Abnormal Trading Behavior of Specific Types of Shareholders Before Firm Bankruptcy and its Ramifications for Firm Bankruptcy Prediction

Transcription

1 Abnormal Trading Behavior of Specific Types of Shareholders Before Firm Bankruptcy and its Ramifications for Firm Bankruptcy Prediction Abstract: This paper examines the behavior of corporate insiders and certain groups of institutional investors (short-term transient top-performing and those with fiduciary responsibility) in the eight quarters leading up to a firm s bankruptcy filing. Using a matched sample based on year industry and probability of future bankruptcy we find that corporate insiders display reduced net trading activity in the quarters before bankruptcy with corporate insiders going quiet with no buying or selling in the quarters immediately preceding bankruptcy. In contrast we find that our identified groups of institutional shareholders display abnormal selling activity several quarters before bankruptcy. We then use this information to enhance the predictive capabilities of Campbell et al. s (2008) bankruptcy model and find the variables measuring the absolute value of net purchases by corporate insiders along with the changes in ownership by specific groups of institutional shareholders improves the predictive capabilities of the Campbell et al (2008) bankruptcy prediction model. Keywords: Insider Ownership Institutional Ownership; Bankruptcy Prediction; Financial distress JEL Classifications: G11 G20 G33 1

2 Abnormal Trading Behavior of Specific Types of Shareholders Before Firm Bankruptcy and its Ramifications for Firm Bankruptcy Prediction In the 1980 s pre-bankruptcy shareholders were able to recover at least part of their investment approximately 78% of the time after a firm went through the bankruptcy process (LoPucki and Whitford 1990). However a new survey using more recent data indicates that less than 10% of the pre-bankruptcy shareholders were able to recover any portion of their investment (Wood 2011). Additionally according to data provided by New Generation Research Inc. the combined asset value of the 52 public companies that filed for bankruptcy in 2014 was $71.8 billion. The convergence of the decline in recovery rate for equity holders in bankruptcy and the potential for substantial losses makes bankruptcy a catastrophic event for shareholders. As a result shareholders have increased and renewed interest in improving methods for predicting impending bankruptcy. Lennox (1999) finds that bankruptcy prediction models such as those developed by Altman (1968) Merton (1974) Ohlson (1980) Shumway (2001) Neophytou and Mar Molinero (2004) Dewaelheyns and Val Hulle (2006) and Campbell et al. (2008) are generally able to outperform bankruptcy predictions based on auditor issued going concern opinions. Therefore Lenox (1999) suggests that stakeholders rely on these bankruptcy prediction models for decision making. These existing bankruptcy predication measures focus primarily on accounting financial statement information and market return data. However Piotroski and Roulstone (2004) identify two potentially important sources of firm-specific information: (1) corporate insider trading activity and (2) trading behavior of specific categories of institutional investors. The findings by Piotroski and Roulstone (2004) suggest that adding information related to investor trading activity to the current bankruptcy prediction models may enhance their predictive capabilities. In this study we investigate whether certain categories of investors 1

3 display abnormal trading behavior in the quarters leading up to a firm s bankruptcy filing. We then consider whether the inclusion of the trading behavior of the specific groups of investors that do demonstrate abnormal trading behavior improves the predictive capabilities of Campbell et al. s (2008) bankruptcy prediction model. Corporate insiders have access to information not necessarily disclosed to the market. Conventional wisdom suggests that corporate insiders would act upon this non-public information to sell their equity interests in a firm prior to bankruptcy (Seyhun and Bradley 1997). However Marin and Olivier (2008) report that lower insider selling volume over the recent past is associated with a higher probability of stock price crash. Further more recent research by Gao et al. (2015) speculates that corporate insiders may face litigation risk if they sell their equity interests before the release of bad news. Based on the inconsistent findings of Seyhun and Bradley (1997) along with Marin and Olivier (2008) and Gao et al (2105) it is not clear whether firm insiders will take advantage of their insider information and liquidate their equity interests before impending bankruptcy or fearing litigation risk corporate insiders will abstain from trading before impending bankruptcy. Regardless of whether corporate insiders demonstrate abnormal selling or are abnormally quiet in their trading activity in the quarters leading up to a bankruptcy filing their behavior has the potential to inform the market. Therefore the inclusion of their trading behavior may improve Campbell et al. s (2008) bankruptcy predication model. Unlike managers who have private information institutional investors may expend considerable time and effort to acquire and process all available information. Initial research in the area of institutional investors or institutional ownership in the periods preceding bankruptcy treats these investors as a homogeneous group. For instance Frino et al (2014) examines aggregate institutional ownership in the periods leading up to firm bankruptcy. Using 2

4 proprietary ASX data Frino et al (2014) find that aggregate institutional ownership is gradual commencing approximately 115 days prior to bankruptcy. However more recent research recognizes the heterogeneity of institutional investors and attempts to partitions them into categories based on different investment horizons different investment objectives and styles and different fiduciary or legal restrictions (Yan and Zhang 2009). Previous research shows that specific types of institutional shareholders are able to consistently display superior information processing capabilities by outperforming other types of institutional shareholders in particular information environments (Yan and Zhang 2009). We focus on four institutional investor groups empirically identified by previous research as outperforming other groups of institutional investor ahead of certain information events either due to superior information processing 1 or an aversion to risk: (1) institutional investors with a short-term investment horizon (Yan and Zhang 2009) (2) transient institutional shareholders (Bushee 1998) (3) top performing institutional investors (Adebambo et al. 2015) and (4) those institutional investors with fiduciary responsibilities (DelGuercio 1996). Using a matched sample design we investigate abnormal trading behavior for each of our investor types. We match firms that filed for bankruptcy between 1992 and 2012 with similarly distressed firms that survive financial distress. Our evidence suggests that litigation risk may influence corporate insiders to forgo trading several quarters ahead of a bankruptcy filing extending Gao et al. s (2015) work into the more specialized bad news setting of bankruptcy. Furthermore extending the work of Frino et al (2014) that aggregate institutional shareholders begin a gradual withdrawal from firms approximately 115 banks before bankruptcy we find that 1 Consistent with Yan and Zhang (2009) Ramalingegowda (2014) and Adebambo et al. (2015) we define the term superior information processing as the institutional shareholders superior processing of both non-public and public information. 3

5 each of the separate categories of institutional investors that we analyze demonstrate abnormal trading behavior for firms that ultimately declare bankruptcy as compared to those firms that ultimately survive financial distress. Since our results indicate that corporate insiders dramatically reduce their trading volume in the periods leading up to bankruptcy while specific categories of institutional shareholders engage in abnormal selling in the quarters leading up to a firm s bankruptcy filing we next explore whether the trading volume and trading behavior of each of these investor types could be used to improve Campbell et al. s (2008) bankruptcy prediction model. We augment Campbell et al. s (2008) bankruptcy prediction model with variables measuring net trading activity for corporate insiders in the previous six months along with changes in percentage ownership for specific groups of institutional shareholders in the previous two quarters. We find that including these additional variables strengthens the predictive capabilities of the Campbell et al. s (2008) bankruptcy prediction model for firms filing for bankruptcy within the future two quarters. Lower corporate insiders net trading activity results in an increased probability of firm bankruptcy in the next six months. Conversely larger decreases in ownership percentage by specific groups of institutional investors in the previous two quarters indicates a higher probability of firm bankruptcy within the future two quarters. This study makes several contributions. First our results extend Piotroski and Roulstone s (2004) suggestion that both corporate insider ownership changes and institutional investor ownership changes provide important firm-specific information to the bankruptcy setting. In doing so we make a significant contribution to the bankruptcy prediction literature by providing evidence that the absence of corporate insider trading along with changes in ownership by specific categories of institutional shareholders can be used to enhance the bankruptcy 4

6 prediction model. To our knowledge our study is the first to employ the trading behavior of corporate insiders and specific categories of institutional investors to a forward-looking bankruptcy prediction analysis. Second our results add to the growing body of research investigating institutional investor informed trading such as Yan and Zhang (2009) Puckett and Yan (2011) and Adebambo et al. (2015). Finally our work also specifically expands upon Ramalingegowda s (2014) research of the bankruptcy setting by examining the trading behavior of additional groups of institutional investors. Beyond our contributions to the existing literature our study should be of interest to agencies or individuals who seek to evaluate financially distressed firms. Lennox (1999) calls auditors to consider information in bankruptcy prediction models to reduce the potential of failing to issue a going concern opinion when the firm actually files for bankruptcy within the future year. Thus both credit monitoring agencies and auditors could apply our enhanced bankruptcy prediction model to inform credit decisions or going concern reports for distressed firms. Our evidence also provides insights to regulators by indicating the presence of substantial relevant information which certain institutional investor groups are using to enlighten their trading behavior as distressed firms approach bankruptcy. Regulators concerned with ensuring parity in information access to all investors should be interested in such evidence. Finally our results should be of interest to non-institutional investors who could garner additional firmspecific information from institutional investors trading behavior and potentially avoid being the investors left holding the empty bag when firms files for bankruptcy. The remainder of the paper is organized as follows. The next section presents the background research and develops the hypotheses. Section three presents the research design and 5

7 sample selection while section four presents our results. Section five discusses the implications and conclusions. 2. Hypothesis development Piotroski and Roulstone (2004) identify both changes in corporate insider ownership and changes in institutional ownership as firm-specific sources of information. In this section we develop hypotheses to examine whether corporate insiders engage in abnormal trading or dramatically curtail trading in the quarters leading up to firm bankruptcy. Furthermore we investigate whether specific groups of institutional investors engage in abnormal trading behavior as a firm approaches bankruptcy. Subsequently we use our findings regarding the trading behavior of corporate insiders and specific groups of institutional shareholders to ascertain whether information regarding the trading or absence of trading by corporate insiders or specific types of institutional shareholders over the previous six months can be used to enhance Campbell et al. s (2008) bankruptcy prediction model in forecasting bankruptcies in the future six months. 2.1 Hypothesis 1a - Corporate Insider Trading Behavior Before Bankruptcy The notion that managers can profit by trading on their private information is well established in existing research (Jaffe 1974 Elliot Morse Richardson 1984 Givoly and Palmon 1985 Seyhun 1986 Rozeff and Zaman 1988 Ke Huddart and Petroni 2003 Piotroski and Roulstone 2004 Piotroski and Roulstone 2005). Furthermore Piotroski and Roulstone (2005) provide strong evidence that corporate insider trades are positively related to the firm s future earnings performance a proxy for superior cash flow. Seyhun and Bradley s (1997) evidence that managers engage in significant selling behavior prior to bankruptcy suggests that managers profit by trading on their private information as their firm approaches bankruptcy. Furthermore 6

8 using a sample of bankruptcies from Stanley et al (2009) find that top executives engage in net selling prior to bankruptcy. However subsequent to the study from Seyhun and Bradley (1997) and Stanley et al. (2009) changes to the legal and regulatory environment may have altered the trading behavior of corporate insiders of firms facing impending bankruptcy. The Cornerstone Research Database maintained by Stanford Law School reports a dramatic increase in the number of Security Class Action Lawsuits against Corporate Insiders in 2001 and 2002 corresponding with the collapse of Enron Corporation WorldCom Inc. and Tyco International Ltd. Brochet and Srinivasan (2010) find that after passage of the Sarbanes-Oxley Act of 2002 (SOX) corporate insiders were less likely to sell shares immediately prior to negative stock returns and ahead of earnings news that falls short of analyst forecasts. In the current legal environment corporate insiders willingness to profit from private information may be tempered by litigation risk political costs (Piotroski and Roulstone 2005) or even the threat of SEC enforcement actions (Cohen et al 2012). Marin and Olivier (2008) find that sales by corporate insiders may indicate future bad news and future declining stock price. However they further find that the absence of selling by corporate insiders may indicate the existence of substantially worse news that when disclosed in the future leads to a higher probability of crash in stock price. Cohen et al. (2012) notes that several high-profile insider trading cases in recent years indicates that the SEC continues to vigorously investigate and prosecute cases involving inappropriate use of information by corporate insiders. Following this sentiment Gao et al. (2015) theorize and find evidence suggesting that fearing litigation risk corporate insiders neither buy nor sell company owned shares when they anticipate future bad news. During periods of corporate insider trading silence Gao et al. (2015) find that future stock returns are significantly lower. In contrast they find that corporate insiders 7

9 sell shares when they do not anticipate significant future bad news. The fear of litigation risk may assist in explaining Adebabmbo et al. s (2015) finding that unlike specific types of institutional shareholders corporate insiders in financial institutions did not sell company owned stock during the financial crisis. Based on these mixed findings it is unclear whether Seyhun and Bradley s (1997) results that corporate insiders engage in significant selling behavior prior to bankruptcy using data from will hold in a more recent setting. Based primarily on more recent work including Gao et al. (2015) our first hypothesis states: H1a: In the periods leading up to bankruptcy corporate insiders will engage in less net trading activity as compared to corporate insiders in firms that ultimately survive financial distress. Results supporting our hypothesis indicate that in the quarters before corporate bankruptcy corporate insiders substantially curtail their net trading activity with lower absolute values of the number of shares purchased minus the number of shares sold scaled by total shares outstanding as compared to the matched sample firms. 2.2 Hypothesis 1b Trading by Specific Groups of Institutional Shareholders Before Bankruptcy Initial research investigating the informational role of institutional shareholders in the market focused on institutional investors in the aggregate (Gompers and Metrick 2001; Nofsinger and Sias 1999; Cai and Zheng 2004; Bennett Sias and Starks 2003). In specific reference to the reaction of institutional shareholders to impeding firm bankruptcy using proprietary ASX data Frino et al (2014) find that aggregate institutional shareholders gradually sell their interest in impending bankrupt firms starting approximately 115 days before bankruptcy. 8

10 Instead of examining aggregate institutional ownership research from Yan and Zhang (2009) along with Bushee (1998) and Bushee (2001) suggest that institutional shareholders are not a homogeneous group. While institutional shareholders may share some common characteristics Yan and Zhang (2009) point out that institutional shareholders may have different investment horizons different investment objectives and styles and different fiduciary or legal restrictions. Therefore instead of examining aggregate institutional ownership it is important to identify specific subcategories of institutional shareholders that have demonstrated superior information processing capabilities based on prior research. Therefore we examine the abnormal trading activity of institutional investors with a short-term investment horizon (Yan and Zhang 2009;Bushee 1998) top performing institutional investors (Adebambo et al. 2015) and those institutional investors with fiduciary responsibilities (DelGuercio 1996) to determine if they demonstrate abnormal trading for bankrupt firms as compared with matched sample firms in the quarters leading up to bankruptcy Institutional Shareholders with Short-Term Investment Horizons Short-term institutional investors as categorized by Gaspar et al. (2005) and Yan and Zhang (2009) have high portfolio churn ratios. Along with short-term institutional shareholders transient institutional shareholders as classified by Bushee (1998) generally have high portfolio turnover they also purchase small equity interests in a large number of firms focus on shortterm information and engage in trading behavior based on momentum strategies such as selling firms with bad earnings news (Porter 1992 Bushee 1998). Yan and Zhang (2009) argue that institutional investors with superior information processing capabilities are likely to use their information advantage to trade frequently within a short investment horizon and earn positive abnormal returns. 9

11 Previous work from Ke and Ramalingegowda (2005) along with Yan and Zhang (2009) support this argument finding that institutional shareholders with a short-term investment horizon demonstrate superior information processing capabilities as compared to long-term institutions by earning larger positive abnormal returns. Further supporting this finding Puckett and Yan (2011) find that institutional shareholders that make inter-quarter trades also earn positive abnormal returns. Expanding on these findings Adebambo et al. (2015) find that shortterm institutional investors were able to use their superior information processing capabilities to shift their portfolios away from financial stocks before the financial crisis of Other research supports the notion that transient institutional investors as classified by Bushee (1998) engage in informed trade prior earnings announcements (Ali et al. 2008) breaks in earnings strings (Ke and Petroni 2004) and restatements (Hribar et al. 2009). This line of research supports a notion that transient institutional investors may also possess superior information processing. Thus we expect that both short-term and transient institutional investors will engage in superior information processing which will then inform their trades in the quarters leading up to a firm s bankruptcy setting. In the alternative due to their tendencies to make small investments in numerous firms short-term or transient institutional investors may have a higher risk tolerance forgoing sales of shares in distressed firms or even investing in distressed firms in the hope that they will be rewarded with a substantial payout in the case that the firm recovers. Ultimately it is unclear whether short-term and transient institutional investors will engage in abnormal trading behavior in the quarters leading up to a firm s bankruptcy filing Top-Performing Institutional Shareholders 10

12 Adebambo et al. (2015) defines top-performing institutional shareholders as institutions that have an alpha value (based on the Carhart (1997) four factor model) estimated over the previous 36 months that ranks in the top tercile of all institutions. Adebambo et al. (2015) found that top-performing institutional shareholders demonstrated foreknowledge of the Financial Crisis by shifting their portfolios away from financial stocks and into non-financial stocks. While Adebambo et al (2015) found that top-performing institutional shareholders displayed superior information processing capabilities for an industry-wide distress it is unclear whether Adebambo s (2015) findings would apply to individual firms in the pre-bankruptcy setting Institutional Shareholders Subject to Fiduciary Responsibilities Certain institutional investors are subject to fiduciary responsibility standards based on state regulations common law and the Employee Retirement Income Securities Act (ERISA). We classify institutional investors that report as banks insurance companies and other institutional shareholders such as pension funds and university endowments as institutional shareholders subject to fiduciary standards. The individual managers within institutions as well as the institutions themselves subject to fiduciary responsibility can be held personally liable for losses if they do not act prudently. Additionally as Badrinath et al. (1989) states court rulings mandate that these institutions cannot hide behind portfolio diversification but instead must evaluate each stock holding individually. 2 As a result institutional investors subject to fiduciary standards are likely to be more risk-averse and generally tilt their portfolios away from investments they view as volatile (Del Guercio 1996; Abarbanell Bushee and Raedy 2003). 2 Lakonishok et al. s (1992) evidence that pension funds do not succumb to herding behavior or positive-feedback trading provides support for the notion that institutional investors subject to fiduciary responsibilities rules do evaluate each stock holding individually. 11

13 Thus institutional investors with fiduciary responsibilities may invest significant resources to obtain firm specific information and in an effort to avoid catastrophic losses from individual firm bankruptcy. In contrast institutional shareholders subject to fiduciary standards may not exhibit abnormal trading when paired with a matched sample before firm bankruptcy. Generally institutional shareholders subject to fiduciary responsibilities are likely to have lower risk tolerance and as a result liquidate their equity interest in both the bankrupt firm and the matched sample firm at the first sign of financial distress. Unlike other institutional shareholders that may display abnormal trading due to superior information processing capabilities abnormal trading behavior by institutional investors subject to fiduciary responsibilities may be the result of lower risk tolerance associated with the prudent investor standards. We develop our next hypothesis which covers abnormal selling behavior expectations for each of our groups of institutional investor groups: H1b: Specific types of institutional investors (based on investment horizon past performance or fiduciary responsibilities) may possess superior information processing capability and will engage in abnormal selling behavior at least one quarter ahead of the bankruptcy. Results that support this hypothesis may be interpreted as being consistent with specific types of institutional investors possessing superior information processing capabilities and shifting their portfolios away from equity ownership in firms prior to bankruptcy. Thus support for this hypothesis indicates that abnormal trades of institutional investors potentially provides significant firm-specific information to the market as firms approach bankruptcy. 12

14 2.3 Hypothesis 2 Information Content of Shareholder Trading Behavior for the Campbell Bankruptcy Prediction Model Due to the potential for significant loss related to bankruptcy researchers have attempted to develop models to accurately predict the probability that a firm will file for bankruptcy. Altman (1968) uses a discriminant-ratio model to predict bankruptcy and develop the Z-score. Merton (1974) creates a distance to default measure. Ohlson (1980) develops the O-score. Shumway (2001) develops a hazard model to predict bankruptcy. Campbell et al. (2008) expends on these previous measures developing a bankruptcy predication model that includes a number of predictor variables. These variables generally include net income liabilities prior stock returns volatility of prior stock returns firm size percent of assets held in cash growth opportunities through the market-to-book ratio and stock price. However at the current time bankruptcy predication models like the one from Campbell et al. (2008) do not use information regarding previous trading behavior by corporate insiders or previous trading behavior by specific groups of institutional investors. As Gao et al. (2015) points out if managers have private information regarding their firm s future prospects and wish to limit their losses they may avoid purchasing additional shares while simultaneously limiting their selling behavior to avoid the potential litigation risk. Thus information regarding the absence or substantial reduction in trading activity may enhance Campbell et al. s (2008) bankruptcy prediction model by providing private firm-specific information that can be used to better discern the future prospects of distressed firms. This leads to our second hypothesis: H2a: The trading behavior by corporate insiders in the previous six months increases the predictive capabilities of Campbell et al. s (2008) bankruptcy prediction model for firms filing for bankruptcy in the future six months. 13

15 Results that support H2a indicate that reduction in the net trading activity by corporate insiders over the previous six months indicates an increased probability of firm bankruptcy in the future six months. After establishing that specific groups of institutional investors are able to demonstrate their superior informational processing capabilities by reducing their ownership percentages in firms prior to bankruptcy while retaining equity in firms that survive financial it is possible that a variable measuring the changes in ownership by certain types of institutional investors in the quarters leading up to bankruptcy may improve the future bankruptcy predictive abilities of the Campbell et al. (2008) bankruptcy prediction model. With respect to certain groups of institutional shareholders we hypothesize: H2b: Changes in ownership by specific types of institutional investors over the previous six months increases the predictive capabilities of Campbell et al. s (2008) bankruptcy prediction model for firms filing for bankruptcy in the future six months. Results that support H2b indicate that larger reduction in equity ownership by certain groups of institutional shareholders in the previous six months provides incremental information suggesting a higher probability of bankruptcy in the future six months. 3. Research design sample selection and data sources 3.1 Sample Selection Corporate Insider Sample For our corporate insider sample we begin with 2301 first time bankruptcies of publicly traded firms between 1992 and 2012 from the UCLA LoPucki Bankruptcy Research Database SDC Platinum Bankruptcy database and Compustat s Audit Analytics bankruptcy database. We 14

16 eliminate 746 bankrupt observations that fail to report quarterly financial statement information in Compustat or market return data in CRSP within two quarters of the bankruptcy filing date. From there we eliminate 968 observations that either failed to report insider ownership information or reported that corporate insiders owned 0% of the outstanding corporate stock at both one year before bankruptcy and two years before bankruptcy. After that we eliminate 230 observations that did not report data for the 11 consecutive quarters necessary to calculate our control variables we obtained our corporate insider ownership sample of 244 bankrupt observations. Table 1a summarizes the bankrupt sample selection process for the corporate insider sample Institutional Ownership Sample For our institutional ownership sample we once again identify an initial sample of 2301 first time bankruptcies of publicly traded firms between 1992 and 2012 from the UCLA LoPucki Bankruptcy Research Database SDC Platinum Bankruptcy database and Compustat s Audit Analytics bankruptcy database. We eliminate 746 bankrupt observations that fail to report quarterly financial statement information in Compustat or market return data in CRSP within two quarters of the bankruptcy filing date. Then we eliminate 721 bankrupt observations that did not report institutional ownership data from Thompson Financial. Finally we eliminate 468 bankrupt observations that did not have report data for the 11 consecutive quarters necessary to calculate our institutional ownership variables and our control variables for each quarter. After eliminating these firms 366 bankrupt observations remain. Table 1b summarizes the bankrupt sample selection process for the institutional shareholder sample Matched Sample Selection Process 15

17 Our initial sample of matched firms from contains firm quarter observations. We eliminate firm quarter observations that do not have at least one million in book value of assets and at least one million in market value of equity eight quarters (2-years) into the future to ensure the matched sample firms survive. Furthermore we eliminate firm quarter observations due to missing period institutional ownership financial statement data from Compustat or market return data from CRSP. Finally we eliminate observations that do not have the 11 consecutive quarters of data necessary to calculate changes in institutional ownership data necessary to calculate the probability of filing for bankruptcy within 24 months (as calculated by Campbell et al. 2008) or data necessary to calculate our control variables from Compustat and CRSP. We are left with potential firm quarter matching observations. In order to identify the appropriate quarter to begin the collection of data for a matched sample we first identify a period prior to observable changes in distress. Seyhun and Bradley (1997) indicate that firms that file for bankruptcy generally experience a drop in price and earning performance starting approximately two years before the bankruptcy. As a result we use a matched sample design similar to Ramalingegowda (2014). For each bankrupt firm we complete a matching without replacement design based on year industry (1-digit SIC code) and probability of bankruptcy within the next 24 months as calculated by Campbell et al. (2008). We then choose the matching firm quarter observation that minimizes the difference between the Campbell et al. (2008) model s bankruptcy probability for the bankrupt firm and the matched sample firm. This procedure controls for financial distress eight quarters before bankruptcy. We collect 11 quarters of data for both the bankrupt firms and the matched sample firms yielding the 2684 firm quarter observations for the corporate insider ownership sample and the 4026 firm quarter observations for the institutional ownership sample. 16

18 Each of our variables in our first set of analyses are calculated as the difference between the value for that particular variable for the bankrupt firm minus the value for that particular variable for the matched sample firm for each of the 11 quarters before bankruptcy. See Appendix A for a summary of the Campbell et al. (2008) model for calculating the probability of filing for bankruptcy within the future 24 months. (Insert Table 1 Here) Our bankrupt and matched sample observations for the institutional ownership sample and the corporate insider ownership sample originate from a variety of industries. Our largest number of bankruptcies originate from manufacturing (SIC = 3) and from wholesale and retail trade (SIC = 5). In terms of years we find a significant number of bankruptcies in both our institutional ownership sample and our insider ownership sample between 2000 and 2002 along with 2008 and These findings are consistent with downturns in the U.S. economy. In our institutional ownership sample we have 217 bankrupt observations from firms that filed for bankruptcy within one quarter of their last reported quarterly financial statement data in Compustat or market return data in CRSP. In contrast we have 147 bankrupt observations that filed for bankruptcy within two quarters of their last reported quarterly financial statements in Compustat or market return data in CRSP. (Insert Table 2 Here) 3.2 Corporate Insiders Using our matched sample design we examine the changes and levels of corporate insider ownership for our bankrupt firms and our matched sample firms in the quarters leading up to bankruptcy. To obtain ownership data on corporate insiders we use the Thompson Reuters Insiders Data from WRDS. We define corporate insiders as: Chairman of the Board (CB) Chief Executive 17

19 Officer (CEO) Chief Operating Officer (CO) General Council (GC) Chief Financial Officer (CFO) and President (P). Following Lakonishok and Lee (2001) and Adebambo et al. (2015) we define the variable NetInsiderActivty as shares purchased by corporate insiders minus shares sold by corporate insiders over the fiscal quarter scaled by total shares outstanding at the beginning of the quarter. If corporate insiders did not buy or sell any stock for the quarter NetInsiderActivity is set equal to zero. Since we are interested in level of net trading by corporate insiders in the periods leading up to bankruptcy we create the variable ABS(NetInsiderActivity) by taking the absolute value of NetInsiderActivity. Larger values of ABS(NetInsiderActivity) indicate that corporate insiders either had larger net purchases or net sales for the quarter. Instead if corporate insiders substantially curtailed or even ceased trading in anticipation of firm bankruptcy then the value of ABS(NetInsiderActivity) should be close to zero. 3.3 Institutional Ownership We use data provided by Thompson Financial to calculate institutional ownership as the percentage of shares owned of each firm by institutional shareholder. The Securities and Exchange Commission (SEC) requires all institutional shareholders with more than $100 million in total equities to report individual equity positions in excess of shares or $ in value to the SEC at the end of each quarter. With this information we calculate institutional ownership for each firm as the number of shares held by institutional investors at the end of each quarter (March 31 June 30 September 30 and December 31 of the calendar year) divided by the total number of shares outstanding on each of those dates. Firms reporting institutional ownership in excess of 100% are limited to institutional ownership percentages of 100%. Our first set of tests examines the abnormal trading behavior of institutional shareholders with a short-term investment horizon. To identify those institutions with a short-term investment 18

20 horizon we take two different approaches. First we follow the procedures outlined in Yan and Zhang (2009) and calculate the churn ratio for each institution. We classify an institutional shareholder as short-term if they have a one year average churn ratio in the top tercile of all institutions (See Appendix C for the calculation of the Yan and Zhang (2009) Churn Ratio). Next we use the Bushee (1998) institutional shareholder classification system and specify Transient shareholders as institutions with a short-term investment horizon. We then find ownership percentages and changes in ownership percentages for these two categories for each firm in the bankrupt sample and the matched sample. To find top performing institutional shareholders we follow the procedures outlined by Adebambo et al. (2015) and classify an institution as a top performing institutional investor if the institution s alpha value as calculated from the Carhart (1997) four factor model over the previous 36 months is in the top tercile among all institutions. Finally we classify institutional shareholders who report their status as: banks and trust companies (BNK) insurance companies (INS) or miscellaneous (which includes corporate pension funds public pension funds university endowments and various other types of institutional shareholders) as institutional investors subject to fiduciary responsibilities. From our fiduciary responsibility category we eliminate any institution classified as a hedge fund Empirical model For our tests of abnormal trading for corporate insiders we use a matched sample model similar to Ramalingegowda (2014). Specifically we use the following model: dabs(netinsideractivity) = α + β1 QTR0 + β2 QTR1 + β3 QTR2 + β4 QTR3 + β5 QTR4 + β6 QTR5 + β7 QTR6 + β8 QTR7 + β8 dcampbellprob + β9 dmom_t3_t1 + β10 dmom_t6_t4 + β11 dmom_t12_t7 + β12 dtvol_t6_t4 + β13 dprice + β14 dlnmve + β15 dbmratio + β16 ddivyield + β17 dinsiderown_per + Fixed Effects Indicator Variables + ε. 3 We thank Chris Clifford from the University of Kentucky for providing us a list of Hedge Funds. 19

21 In addition for our tests involving the abnormal trading for specific groups of institutional shareholders we once again use a matched sample design model similar to Ramalingegowda (2014). Our model for specific categories of institutional shareholders is: dδinstown_per = α + β1 QTR0 + β2 QTR1 + β3 QTR2 + β4 QTR3 + β5 QTR4 + β6 QTR5 + β7 QTR6 + β8 QTR7 + β8 dcampbellprob + β9 dmom_t3_t1 + β10 dmom_t6_t4 + β11 dmom_t12_t7 + β12 dtvol_t6_t4 + β13 dprice + β14 dlnmve + β15 dbmratio + β16 ddivyield + β17 dinstown_per + Fixed Effects Indicator Variables + ε. where: dabs(netinsideractivity) is the difference in the absolute value of ABS(NetInsiderActivity) for the bankrupt firm minus the absolute value of ABS(NetInsiderActivity) for the matched sample firm. dδinstown is the change in ownership by a particular category of institutional investor for the bankrupt firm minus the change in ownership by a particular category of institutional investor for the matched sample for that particular quarter. For our second set of results we use the following different groups of institutional shareholders: institutional investors with a short-term investment horizon separated in to two groups short-term institutional investors as defined by Yan and Zhang (2009) and transient institutional shareholders as identified by Bushee (1998); top performing institutional shareholders; and institutional shareholders subject to fiduciary standards. Each of our variables is calculated as the difference between the change in ownership percentage for a particular category of institutional shareholder for the bankrupt firm and the change in ownership percentage for the corresponding category of institutional shareholder for the matched sample firm for one of 11 different quarters Independent variables of interest In our matched research design our variables of interest are: 20

22 QTR0 Equals 1 if the observation occurred on the day the firm last reported data before filing for bankruptcy QTR1 Equals 1 if the observation occurred approximately days before the last reporting period in which the firm reported financial statement data and market return data QTR2 Equals 1 if the observation occurred approximately days before the last reporting period in which the firm reported financial statement data and market return data QTR3 Equals 1 if the observation occurred approximately days before the last reporting period in which the firm reported financial statement data and market return data QTR4 Equals 1 if the observation occurred approximately days before the last reporting period in which the firm reported financial statement data and market return data QTR5 Equals 1 if the observation occurred approximately days before the last reporting period in which the firm reported financial statement data and market return data QTR6 Equals 1 if the observation occurred approximately days before the last reporting period in which the firm reported financial statement data and market return data QTR7 Equals 1 if the observation occurred approximately days before the last reporting period in which the firm reported financial statement data and market return data Observations for the remaining three quarters (days ) before the last reporting period in which the firm reported financial statement data and market return data are part of the leave out sample and do not have indicator variables. Unreported univariate results indicate few differences in the change in various categories of institutional ownership for the bankrupt observations and the matched sample observations. A negative coefficient on any of these indicator variables indicates a larger abnormal reduction in that particular category of institutional ownership for the bankrupt firm as compared to the abnormal change in institutional ownership for the matched sample firm for that quarter Control variables Prior research provides evidence that numerous variables influence changes in investment by a particular category of institutional investor or corporate insider in a particular firm. We control for several of these variables in our matched sample design. The control variables that we use are: CampbellProb - Beginning of the quarter probability of the firm declaring bankruptcy within the future 24 months measured using the Campbell et al. (2008) bankruptcy prediction model 21

23 MOM_t3_t1 - Buy-and-hold raw return for the firm in the current quarter (months m-1 m-2 and m-3) MOM_t6_t4 - Buy-and-hold raw return for the firm in the previous quarter (months m-4 m-5 and m-6). MOM_t12_t7 - Buy-and-hold raw return for the firm in the previous half year (months m-7 m-8 m-9 m-10 m-11 and m-12). TVOL_t6_t4 - Previous quarter average monthly trading volume divided by shares outstanding (months m-4 m-5 and m-6). PRICE - Price of the firm at the end of the current quarter. LnMVE Log of the Market Value of Equity for the previous quarter. BMRatio - Book value of equity divided by market value of equity measured at the beginning of the quarter DivYield Prior quarter divided scaled by beginning of the period market value of equity InsiderOwn_Per - The percentage ownership by corporate insiders for the beginning of the quarter InstOwn - The percentage ownership by the particular category of institutional shareholder for the beginning of the quarter. In terms of control variables we include a variable that measures the probability of the firm going bankrupt within the next 24 months. While Ramalingegowda (2014) uses the Shumway (2001) model to predict firm bankruptcy 24 months into the future we use the Campbell et al. (2008) model to calculate the probability the firm will file for bankruptcy within the future 24 months (see Appendix B for a discussion of calculating the probability of bankruptcy 24 months into the future as calculated by Campbell et al. (2008)). Following Ramalingegowda (2014) and Ke and Petroni (2004) we include control variables for a firm s return over the current quarter 22

24 (MOM_t3_t1) the return from the previous quarter (MOM_t6_t4) and the firm s return in the previous half year (MOM_t12_t7). Ke and Petroni (2004) find that returns over the prior year affect changes in the firm s ownership by transient and dedicated institutional shareholders. Ke and Petroni (2004) also find that firms with larger market values of equity generally have larger increases in both transient and dedicated institutional ownership percentage from one quarter to the next. Consistent with Ramalingegowda (2014) we also include control variables for prior quarter average monthly trading volume (TVOL_t6_t4) and end of the quarter stock price (Price) to proxy for stock liquidity. We also include a control variable for the firm s prior quarter book-to-market ratio (BMRatio) because prior research from Bushee (2001) and Ke and Petroni (2004) find that some types of institutional shareholders have a preference for value versus growth firms. Finally we control for the dividend yield of each of the firms. 3.5 The Modified Campbell et al (2008) Bankruptcy Prediction Model To test our second hypothesis we replicate the Campbell et al. (2008) bankruptcy prediction model and add additional terms for the absolute value of changes in the percentage ownership by corporate insiders over the previous six months or changes in ownership by specific groups of institutional investors. Consistent with Campbell et al. (2008) we define the indicator variable Failure as equal to 1 if the firm filed for bankruptcy had a financially driven delistings or D ( default ) ratings issued by a leading credit rating agency within the future 6 months. Otherwise the value of Failure is set equal to 0. Specifically consistent with Campbell et al. (2008) we use the following logit models: Failure = α + β1 NIMTAAVG + β2tlmta + β3 EXRETAVG + β4sigma + β5rsize + β6cashmta + β7mb + β8price + β9abs(netinsideractivity) + ε Failure = α + β1 NIMTAAVG + β2tlmta + β3 EXRETAVG + β4sigma + β5rsize + β6cashmta + β7mb + β8price + β9change0_instown_per + β10change1_instown_per + ε 23

25 Where the variables defined by Campbell et al. (2008) are: NIMTAAVG it is the geometric decaying average of (Net Income / (Market Value of Equity + Total Liabilities)) TLMTA it is Total Liabilities / (Firm Market Equity + Total Liabilities) EXRETAVGit = log(1 + Rit ) log(1 + RS&P500t ) where R is the return on the stock for the month and RS&P500 is the return on the S&P 500 Index for the month. This variable is calculated as the geometric decaying mean for the previous 12 months. SIGMAit 1t 3 is the computed as an annualized 3-month rolling sample standard deviation which is calculated as = (252 1 r N 1 k (t 1t 2t 3 ik 2 ) RSIZE = log (Market Value of Equity / Total S&P 500 Market Value of Equity) CASHMTA = Cash and Short-Term Investments / (Market Value of Equity + Total Liabilities) MB = Market-to-Book ratio of the firm Price = log of price per share winsorized above $15 In addition to the variables calculated by Campbell et al. (2008) we include additional variables to test our second hypothesis: Abs(NetInsiderActivity) The absolute value of the number of shares purchased minus the number of shares sold by corporate insiders over the previous six months divided by the number of shares outstanding at the beginning of the period. Change0_InstOwn_Per The change in institutional ownership for a particular category of institutional shareholder in the current quarter Change1_InstOwn_Per The change in institutional ownership for a particular category of institutional shareholder in the previous quarter. 24

26 4. Results 4.1 Descriptive Statistics Table 3 presents our descriptive statistics for our final sample. We present mean values for each of our variables for QTR0 QTR1 QTR2 QTR3 and QTR8. The period QTR8 is the period in which we match each bankrupt firm with a matched sample firm. Insert Table 3 The descriptive statistics presented for QTR 8 indicates very few significant differences between the means of the variables for the bankrupt firms and the means of the variables for the matched firms. Both the change in particular categories of percentage of institutional ownership and the beginning of the period level of institutional ownership for each of our institutional ownership classifications is generally similar. Since our matched procedure is done based on QTR8 the lack of differences between bankrupt firms and matched firms is consistent with the matching procedure controlling for future predicted financial distress. Eight quarters before bankruptcy we find that the mean probability based on the Campbell et al. (2008) model of filing for bankruptcy for the bankrupt sample of firms is 5.39% while the mean probability of filing for bankruptcy for the matched sample of firms is 5.13%. Our t-test reports that this was not a significant difference between the mean probability of filing for bankruptcy within the bankrupt group and the matched sample group. We find that in QTR8 or 2 years before bankruptcy the matched sample firms have slightly higher stock prices slightly higher trading volume higher slightly larger market value of equity and smaller corporate insider ownership. In Table 3 Panel A we find less corporate insider trading and larger negative changes in various categories of institutional ownership for the bankrupt sample as compared to the matched sample in the quarters leading up to bankruptcy. While the univariate statistics presented in 25