of 1 to indicate that the firm s auditor was one of the 8/6/5/4 major accounting firms. In Models 5 and 6 of Panel A, we include the interaction term between Post fraud and Major accounting firm dummy in the regression. In Model 5, the dependent variable is Quoted spread (%); the coefficient for the interaction term is and is statistically insignificant. In Model 6, the dependent variable is Effective spread 17

18 (%). Here, the coefficient for the interaction term is and remains significant at 1%. These results suggest that the short-term liquidity of fraudulent firms with weak governance, as evidenced by inadequate monitoring, is more adversely affected than the short-term liquidity of fraudulent firms with better governance. In Panel B of Table 9, we look at the interaction effect of corporate fraud and governance on long-term liquidity. The sample in Panel B contains 188 firms. In Models 1 and 2 of Panel B, we include the interaction term between Post fraud and Outside director % in the regression. In Model 1, the dependent variable is Quoted spread (%); the coefficient for the interaction term is and is statistically significant at 1%. This indicates that the revelation of fraud seems to affect the long-term liquidity of firms with good governance less adversely than it does firms with weak governance. In Model 2, the dependent variable is Effective spread (%). The coefficient for the interaction term is and remains significant at 1%. In Models 3 and 4 of Panel B, we include the interaction term between Post fraud and Institutional holdings % in the regression. In Model 3, the dependent variable is Quoted spread (%); the coefficient for the interaction term is and is statistically significant at 5%. In Model 4, the dependent variable is Effective spread (%). The coefficient for the interaction term is and remains significant at 10%. In Models 5 and 6 of Panel B, we include the interaction term between Post fraud and Major accounting firm dummy in the regression. In Model 5, the dependent variable is Quoted spread (%); the coefficient for the interaction term is and is statistically significant at 10%. In Model 6, the dependent variable is Effective spread (%). The coefficient for the interaction term is and remains significant at 10%. Thus, we find that the liquidity effect of corporate fraud is more severe for firms with weak governance than for those with strong governance. These results highlight the 18

19 importance of corporate governance as a channel through which the adverse effect of fraud on liquidity can be moderated. [INSERT TABLE 9 HERE] V. Robustness We also look at the impact of fraud revelation on information asymmetry. Spread measures can be decomposed into order processing, inventory, and information asymmetry components. There are several methods (e.g., Glosten and Harris 1988; Huang and Stoll 1997; Easley, Keifer, O'Hara, and Paperman 1996) that extract the adverse selection component from the spread. Unfortunately, Vanderbilt's Market Microstructure Database (Financial Markets Research Center) only contains daily observations of effective and quoted spreads and does not decompose the spread measures into their adverse selection components. To calculate the information asymmetry measures, we used an alternative database that is publicly available. We used monthly data disclosed under Securities and Exchange Commission Rule 605 of Regulation NMS from 2001 to This rule requires that exchanges and market centers release order-based performance statistics by stock, order type, and order size. Price impacts of trades are commonly used in the market microstructure literature to examine the extent of information-based trading. Price impact of a trade will be zero if the trade has no new information on the share's value. However, if it is an information-motivated trade, stock price would increase if the trade is initiated by a buyer and decrease if the trade is initiated by a seller. Thus, price impact shows the value of private information held by informed traders. We used two different high- 2 The reporting of execution quality was initially required under SEC Rule 11Ac1-5 starting in mid Regulation NMS (approved on 8/29/2005) changed the numbering of Rule 11AC1-5 to Rule

20 frequency price impact measures as proxies for information asymmetry. Our first price impact measure was developed by Goyenko, Holden, and Trzcinka (2009). Goyenko et al. (2009) stated that "A static version of price impact is the slope of the price function at a moment in time. Essentially, this is the cost of demanding additional instantaneous liquidity and can be thought of as the first derivative of the effective spread with respect to order size. The price impact benchmark uses two (aggregated) points on this curve to measure the slope (p. 156)." Static price impact based on Rule 605 data for a given stock is calculated as follows: Static price impact (650) = [($Effective spread (605) Big Orders /P) - ($Effective spread (605) Small Orders /P)] / [(Ave trade size (605) Big Orders ) - (Ave trade size (605) Small Orders )]* , Effective spreads are reported only for market orders and marketable limit orders in the 605 data. Effective spread (605) is the dollar share-weighted average monthly spread reported by market centers. P is the average price in a month. Big orders is the set of all orders in the range of while small orders is the set of all orders in the range of Our second price impact measure was used by Boehmer (2005) and Bessembinder (2003). Proportional price impact (%) based on Rule 605 data for a given stock is calculated as follows:, Effective spread can be interpreted as the total price impact of a trade and can be decomposed into a permanent and a temporary component (Boehmer 2005 p. 560). Realized spread can be considered a temporary or non-informational price impact. Thus, Proportional price impact reflects the private information conveyed by informed traders. Rule 605 data is reported monthly. Thus, for the pre-fraud period (Before [-12,-1]), the price impact is calculated as the mean of the price impact on months -12 through -1. For the post-fraud period (After [0,+12]), the price impact is calculated as the 20

21 mean of the price impact on months 0 through +12. We removed firms that did not have price impact data for both periods. Our sample size has been reduced significantly since the Rule 605 data is available only from 2001 until In Model 1 of Table 10, the dependent variable is Static price impact. Post fraud dummy is and is statistically significant at 10%. In Model 2 of Table 10, the dependent variable is Proportional price impact (%). Post fraud dummy is and is statistically significant at 1%. The results show that the revelation of fraud increases information asymmetry in fraudulent firms. Thus, our results are robust to alternative measures of liquidity. [INSERT TABLE 10 HERE] VI. Conclusion In this paper, we looked at whether and how corporate fraud affects the short-term and long-term liquidity of a firm s stock. We documented a new channel through which corporate fraud affects shareholders' wealth (via a reduction in liquidity). We found that the revelation of corporate fraud has an adverse effect on liquidity not only in the short term but also in the long term. In the short run, the quoted spread (effective spread) increases by 0.138% (0.108%) after the revelation of fraud. In the long run, the quoted spread (effective spread) increases by 0.152% (0.098%) after the revelation of fraud. Fraudulent activities differ in terms of complexity and the degree of harm caused to the shareholders. Thus, we also examined the impact of the severity and complexity of fraud on firms liquidity. The results suggest that the complexity and severity of fraudulent activities does not further reduce liquidity; rather, the existence of fraud is what matters to investors. 21

22 Apart from analyzing the effect of fraud on liquidity, we also studied the interaction effect of corporate fraud and governance on liquidity. We examined whether firms with weak corporate governance have more severe liquidity problems after the revelation of fraud than those with better corporate governance. Our results suggest that the liquidity of fraudulent firms with weak governance is more adversely affected than the liquidity of fraudulent firms with better governance. These results highlight the importance of corporate governance as a channel through which the adverse effect of fraud on liquidity can be moderated. 22

23 TABLE 1. Description of Our Fraud Sample. Our fraud sample consists of 202 SEC enforcement actions from 1993 through The violation period extends from the beginning of the fraud until the end of the fraudulent activity. The first public revelation date is the earliest date that information about the fraud became public. First public revelation dates are taken from the Karpoff et al. (2008a, b) database. First public revelation events include firm disclosures, internal investigations, restatements, auditor changes, management changes, third-party actions, SEC filing delays, whistle-blower charges, related investigations, related litigation, class action lawsuit filings, informal inquiries, formal investigations, and Wells Notices. Panel B reports the number of events each year by first public revelation event types. Panel A: Violation Period (Days) Violation Beginning to First Public Announcement (Days) Violation Ending to First Public Announcement (Days) Year Number of Events Mean Median Mean Median Mean Median Total

24 Panel B: Year Announce ment Auditor Change/With drawn Audit Internal Investigation Management Change Registration/ Reports Related Investigat ion Related Litigation Restat ement Third Party Unusual Trading Class Action Lawsuit Filing Informal Inquiry Formal Investigat ion Total Wells Notice 24

25 TABLE 2. Distribution of the Fraud Sample by Industry. This table shows the distribution of the fraud sample by industry based on two-digit SIC code. Two-Digit SIC Code Industry Number of Firms 13 Oil and Gas Extraction 3 14 Nonmetallic Minerals, Except Fuels 1 20 Food and Kindred Products 5 22 Textile Mill Products 1 23 Apparel and Other Textile Products 5 26 Paper and Allied Products 3 27 Printing and Publishing 3 28 Chemical and Allied Products 8 32 Stone, Clay, and Glass Products 1 33 Primary Metal 2 34 Fabricated Metal Products 5 35 Industrial Machinery and Equipment Electronic and Other Electric Equipment Transportation Equipment 2 38 Instruments and Related Products Miscellaneous Manufacturing 1 45 Transportation by Air 1 47 Transportation Services 1 48 Communications 6 49 Electric, Gas, and Sanitary Services 9 50 Wholesale Trade - Durable Goods 7 51 Wholesale Trade - Nondurable Goods 8 53 General Merchandise Stores 1 56 Apparel and Accessory Stores 1 57 Furniture and Home Furnishings Stores 1 58 Eating and Drinking Places 1 59 Miscellaneous Retail 6 60 Depository Institutions 7 61 Nondepository Institutions 3 62 Security and Commodity Brokers 1 63 Insurance Carriers 2 64 Insurance Agents, Brokers, and Services 1 67 Holding and Other Investment Offices 2 72 Personal Services 2 73 Business Services Miscellaneous Repair Services 1 79 Amusement and Recreation Services 2 80 Health Services 3 87 Engineering and Management Services 3 Total

26 TABLE 3. Liquidity Response to the Revelation of Fraud. This table shows summary statistics for the 1-day quoted spread (%), effective spread (%), and trading halt for important dates in our fraud sample from 1993 to Quoted Spread (%) is the average trade-weighted half-spread of a stock. Effective Spread (%) is the average trade-weighted effective half-spread of the stock. Halt is a dummy variable that is equal to 1 if the firm experienced a trading halt and 0 otherwise. Panel A presents quoted spread (%), effective spread (%), and trading halt for the first public announcement date, which is the earliest date that information about the fraud was available to the public. Most of the first public announcements are SEC-identified events such as firm disclosures, auditor changes, internal investigations, management changes, SEC filing delays, related investigations, related litigation, restatements, and third-party actions. In 11 cases, the first public revelation date is identified in the Karpoff et al. (2008a, b) database. Other first revelation events consist of announcements of SEC informal inquiries, formal investigations, Wells Notices, and class action lawsuits. Panel B reports on important follow-up announcements about the frauds. We only use those firms that have at least three announcements in Panel B. These follow-up announcements include notices of informal SEC inquiries, formal SEC investigations, Wells Notices, initiation of regulatory proceedings, initiation of class action lawsuits, and bankruptcy. In Panel C, we divide our sample into low, medium, and high levels based on market cap, volatility, and volume and show quoted spread (%), effective spread (%), and trading halt for the first public announcement date. N Quoted Spread (%) 26 Effective Spread (%) Panel A: First Public Announcement All first public announcements SEC-identified trigger events Announcement Auditor change/withdrawn audit Internal investigation Management change Registration/reports Related investigation Related litigation Restatement Third party Unusual trading Other first revelation events SEC informal inquiry SEC formal investigation SEC Wells Notice Class action lawsuits begin Panel B: Important Follow-Up Announcements 2nd announcement rd announcement Panel C: First Public Announcement by Market Cap, Volatility, and Volume N Quoted Effective N Halt Spread (%) Spread (%) Low Market Cap Medium Market Cap N Halt

27 High Market Cap Low Volatility Medium Volatility High Volatility Low Volume Medium Volume High Volume

28 TABLE 4. Liquidity around the Revelation of Fraud. Panel A of this table shows the mean quoted spread (%) and effective spread (%) for our fraud sample during the 401- trading-day window around the revelation of fraud. Day 0 is the trading day on which the fraud was publicly revealed. Quoted spread (%) is the average trade-weighted half-spread of a stock. Effective spread (%) is the average tradeweighted effective half-spread of the stock. Number of events indicates the number of firms used in calculating the mean for quoted spread (%) and effective spread (%) for each trading day. Number of events changes due to the limited availability of data on liquidity. Panel B shows the difference in spread across time. The notations ***, **, and * in Panel B represent 1%, 5%, and 10% significance levels, respectively. Panel A: Trading Day Relative to the Revelation of Fraud Number of Events Quoted Spread (%) Effective Spread (%)

29 Panel B: Quoted Spread (%) Effective Spread (%) Day + 1 minus Day ** 0.25** Day + 2 minus Day Day + 3 minus Day * Day + 4 minus Day * 0.161** Day + 5 minus Day *** 0.228*** 29

30 TABLE 5. Changes in Liquidity around the Revelation of Fraud. This table shows changes in liquidity around the revelation of fraud. Panel A presents short-term changes in liquidity. The means of those variables from day -5 to day -1, Before [-5,-1], are used to calculate short-term liquidity levels for the pre-revelation period. The means of the variables from day 0 to day +5, After [0,+5], are used to calculate short-term liquidity levels for the period following the revelation of fraud. Panel B shows long-term changes in liquidity. The means of the variables from day -200 to day -6, Before [-200,-6], are used to calculate long-term liquidity levels for the prerevelation period. The means of the variables from day +6 to day +200, After [+6,+200], are used to calculate long-term liquidity for the period following the revelation of fraud. Difference (After - Before) indicates short-term and long-term changes in liquidity. The notations ***, **, and * represent 1%, 5%, and 10% significance levels, respectively. Panel A: Short-Term Changes Before [-5,-1] After [0,+5] Difference (After - Before) Quoted Spread (%) *** Effective Spread (%) *** Halt *** Panel B: Long-Term Changes Before [-200,-6] After [+6,+200] Difference (After - Before) Quoted Spread (%) *** Effective Spread (%) *** Halt *** 30