Informational efficiency and market reaction to bankruptcy announcement

(1)

“Informational efficiency and market reaction to bankruptcy

announcement”

Name: Reinout Mensing Student Number: 10084525

December 2015

Supervisor: Vladimir Vladimirov

(2)

Statement of Originality

This document is written by Reinout Mensing who declares to take full responsibility for the contents of this document.

I declare that the text and work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Abstract

This thesis researches price reactions to bankruptcy announcements and the influence of media and analyst coverage on informational efficiency. Using event study analysis on a sample of US bankruptcy cases between 2000 and 2013, this thesis finds cumulative abnormal returns of -33.12% within a window of seven trading days surrounding the bankruptcy filing date. These abnormal returns are concentrated on the filing date and its consecutive trading day, which together account for -25.95% abnormal returns. Firms that have been mentioned in the Wall Street Journal as a possible bankruptcy case, report lower losses around the filing date, indicating improved market anticipation of bankruptcy. In addition, this thesis finds evidence for analyst recommendations influencing the magnitude of abnormal returns and the existence of market inefficiencies when dealing with corporate bankruptcy.

(4)

1. Introduction

Corporate bankruptcy of publically listed firms largely impacts financial markets and has severe consequences for all involved stakeholders. The most recent financial crisis has stressed the importance of monitoring stocks of financially distressed firms. Shareholders and bondholders of financially distressed firms are likely to incur large losses is the period preceding bankruptcy (Lang & Stulz, 1992). When the firm eventually files for bankruptcy, investors stand to lose even more value. The announcement of bankruptcy filing is commonly accompanied by significant negative abnormal returns. Clark and Weinstein (1983) reported average cumulative abnormal returns of -47% over the three day interval around the bankruptcy date and Lang and Stulz (1992) found average cumulative abnormal returns of -22% on the day before bankruptcy and on the filing date. The magnitude of abnormal returns indicates the degree to which financial markets are surprised by the bankruptcy. High negative abnormal returns indicate a low levels of market anticipation, whereas low or zero abnormal returns indicate that most of the information regarding bankruptcy filing was already incorporated into the stock price.

The Efficient Market Hypothesis (EMH) states that security prices always reflect all available information (Fama, 1970). According to advocates of the EMH, stock prices do not systemically deviate from their fundamental value. In an efficient market, arbitrage serves as a corrective mechanism to quickly incorporate new information into stock prices. The EMH assumes that security prices reflect all available information and it is therefore impossible to consistently make economic profits based on a specific information set (Jensen, 1978). Behavioural ecnomists however, explain that investors can suffer from biases and that arbitrage does not always revert prices to fundamental value. According to behavioural economists, security prices are at least partially predictable and can systemically deviate from fundamental value (Shleifer, 2000). Previous research on bankruptcy announements and valuation of bankrupt stocks has found significant evidence for market inefficiencies (Hubbard & Stephenson, 1997).

This thesis builds upon existing research on stock price reactions to bankruptcy announcements by using an up-to-date dataset that considers corporate bankruptcies in the United States from 2000-2013. By employing event studies with different time windows, this paper contributes to the ongoing debate about market efficiency. Even though numerous academics have studied bankruptcy announcements, there is still a gap in the understanding of the magnitude of price reactions to bankruptcy filings. This thesis contributes to existing

(6)

research by exploring this gap and investigating the influence of analyst and media activity on price reactions to bankruptcy announcements. Media sources such as the Wall Street Journal are in the position to influence stock prices because their articles are widely disseminated and generally relied on by financial markets (Dawkins & Rose(Green, 1998). Analysts provide recommendations and forecast future earnings and stock values and are therefore in the position to influence share prices as well. Using cross-sectional regression analysis on the abnormal returns resulting from the event study, this paper establishes a relation between analyst and media coverage and abnormal returns associated with bankruptcy filing. In addition, a long-term event study using Buy-And-Hold Abnormal Returns (BHAR) is employed to research the influence of analysts and media on the timing and speed of shareholder loss accumulation of firms on the verge of bankruptcy. The research question that this thesis addresses is therefore:

What is the influence of analyst and media coverage on informational efficiency and the magnitude of abnormal returns associated with bankruptcy announcement?

This thesis finds cumulative abnormal returns of -33.12% within a window of seven trading days surrounding the bankruptcy filing date. These abnormal returns are concentrated on the filing date and its consecutive trading day, which together account for -25.95% abnormal returns. Furthermore, this thesis finds evidence for analyst and media coverage influencing the magnitude of abnormal returns and the existence of market inefficiencies when dealing with corporate bankruptcy. These findings could incentivize firms to re-evaluate their relationship with analysts and media. Especially for financially distressed firms, negative impact on stock prices could force firms into bankruptcy. The findings of this thesis could lead to new insights for security analysts and investors who are constantly looking for predictable patterns.

This paper is structured as following: Section 2 discusses relevant literature about the efficient market hypothesis, bankruptcy and analyst and media coverage. Section 3 provides a detailed description of the research methodology and hypotheses. Section 4 reports descriptive statistics of the dataset and lists all data sources that were used in this thesis. Section 5 reports and discusses the results of the data analysis. Section 6 employs several robustness checks to examine the validity of the results. The thesis is discussed and concluded in section 7. Section 8 lists all academic papers that were used in this thesis and section 9 provides a summary of key coding that was used to produce the empirical results.

(7)

2. Literature Review

In order to understand bankruptcy-related stock price movements it is essential to understand what security prices are based on and how markets value financial assets. Furthermore, it is of importance to understand how markets process news and incorporate it into security prices. This section reviews existing literature on capital market efficiency and studies the implications of corporate bankruptcy. Moreover, this section examines the influence of analysts and media on share prices and reviews academic papers on bankruptcy announcement and associated stock returns.

2.1 Capital Market Efficiency

Capital markets exist to facilitate interaction between providers and users of financial capital. In the primary market companies sell new stocks or bonds to initial buyers such as banks, pension funds and other investors. In the secondary market, investors trade securities. In an efficient capital market security prices reflect their true value. The efficient market hypothesis (EMH) assumes that at any time security prices reflect all available information (Fama, 1970). Efficient capital markets allow investors to allocate their resources under the assumption that security prices do not deviate from their true value. According to Jensen (1978) it is impossible to consistenly make economic profits by trading based on a specific information set in an efficient market. Since all available information is already incorporated in the price of an asset it is impossible to predict future stock price movements based on this information. The EMH is in line with the random walk theory which states that succcessive price movements are independent from each other, meaning that stock price history cannot be used to predict future stock prices (Fama, 1995). An underlying assumption is that if returns were predictable, investors would use their forecasts to generate unlimited profits (Timmermann & Granger, 2004). If asset mispricing occurs, arbitrageurs will immediately assume a position in the asset that exploits the mispricing, causing prices to almost immediately adjust to new levels corresponding to the new true value of the financial asset. The EMH therefore suggests that investors cannot consistently beat the market and that financial analysis is therefore only useful if an analyst has new and private information. When new and private information arrises that is relevant to the value of a stock, trading activity in an efficient market makes sure that markets incorporate the new information and that stock prices quickly move to their new intrinsic value.

(8)

Three forms of market efficieny commonly prevail for existing literature on the Efficient Market Hypothesis. Markets are weak form efficient if security prices reflect all historical price information and future market returns are independent of historical returns. Semi-strong form efficiency implies that security prices reflect all publicly available information and stock prices quickly adjust when new information becomes available. In strong form efficient markets, security prices reflect all public and private information (Timmermann & Granger, 2004). Market efficiency relies on underlying market conditions. Firstly there are no transaction costs for trading securities. Secondly all information is costlesly available to all market participants and thirdly all market participants must agree on the implications of current information on current prices and the distribution of future returns (Fama, 1970).

The efficient market hypothesis has been frequently challenged by academics and behavioural economists who believe stock prices can systematically deviate from their true value and that returns are at least partially predictable. A major critique on the EMH is that it expects market participant to behave rationally and to be able to value securities correctly (Malkiel, 2003). Advocates of the EMH counter this critique by explaining that markets can still be efficient when not all participants are rational. As long as the irrational market participants trade randomly and their trading strategies are uncorrelated, the effects of irrational market participants are likely to cancel each other out resulting in prices that are still close to their fundamental value (Shleifer, 2000).

Behavioural finance explains how systemic market mispricing occurs and how it can be attributed to behavioural and psychological elements. Furthermore, behavioural finance suggests that driving forces of market efficiency such as arbitrage are not as strong as advocates of EMH suggest. In their paper on the limits of arbitrage, Shleifer & Vishny (1997), find evidence that arbitrage is not always completely effective at reverting security prices to their fundamental value. Especially in extreme circumstances, professional arbitrageurs may avoid volatile arbitrage positions due to the risk of losses and the need to liquidate their portfolio under pressure of investors in their fund. Moreover behavioural finance expects significant and systemic deviations from effiency to persist for long periods of time (Shleifer, 2000). For example De Bondt & Thaler (1985) found that stock markets tend to overreact to events or news and found substantial evidence for weak form efficiency. Their paper suggests that markets do not respond rationally to news and provides evidence for the existence of anomalies in financial markets.

(9)

2.2 Corporate Bankruptcy

Companies filing for bankruptcy have likely been in financial distress for an extended period of time. When choosing a restructuring mechanism, firms consider the costs of restructuring and the extend to which financial issues will be resolved (Eckbo, 2008). Default occurs when companies are no longer able to meet debt obligation to their creditors. This could entail failure to make interest payments or to pay back principal as stated in the debt contract. Debtor and creditors can arrange out of court settlement in which they renegotiate contractual terms. Often firms first attempt to resolve their financial issues informally before filing for bankruptcy (Franks & Torous, 1994). However, when companies in default are unable to resolve financial issues, the debtor and creditors who meet certain conditions can file for bankruptcy. Firms filing for bankruptcy have the choice between liquidating under Chapter 7 or reorganizing under Chapter 11 of the US bankruptcy code (White, 1989).

Chapter 11 of the US bankruptcy code allows companies to reorganize under protection from its creditors. Throughout the reorganization, companies can continue their operations and are exempted from payments to debt holders. The debtor may also obtain debtor-in-possession financing, which provides a new line of credit to finance routine business operations (Eckbo, 2008). This provides firms the opportunity to reorganize its business in order to make it viable again. The bankruptcy petition can be filed voluntarily by the debtor or it can be filed by creditors who meet certain conditions, in which case it is called an involuntary bankruptcy (Bris, Welch, & Zhu, 2006). Companies operating under chapter 11 protection have 120 days to file their reorganization plan and to approach their creditors in an attempt to renegotiate the terms on outstanding debt such as payment schedules, interest payments and dollar value. Repayments are typically lower than stated in the original debt contracts. Creditors have incentive to cooperate with the debtor since the outcome of Chapter 11 is often more favorable for creditors than its alternative, chapter 7 (White, 1989).

Chapter 7 of US bankruptcy code liquidates assets in order to pay creditors. Firms that file for chapter 7 bankruptcy are past the stage of reorganizing and must liquidate assets to pay off creditors. In chapter 7 liquidation, creditors collect their outstanding loans following the absolute priority rule. Creditors who own secured debt are paid based on the value of the secured asset. Remaining assets and cash are distributed to creditors in order of seniority. In case of Chapter 7 liquidation, creditors often recover less value than in case of Chapter 11 reorganization. Chapter 7 liquidation could lead to fire-sales of assets, causing assets not to realize their full value due to unfavorable market conditions or due to the immediacy of the

(10)

asset sales (Shleifer & Vishny, 1992). Moreover, Chapter 11 reorganization could preserve going-concern value that would cease to exist under Chapter 7 liquidation. Even though Chapter 11 can prevent firms from declaring total bankruptcy and possible preserves more value for creditors, the outlook for bondholders and shareholders is still unfavorable. Direct costs associated with reorganization under Chapter 11 are estimated to be 6.5% of the book value of assets and bondholders and shareholders are expected to incur substantial losses (Altman & Hotchkiss, 2006).

When companies file for Chapter 11, their share prices commonly drop significantly. Companies in Chapter 11 are often delisted from major stock exchanges after failing to meet listing standards and can continue trading on the OTCBB (Over-The-Counter Bulletin Board). Share prices show significant declines around bankruptcy announcement because shareholders are last in order of seniority after unsecured and secured creditors. According to the SEC, most reorganization plans under Chapter 11, cancel existing equity shares. In reorganization plans where shareholders do participate, shares are often substantially diluted. Furthermore, shareholder do not receive dividends during reorganization. As part of the reorganization plan, bankrupt companies can issue new stock. Shareholders may possibly trade their cancelled stock for newly issued stocks. The reorganization plan decides upon the rights and payments for shareholders (Bris, Welch, & Zhu, 2006).

2.3 Market Reaction to Bankruptcy Announcement

Bankruptcy filing may convey new information about the bankrupt firm or the industry it operates in. When the bankruptcy filing is not completely anticipated by the market, this could lead to reassessment of the firm’s value resulting in abnormal returns around bankruptcy announcement. When companies file for Chapter 11, this usually results in significant negative abnormal returns, indicating strong negative informational content (Datta & Iskandar-Datta, 1995). Not only shareholders of the bankrupt firm are affected by Chapter 11 filing. Other stakeholders such as rival firms, client firms and creditors are affected as well (Hertzel, Li, Officer, & Rodgers, 2008). In their study on contagion effects of bankruptcy announcements, Lang & Stulz (1992), found that on average, bankruptcy announcements decrease the value of a value-weighted portfolio of competitors by 1%.

Bankruptcy of a publicly listed firm is an impactful event that affects many different stakeholders. Stock prices of firms on the verge of bankruptcy often reflect the unsecure nature of the bankruptcy event. The market’s assessment of the probability of bankruptcy filing is an important factor for stock price valuation in the period preceding a Chapter 11

(11)

filing. Commonly, shareholders experience significant losses for long periods preceding bankruptcies due to poor performance and financial distress (Clark & Weinstein, 1983). In their paper on indirect costs of financial distress and bankruptcy law, Sautner & Vladimirov (2014) found that once firms start accumulating indirect distress costs, default and bankruptcy become more likely. This reinforces bankruptcy fears of stakeholders which can be incorporated into share prices. Furthermore, the potential outcome for shareholders in case of bankruptcy is a driving force behind share prices of distressed firms. For example when creditors are relatively weak, managers can extract more value for shareholders in case of bankruptcy (Sautner & Vladimirov, 2014). More value for shareholders in case of bankruptcy could translate into higher share prices.

An empirical study by Gilson, Hotchkiss, & Ruback (2000) found evidence that there can be great dispersion between the marketvalue of bankrupt stocks and the value of the bankrupt firms based on the estimated cash flow projections. Even though every bankruptcy case is unique and has different implications for shareholders, studies have shown that shareholders consistently overestimate the performance of the stocks they hold in bankrupt firms. Stock prices grossly overstate the actual provisions for shareholders as stated in the reorganization plan (Hubbard & Stephenson, 1997). This indicates that markets do not completely comprehend the implications of bankruptcy with respect to shareholder value, suggesting the existence of market inefficiencies.

Several studies have researched the announcement effect of bankruptcy filing on stock prices. The announcement of bankrupty filing can convey important information about value and risk of common shares. The announcement signals probabilitites of alternative future share value. For example when a firm files for Chapter 11, the probability that its common shares become worthless increases (Clark & Weinstein, 1983). The value of shares is largely determined in court as the reorganization plan states what shareholders can expect to receive. The magnitude of negative returns might be larger for firms with high levels of debt. Since debtors have seniority over shareholders, bankruptcy of highly leveraged firms leaves its shareholder with little value (Clark & Weinstein, 1983). When bankruptcy is completely anticipated by the market and does not convey new information, abnormal returns around the bankruptcy announcement should be zero according to the Efficient Market Hypothesis. If the bankruptcy is not completely anticipated by the market, abnormal returns are expected to occur around the bankruptcy filing.

(12)

The degree to which the market anticipates the bankruptcy determines the magnitude of abnormal returns. High abnormal returns indicate a surprise-effect and low or zero abnormal returns indicate that the information conveyed in the bankruptcy announement was already incorporated into share prices (Dawkins & Rose(Green, 1998). The informational content of Chapter 11 filing and associated abnormal return is therefore expected to be larger if there has been little signaling about possible bankruptcy or poor performance in the period preceding the bankruptcy. Clark and Weintstein (1983) emphasize in their research that shareholders lose large amounts of money during the month in which the bankruptcy occurs and find that these losses are mainly concentrated in the three-day trading interval around the bankruptcy date.

Previous event studies on bankruptcy announcement have found significant abnormal returns around the Chapter 11 filing date. Clark and Weinstein (1983) reported average cumulative abnormal returns of -47% over the three day interval around the bankruptcy date. Lang and Stulz (1992) found average cumulative abnormal returns of -22% on the day before bankruptcy and on the filing date. Both these studies however do not investigate the possible effect of analyst and media coverage on the magnitude of the abnormal returns associated with bankruptcy announcement. Dawkins and Rose-Green (1998) investigate the relationship between prior announcement of possible bankruptcy and price reactions to subsequent Chapter 11 filings. They find that firms with a prior announcement in the Wall Street Journal experience significantly smaller price reactions at the event date when compared to firms without such an announcement. This is consistent with semi-strong form market efficiency.

2.4 Analyst Influence on Share Prices

Financial markets are constantly monitored by financial analysts. Analysts are considered to be informed market participants and their analysis and forecasting influences other market participants. Financial analysts participate in trading and trade-generating activities and are in the position to influence the amount of firm-specific, industry-level and market-level information impounded into stock prices (Piotroski, 2004). Disclosure of information could lead to securities trading closer to their fundamental value. In case of bankruptcies, the surprise effect of bankruptcy filing is expected to be lower when more information is incorporated into stock prices. Analyst influence on stock prices is dependent on several factors. Determinants of analyst impact include strength of recommendation, analyst reputation and size of the recommended firm (Piotroski, 2004).

(13)

A study by Doukas, Kim and Pantzalis (2005) finds that positive excess analyst coverage is associated with overvaluation and low future returns. The underlying reasoning is that excessive analyst coverage, driven by the incentives of investment banks or the analysts’ self-interest, raises optimism among investors leading to stock prices trading above their fundamental value. On the other hand, weak analyst coverage increases the probability of information asymmetries. Moreover, weak analyst coverage causes stock prices to trade below fundamental value (Doukas, Kim, & Pantzalis, 2005).

The intensity of analyst coverage that firms receive is dependent on several factors. In a study about firm characteristics that determine analyst following, Bhushan (1989) finds significant statistical evidence that the number of analysts following a firm is determined by firm size, insitutional share holdings and return variability of the firm. Larger firms receive more analyst coverage. Furthermore there is more demand for analyst coverage for firms with higher return variability and higher institutional holdings. This is due to higher possible trading profits for firms with high return variability and because larger institutional holdings can justify the costs of analyst services, while for small investors this expenditure might not be justified (Bhushan, 1989).

Several studies have shown that analysts can influence share prices and market sentiment. Brennan, Jegadeesh, & Swaminathan (1993) find evidence that analysts also influence the speed of adjustment of share prices. They find that firms with a higer number of analysts following respond more rapidly to market returns. Moreover, the marginal effect of the number of analysts on the speed of share price adjustment increases with the number of analysts. Analysts could therefore have a significant impact on abnormal returns around bankruptcy announements. Analysts can increase the amount of information impounded into stock prices and the speed of share price adjustments. These findings suggest that strong analyst coverage could lead to smaller abnormal returns around bankruptcy filings. However, as Doukas, Kim and Pantzalis (2005) found, strong analyst coverage could also lead to optimism in investor sentiment and overvaluation of shares.

2.5 Media Influence on Share Prices

Mass media, such as newspapers, have the potential of reaching a large audience and therefore play an important role in disclosing information. Mass media can resolve informational friction and affect share prices even if it does not supply genuine news (Fang & Peress, 2009). News in newspapers such as the Wall Street Journal is widely disemminated and generally relied on by capital market participants (Dawkins & Rose(Green, 1998).

(14)

Publication of information could therefore lead to improved investors’ assessment of specific firms and industries. The publications of news events leads to improved availability of information and causes some investors to react more quickly. Several studies have researched the influence of media on stock prices and found significant evidence that media affect stock prices and contribute to market efficiency.

Fang & Peress (2009) established a relation between stock performance and media coverage. They found that a portfolio of stocks without media coverage outperforms a portfolio with high media coverage by 3% annualy. Media’s impact of information disemmination is more pronounced for stocks that have low analyst coverage, high fraction of individual ownership and high idiosyncretic volatility, indicating lower information availability. In their paper on media and assets prices, Dyck & Zingales (2003) research the impact of media on market reactions to earnings announcements. They find that market reaction to earnings announcements is stronger when earnings announcements are emphasized by the media. Consistent with Fang & Peress (2009), their results are more pronounced for companies with fewer analysts following. They conclude that media have more impact on asset prices when less alternative information sources are available.

Tetlock (2007) finds that high media pessimism predicts downward pressure on share prices, followed by reversion to fundamental value and that unsually high or low pessimism predicts high trading volume. Tetlock (2007) uses daily content from the Wall Street Journal and concludes that his findings are inconsistent with theories of media content as a proxy for new information about fundamental asset values. However, A study by Chang & Suk (1998) that researches the impact of secondary information disemmination on stock markets, finds that publication of secondary information in the Wall Street Journal generates abnormal returns and an increase in trading volume. They research market reactions to secondary publications of insider trading reports and find that secondary information specifically impacts stock markets if the initial disclosure attracted only limited market attention. Their research suggests that publications in the Wall Street Journal increase information availability and trigger market reactions. Publication about possible bankruptcy announemencts in the Wall Street Journal could therefore affect stock prices and influence market sentiment, leading to smaller surprise effects associated with bankruptcy filings.

(15)

3. Methodology

This sections discusses the methodology used to analyze stock returns associated with bankruptcy announcement. Section 3.1 and 3.2 explain the event study analysis and corresponding assumptions used to measure the impact of bankruptcy announcement on stock returns. Section 3.3 elaborates on the cross-sectional regression analysis that considers several explanatory variables in order to explain differences in abnormal returns between firms. Section 3.4 extensively discusses the composition of the tested variables and forms hypotheses for their explanatory power with respect to abnormal returns.

3.1 Event Study

Event study analysis is applied in order to measure abnormal returns associated with Chapter 11 bankruptcy filing. Measuring abnormal returns allows to assess the impact of bankruptcy filings on stock prices. Abnormal returns are equal to the difference between realized and expected returns within the period of interest. Using this approach, the bankruptcy event is isolated from other market movements. The event day t = 0 is the official Chapter 11 filing date. The expected returns, or normal returns, are estimated in the period of time called the estimation window. Relative to the bankruptcy filing date (t = 0) the estimation window used to estimate normal returns, comprises trading days [-252, -200]. The returns within the estimation window are unaffected by the bankruptcy event and are used to establish a relationship between firm i’s stock and market returns. Normal returns are estimated by the market model which is based on the assumptions of a constant and linear relationship between individual stock returns and the return of a market index. The market index that is used for analysis is the value-weighted market portfolio provided on CRSP. The market model determines normal returns using ordinary least squares regressions of market returns on firm

i’s historical stock returns within in the estimation window. The ordinary least squares

regression used to estimate the model parameters is as following: !_",$ = & '_"+ )_"∗ !_+,$+&,_",$

Where, - ,",$ = 0 and /0!& ,",$ = & 123"

!_",$ represents the return of firm i at time t. The intercept and the slope of the regression model are estimated by '_" and )_" respectively. ,_",$ is the regression error which has an expected value of zero and a variance equal to the squared standard error. !_+,$ is equal to the

(16)

market return. Using these regression parameters, normal returns are estimated for the trading days within the event window. This study applies multiple event windows to analyze the abnormal returns associated with Chapter 11 filing. The abnormal returns are determined by the difference between the estimated normal returns and actual returns:

&0!_",$ = !_",$− -[!_",$]

Where 0!_",$ represents abnormal returns, !_",$ actual returns and -[!_",$] expected returns based on the estimations from the market model.

This study examines the abnormal returns on individual trading days around the bankruptcy date as well as cumulative abnormal returns (CAR) over different time windows. CAR is a measure of aggregate abnormal returns within a specific event window. This paper uses the CAR approach because it is conceptually well-suited for short-term event studies (Fama, 1998). The alternative approach, Buy-and-Hold Abnormal Returns (BHAR), is more suitable for long-term event studies. Cumulative average abnormal returns (CAAR) are calculated by dividing CAR by the number of days in the event window. The abnormal returns by trading day and the CAR are tested for statistical significance. Using this approach, it can be determined whether bankruptcy filing causes significant abnormal returns and when these abnormal returns are most pronounced. Under the null hypothesis (H0) abnormal returns are equal to zero.

The firms in the sample are categorized into three groups when analyzing abnormal returns. The first group represents all the firms in the sample. The second group consists of all the firms in the sample about which an article has been published in the Wall Street Journal mentioning a possible bankruptcy of the firm. This study only considers articles that were published in the year prior to the actual bankruptcy to ensure the article is not commenting on the financial position of the firm in an unrelated period of time in the firm’s existence. The third group consists of all firms in the sample that have not been associated with bankruptcy by the Wall Street Journal in the year prior to the Chapter 11 filing date. Throughout the entire paper, robust standard errors are used in order to ensure consistency in the presence of heteroskedastic standard errors.

(17)

3.2 Hypotheses Abnormal Returns and Cumulative Abnormal Returns

The null hypothesis (H0) for all three groups is that abnormal returns are equal to zero. The alternative hypothesis (H1) is that abnormal returns are not equal to zero. The hypotheses for abnormal returns and cumulative abnormal returns are listed below:

H0 = 0!789:= 0, H1 = 0!789:≠ 0 H0 = 0!789<= 0, H1 = 0!789<≠ 0 H0 = 0!789== 0, H1 = 0!789=≠ 0 H0 = 0!7892= 0, H1 = 0!7892≠ 0 H0 = 0!789>= 0, H1 = 0!789>≠ 0 H0 = 0!78?= 0, H1 = 0!789?≠ 0 H0 = 0!_78@>= 0, H1 = 0!_78@>≠ 0 H0 = A0!_[9:,@>] = 0, H1 = A0!_[9:,@>] ≠ 0 H0 = A0![9:,92] = 0, H1 = A0![9:,@2] ≠ 0 H0 = A0![9>,@>] = 0, H1 = A0![9>&@>] ≠ 0

In order to test the underlying assumptions that markets incorporate news published in the Wall Street Journal into stock prices, a separate event study is conducted for the group of firms that was announced in the Wall Street Journal as a possible bankruptcy case in the year prior to bankruptcy. The event day for this study is the publication date of the WSJ article and the event window is [-1, +1]. The null hypothesis is that cumulative abnormal returns are equal to zero and the alternative hypothesis is that CAR is not equal to zero:

H0 = A0!_[9>,@>] = 0, H1 = A0!_[9>,@>] ≠ 0

If CAR is negative and significant in the three-day event window around the publication of the Wall Street Journal article, this could be an explanatory factor for smaller subsequent price reactions for this group of firms when the actual bankruptcy filing takes place.

(18)

3.3 Cross-Sectional Regression Analysis

The measured abnormal returns are used as dependent variable in a cross-sectional regression analysis, thereby attempting to explain the differences in magnitude of abnormal returns between firms. The regression analysis considers several variables of interest and also includes control variables. Control variables are variables that could potentially influence the magnitude of abnormal returns but are not the subject of study. Including control variables increases internal validity by isolating the effects of the variables of interest on the dependent variable. The regression model is as following:

A0!_"8&)_?+ )_>BCD + )₂AE/ + )₌FGH!-A + )_<H-0F!-A + )_:-0!FCG! + )_IE − CJKLM + )_NO!-PQC + )_RCQS- + )_TU-/ +∈_"

3.4 Hypotheses Cross-Sectional Regression Analysis WSJ = Wall Street Journal

WSJ is a dummy variable which indicates 1 if an article was published in the WSJ about the company possibly filing for bankruptcy in the year preceding the Chapter 11 filing date. Information appearing in the WSJ has a broad reach and is generally relied on by capital market participants (Dawkins & Rose(Green, 1998). WSJ serves as a proxy for media coverage and publications about possible bankruptcy. If WSJ publishes an article about a possible bankruptcy, markets could become more aware of the financial distress and possibility of bankruptcy. The magnitude of subsequent abnormal returns associated with the bankruptcy filing are therefore expected to be lower since market participants can better anticipate the bankruptcy. Which means that abnormal returns are hypothesized to be less negative for firms with dummy variable WSJ = 1. The coefficient for WSJ is therefore expected to be positive.

Hypothesis: (+)

COV = Analyst Coverage

COV is a proxy for analyst coverage and is equal to the number of unique analysts providing at least 1 EPS forecast in the year preceding bankruptcy. Analysts are expected to improve informational efficiency and market awareness of the distressed situation of a firm. High analyst coverage would therefore reduce the surprise effect of the bankruptcy announcement. Abnormal returns expected to be less negative and the coefficient for analyst coverage is

(19)

therefore expected to be positive. The data on analyst coverage is retrieved from the I/B/E/S detail history file.

Hypothesis: (+)

NUMREC1YR = Total Number of Analyst Recommendations

NUMREC1YR represents the total number of analyst recommendations in the year prior to bankruptcy. NUMREC1YR does not make a distinction for unique analysts but aggregates the total amount of recommendations made. For example, when the same analyst provides five recommendations during the year for the same stock, all these five recommendations are added to the total number of recommendations for that particular company. The data used to determine NUMREC1YR is obtained from the summary recommendations file on the WRDS I/B/E/S database. A higher number of recommendations is expected to improve the market’s assessment of a distressed stock. Subsequent abnormal returns at the filing date are therefore expected to be less negative because of increased market anticipation of the bankruptcy. Hypothesis: (+)

MEANREC1YR = Mean of Analyst Recommendations

Analyst recommendations are obtained from the I/B/E/S recommendations summary file. The recommendations can vary between one and five. The scale for analyst recommendations is as following: 1 = strong buy, 2 = buy, 3= hold, 4 = underperform, 5 = sell. Based on this scaling of recommendations, a lower recommendation score thus reflects a more positive recommendation and a higher recommendation reflects a more negative recommendation. MEANREC1YR represents the mean of all the recommendations within the last year relative to the filing date. When analysts recommend to sell the stock of a distressed firm, this might increase the market’s prior assessment of a bankruptcy filing. Since a higher grade of recommendation is more negative about the situation of the firm, a higher recommendation grade is hypothesized to cause less negative abnormal returns.

(20)

EARNSUR = Earnings Surprise

The variable earnings surprise is a proxy for the degree to which the market is surprised by bankruptcy. This variable is determined by a regression of stock returns on time variables. The variable EARNSUP represents )₂ in the regression specification below. The estimation period to determine this variable is trading day [-252, 0] relative to bankruptcy filing. The regression expression used to determine earnings surprise is as following:

!MWXLY"& = & )?+ )>ZQH- + )2ZQH-2+&∈"

Return represents the daily stock returns for firm i, TIME is simply a time variable that reflects the days relative to bankruptcy filing and TIME2 is the square of TIME. The results of this regression indicate the speed and timing of accumulated losses. Considering that the firms in the sample are all bankruptcy cases it is expected that the coefficient for TIME is negative. This means that the firm is expected to incur negative returns and thus a declining stock price over the year prior to bankruptcy. The coefficient for TIME2 (β2) is an indicator for the degree to which markets are surprised by the bankruptcy. If the coefficient for TIME2_is negative, this means that the curve for returns over time is concave and that losses accumulate faster as time progresses and the firm approaches bankruptcy. The degree of concavity serves as a proxy for the earnings surprise. When the magnitude of the negative coefficient for TIME2 is larger, the curve for stock returns is more concave and thus markets are more surprised by the bankruptcy. These firms accumulate losses at a later stage, closer to the actual bankruptcy filing date.

If the coefficient for TIME2 is positive and statistically significant, the curve becomes convex. This means that the losses associated with the bankruptcy are incurred in an earlier stage relative to the filing date. In this case, financial markets anticipate the bankruptcy longer before it actually takes place. When the coefficient is not statistically significant, this means that the incurred losses are smoothened over time. The expectation for the relation between earnings surprise and abnormal returns is that a larger earnings surprise results in more negative returns around the bankruptcy.

(21)

O-SCORE = Probability of bankruptcy According to Ohlson O-score

This control variable determines the expected probability of bankruptcy, calculated using the Ohlson O-score. The Ohlson O-score predicts the probability that a firm goes bankrupt within the next two years. The O-score is computed using data from the year-end of the fiscal year prior to bankruptcy. The Ohlson O-score and Altman Z-score are the most commonly used accounting based predictors of bankruptcy. Previous academic studies found that overall, the Ohlson O-score is more accurate when predicting bankruptcies (Dichev, 1998). This study therefore uses the Ohlson O-score to estimate the probability of bankruptcy. Prior academic research has frequently used the O-score as a proxy for financial risk (Chen, Chollete, & Ray, 2010). A higher O-score, representing high financial risk, is expected to result in more negative returns in the event of bankruptcy. The formula used to calculate the Ohlson O-score and its corresponding legend are provided below.

Hypothesis: (-) E = & −1.32 − 0.407 ln Z0W + 6.03 ZUW Z0W − &1.43 BAW Z0W + &0.0757 AUW

A0W − 1.72e − 2.37& FQW Z0W − &1.83& ggEW ZUW + &0.285h − 0.521 FQW − FQW − 1 FQW + FQ&Z − 1 &

In order to calculate the actual probability of bankruptcy the following formula is applied: OLKijiklkWm&Kn&ijYoLXpWJm = -qp(E − sJKLM)/(1 + -qp E − s KLM )&

Table 1. Legend for O-Score Formula Abbreviation Definition TA Total Assets TL Total Liabilities CA Current Assets CL Current Liabilities WC Working Capital

= Current assets – Current Liabilities

NI Net Income

FFO Funds from Operations

= Net Income + Depreciation + Amortization X 1 if TL > TA, 0 otherwise

(22)

PREDIS = Pre-disclosed Information

Pre-disclosed information is a control variable that represents the percentage decrease in stock price from one year prior to six trading days prior to bankruptcy. If a firm’s stock has experienced considerable losses in the period preceding bankruptcy, much of the information concerning financial distress and possible bankruptcy is already incorporated into the stock price. If more information about the bankruptcy is already incorporated into the stock price, the surprise effect is smaller when the firm actually files for bankruptcy. A large prior decrease in share prices is therefore expected to reduce the magnitude of abnormal returns associated with the actual Chapter 11 filing, causing less negative abnormal returns.

Hypothesis (+)

SIZE = Market value

This control variable is determined by the market value 1 year prior to bankruptcy filing. The expectation is that more information about a firm’s performance is available for larger firms. Larger firms receive more analyst and media coverage (Bhushan, 1989). This could lead to increased anticipation of bankruptcy by financial markets, reducing the surprise effect of actual bankruptcy filing. The coefficient for SIZE is therefore expected to be positive

Hypothesis: (+) LEV = Leverage

This control variable is determined by dividing total liabilities by total assets using balance sheet data from the fiscal year-end of the year prior to bankruptcy filing. In case of bankruptcy, debt holders enjoy seniority over shareholders. Firms with relatively high levels of debt have to deal with this in a reorganization plan. The high levels of debt reduce potential payout for shareholders and therefore increase potential shareholder losses in case of bankruptcy (Clark & Weinstein, 1983). Leverage is thus expected to increase the magnitude of negative abnormal returns associated with bankruptcy filing. LEV is therefore expected to have a negative coefficient.

(23)

Table 2. provides an overview of the variables included in the regression analysis and their hypothesized effect on (Cumulative) Abnormal Returns.

Table 2. Summary of Hypotheses

Variable Hypothesis WSJ (+) COV (+) NUMREC1YR (+) MEANREC1YR (+) EARNSUR (-) O-SCORE (-) PREDIS (+) SIZE (+) LEV (-)

3.5 Determinants of Accumulation of Losses.

A separate study is performed to investigate the curve of abnormal returns in the year prior to bankruptcy. A long-term event study is conducted using Buy-and-Hold Abnormal Returns (BHAR) of the stocks in the window [-252, 0]. The BHAR approach is used to assess abnormal returns because it is more effective than CAR in long-term studies. The CAR approach can lead to biased estimators of abnormal returns in medium- and long-term event studies. BHAR uses compounded abnormal returns and is therefore considered to represent investor experience more accurately than CAR (Barber & Lyon, 1997). The resulting BHAR is used as dependent variable in the following regression, where time represents days:

vw0!"&= )?+ )>ZQH- + )2ZQH-2+&∈"

The coefficient for ZQH-2_{is then stored for all firms and used as dependent variable in a}

separate regression. When )₂ is negative, the curve of BHAR is concave. A concave curve means that incurred losses accumulate faster as time progresses, closer to the bankruptcy event. When )₂ is positive, the curve of BHAR is convex. A convex curve indicates that losses accumulate faster at an earlier stage, more distant from the bankruptcy filing date. If )₂ is close to zero, losses are incurred gradually over time.

(24)

)₂ is a proxy for curvature of BHAR plotted over time. The variable is therefore named CURV. By performing a regression with CURV as the dependent variable, it can be tested whether media and analysts help to smoothen information incorporation into stock prices. When numerous analysts cover a firm and they provide recommendations on a regular basis, the surprise effect close to bankruptcy is expected to be lower as analysts improve informational efficiency and constantly inform market participants about the position of a firm. Similarly, high media coverage helps to incorporated available information into stock prices at an earlier stage, thereby reducing the surprise effect close to the filing date. It is therefore hypothesized that high analyst and media activity smoothen losses over time and thus have a positive impact on )₂. The regression expression controls for leverage and size. The complete regression expression is as following:

AG!/_"8&&)_? + )_>BCD + )₂AE/ + )₌FGH!-A1h! + )_<H-0F!-A1h! + )_:E − CAE!-+ )I&CQS- + )NU-/ +∈"

A summary of the hypothesized relation of each variable in the regression expression is displayed in table 3.

Table 3. Summary of Hypotheses

Variable Hypothesis WSJ (+) COV (+) NUMREC1YR (+) MEANREC1YR (+) SIZE (+) LEV (-)

(25)

4. Data and Descriptive Statistics

The sample of bankrupt firms consist of publicly listed firms that have filed for Chapter 11 bankruptcy between 2000-2013 in the United States of America. The firms and bankruptcy filing dates have been drawn from the UCLA-LoPucki Bankruptcy Research Database (BRD). The corresponding Cusip identification numbers have been drawn from the Wharton Research Database Services (WRDS). Firms are often delisted before they file for bankruptcy due to failure to meet listing criteria. Firms that were delisted from the stock exchange before the Chapter 11 filing date were not taken into consideration for this study due to a lack of financial data to perform the event study. The firms that were selected for the event study satisfied the following criteria:

1. Stock returns are available on the WRDS Center for Research on Security Prices (CRSP)

2. The firm remains listed on its stock exchange until Chapter 11 filing date.

3. Daily returns for the stock are available on CRSP for at least 30 days out of a 60-day period (day -252 to day -200) relative to the bankruptcy filing date for estimation of normal returns.

4. Daily returns for the stock are available on CRSP for all days within the event window (day -5 to day +1) relative to the bankruptcy filing date. Table 4. displays the sample size that satisfies these criteria.

Table 4. Sample Size

Total bankruptcies cases in LoPucki database 2000-2013 638

Stocks listed on CRSP 374

Returns available in estimation window 140

Returns available in event window 105

Final Sample 105

Financial statement information was retrieved from the WRDS COMPUSTAT database. Analyst data used for the analysis was drawn from the I/B/E/S database. Data on Wall Street Journal publications was retrieved from Wall Street Journal’s online database. An overview of all individual data items used in the analysis and their corresponding data source is provided in Table 5.

(26)

Table 5. Data Sources

Data Source Data Item

CRSP Daily Stock returns

Daily market returns Daily stock prices Market value equity Shares outstanding

UCLA-LoPucki Bankrupt firms

Bankruptcy filing date Chapter 7/Chapter 11

COMPUSTAT Current assets

Total assets Current liabilities Total liabilities Sales/Turnover Retained earnings EBIT Pretax income Wall Street Journal WSJ publications

I/B/E/S/ Number of analysts

Number of recommendations Analyst recommendations

Of the 105 firms in the sample, 29 firms have been mentioned by the Wall Street Journal as a possible bankruptcy case in the year prior to bankruptcy. This study only considers articles that were published in the year prior to the actual bankruptcy to ensure the article is not commenting on the financial position in an unrelated period of time in the firm’s existence. The information retrieved from the WSJ about possible bankruptcy filings is reported in Table 6. This table also lists the sample distribution of bankruptcy cases by year. Table 6. also presents an overview of the sample by year in which the bankruptcy cases occurred. The high frequency of bankruptcies in the early 2000’s and in 2009 coincide with economic crises is the US during these years. Table 7. provides an overview of the descriptive statistics of other variables used for analysis.

(27)

Table 6. Bankruptcy Frequency and WSJ Article by Year

Year Frequency Percentage WSJ Article No WSJ Article

2000 12 11.4% 2 10 2001 17 16.2% 6 11 2002 15 14.3% 2 13 2003 6 5.7% 3 4 2004 3 2.9% 1 2 2005 7 6.7% 2 5 2006 2 1.9% 1 1 2007 3 2.9% 1 2 2008 6 5.7% 1 5 2009 14 13.3% 4 10 2010 3 2.9% 1 2 2011 6 5.7% 1 5 2012 5 4.8% 2 3 2013 6 5.7% 2 4 Total 105 100% 29 76

Table 7. Summary Statistics

Variable Mean St. Dev. Min Max

COV 6.66 8.11 0 50 NUMREC1 54.50 69.25 0 373 NUMREC50 5.16 7.79 0 53 MEANREC1 2.54 0.67 1 3.83 MEANREC50 2.93 0.68 1 4.33 LEV 0.87 0.59 0.10 6.15 O-Score 0.77 0.25 0.00 1 PREDIS -0.77 0.28 -0.99 0.71 SIZE 1219803 4533140 7955.82 3.16e+07

(28)

5. Results

5.1 Abnormal Returns and Cumulative Abnormal Returns

This section discusses the results of the event study performed on stock returns around bankruptcy announcement. The abnormal returns for each trading day individually and the cumulative abnormal returns aggregated over different time windows are reported. Table 8. displays the results of individual trading days relative to the bankruptcy filing date (t = 0). The abnormal returns are calculated for the entire sample of firms (1), firms that have been announced by WSJ as a possible bankruptcy case in the year prior to filing date (2) and firms that have not been announced by WSJ as a possible bankruptcy case in the year prior to filing date (3). The abnormal returns were tested for statistical significance and their corresponding t-statistics are shown between brackets. Column (4) displays the mean difference between the abnormal returns for group WSJ (2) and No WSJ (3). The mean difference is tested for significance using a t-test. The t-statistics corresponding to this t-test are displayed between brackets in column (4). The level of significance is indicated by the asterisks.

Table 8. Abnormal Returns by Trading Day

(1) (2) (3) (4)

Trading day All Firms WSJ No WSJ Mean

Difference [-5] -0.0123 -0.0175 -0.0103 -0.0072 (-0.66) (-0.30) (-0.77) (0.17) [-4] -0.0324** -0.0175 -0.0381** -0.0324 (-2.32) (-0.74) (-2.23) (-0.66) [-3] -0.0249** -0.0225 -0.0258 -0.0032 (-2.05) (-1.31) (-1.66) (-0.12) [-2] 0.0025 -0.0289 0.0142 0.0431 (0.11) (-0.72) (0.55) (0.88) [-1] -0.0261 -0.0131 -0.0310* -0.0179 (-1.04) (-0.17) (-1.68) (-0.32) [0] -0.1610*** -0.1408*** -0.1686*** -0.0278 (-4.17) (-2.93) (-3.37) (-0.32) [+1] -0.1249*** 0.0126 -0.1835*** -0.1961*** (-3.22) (0.22) (-3.86) (-2.39) # Observations 105 29 76

Robust t-statistics in parentheses *** p<0.01, ** p<0.05, * p<0.1

(29)

The abnormal returns per trading for all firms in the sample are significant on trading day [-3], [0] and [-1]. The abnormal returns are most pronounced and concentrated on day [0] and [+1} and are statistically significant with p<0.01. The abnormal returns are -16.10% on day [0] and -12.49% on day [+1]. Of all five days preceding the filing date, only day [-3] registers minor significant abnormal returns of -2.49%. These results reflect that bankruptcy filing is not well anticipated by financial markets and that most of the losses are incurred on the actual filing date and the day after. On average, markets react in response to actual filing instead of aggregating losses in anticipation of bankruptcy. These results indicate that overall, financial markets are incapable of accurately predicting bankruptcy and its timing.

When comparing group WSJ and No WSJ, the only trading day that registers a significant difference between both groups is day [+1]. The mean difference in abnormal returns however, is remarkable. Group WSJ does not incur negative abnormal returns on day [+1] whereas group No WSJ incurs abnormal returns of -18.35%. This difference is both statistically and economically significant and indicates that bankruptcy for firms with prior association with bankruptcy in the Wall Street Journal is better anticipated by financial markets. More informational content regarding Chapter 11 filing has already been incorporated into stock prices for group WSJ. The difference of 19.61% between mean abnormal returns for both groups is significant at p<0.01. Group WSJ does not incur any negative returns at all on trading day [+1]. These results imply that an article in the WSJ can improve informational efficiency with regards to the financial position of a distressed firm and its probability of bankruptcy.

Table 9. shows the cumulative abnormal returns (CAR) aggregated over different time windows. All of the displayed time windows register statistically significant CAR for group (1). Consistent with the results from table 8. it can be concluded that the abnormal returns are concentrated on the filing date and the day after. This window, [0, +1], registers CAR of -25.95%. Total CAR over the seven-day window [-5, +1] is -33.12% on average. WSJ (2) only incurs significant CAR in the window of [-1, +1] and [0, +1] whereas group No WSJ (3) experiences significant CAR in all of the applied event windows. The differences between the mean CAR for both groups is most pronounced for the window [0, +1]. The CAR in different time windows confirm that most losses associated with bankruptcy announcement are incurred in response to actual Chapter 11 filing and that financial market only to a small degree anticipate the bankruptcy with a CAR of -6.7%, aggregated over four days in the window [-5, -2].

(30)

Table 9. Cumulative Abnormal Returns by Event Window

(1) (2) (3) (4)

Event Window All Firms WSJ No WSJ Mean

Difference [-5, +1] -0.3312*** -0.2191 -0.3739*** -0.1548 (-5.95) (-1.62) (-6.58) (-1.25) [-5, -2] -0.0669** -0.0855 -0.0599* 0.02564 (-2.19) (-1.15) (-1.90) (0.37) [-1,+1] -0.2660*** -0.1357 -0.3141*** -0.1783* (-5.98) (-1.43) (-6.41) (-1.80) [0, +1] -0.2595*** (-6.49) -0.1297** (-2.14) -0.3078*** (-6.28) -0.1781** (2.01) # Observations 105 29 76

As described in the methodology, a separate event study is conducted to test the underlying assumption that news about possible bankruptcy, published in WSJ, is incorporated into stock prices. Table 10. displays the results of this event study. Stock prices register significant abnormal returns on the event day, the day that the article about possible bankruptcy is published by WSJ. The results from this event study confirm that the negative news concerning possible bankruptcy in WSJ is incorporated into stock prices. On the event day, day 0, these stocks report abnormal returns of -17.53% on average. The results are significant at p<0.05. The stocks only report significant abnormal returns on day 0. The CAR for the event window [-1, +1] is -21.54%.

These result confirm that articles in the WSJ can have an effect on stock prices. If the available information is incorporated into stock prices at an earlier point in time relative to the Chapter 11 filing date, the price reaction at the actual bankruptcy filing could therefore be reduced. This implies that news in WSJ enhances information incorporation into stock prices and thereby promotes informational efficiency. These result are consistent with the earlier findings that the magnitude of abnormal returns and CAR for group WSJ around bankruptcy is significantly smaller than for group No WSJ.

(31)

Table 10. Event Study WSJ Article

Robust t-statistics in parentheses

*** p<0.01, ** p<0.05, * p<0.1

5.2 Cross-Sectional Regression Analysis

This section provides an overview of the results from the cross-sectional regression analysis that tests potential explanatory variables that determine the magnitude of abnormal returns associated with bankruptcy announcement. The first regression analysis is conducted on CAR over the entire event window [-5, +1] and the second regression analysis is conducted on the abnormal returns on day [+1]. The separate regression analysis, with abnormal returns of day [+1] as dependent variable, is conducted to test the incremental effect of WSJ while controlling for other factors. Day [+1] is of particular interest because this is the only event day with a significant difference in mean abnormal returns between group WSJ (1) and group No WSJ (2).

Table 11. displays the results from the regression analysis with CAR [-5, +1] as dependent variable. This regression model tests the explanatory power of the independent variables with respect to the magnitude of abnormal returns over the entire event window. The regression analysis considers two models. The first model includes all independent variables and the second model excludes the variable O-SCORE. The information necessary to compute the O-score is not available on COMPUSTAT for the entire firm sample. Therefore, the

Trading Day Abnormal

Return [-1] -0.0255 (-0.74) [0] -0.1753** (-2.15) [1] -0.0158 (-0.24) Window [-1, +1] CAR -0.2154* (-2.06) Observations 25

(32)

sample size decreases significantly when including variable O-SCORE. As can be seen in model (1), variable O-SCORE has no significant influence on CAR and is therefore left out in the second model. This increases the sample size and thereby the precision in the second model.

Table 11. Cross-Sectional Regression Analysis Dependent Variable CAR [-5, +1]

VARIABLES Model 1 Model 2

COV 0.0178 0.0184 (0.94) (1.32) SIZE -0.0000* -0.0000*** (-1.75) (-3.77) NUMREC1YR -0.0002 -0.0002 (-0.10) (-0.13) MEAN1YR -0.2967*** -0.2014*** (-3.33) (-2.68) PREDIS -0.2401 -0.3421 (-0.68) (-1.47) EARNSUR 238,014.11*** 173,803.30*** (3.96) (3.82) WSJ 0.0815 0.0549 (0.43) (0.39) LEV 0.1251*** 0.0420 (2.79) (0.55) O-SCORE -0.2688 (-0.88) Constant 0.4321 -0.1051 (1.14) (-0.38) Observations 51 84 R-squared 0.517 0.412

The R-squared of model one and two is 0.517 and 0.412 respectively, indicating that the models are successful at explaining the variation in the dependent variable CAR. Variable COV was hypothesized to have a positive effect on CAR. The effect of increasing analyst coverage on CAR is indeed positive, however it is not statistically significant. SIZE has a negative relation with CAR, which means that firms with a higher market value one year prior to bankruptcy experience higher losses around bankruptcy. The expectation was that there is more market awareness about firms with high market capitalization. This hypothesis is disproven. A reason for this could be that market participants do not expect the largest

(33)

companies to go bankrupt because they are “too big too fail”. NUMREC1YR and PREDIS have no particular explanatory power for CAR, meaning that the number of analysts following the firm and the percentage share price decrease over the last year do not impact CAR.

It is interesting to note that WSJ has no significant explanatory power for CAR over the entire event window. This is consistent with earlier findings that showed that the mean difference of abnormal returns for group WSJ and No WSJ is not statistically significant. Leverage is of significant importance in the first model. In the second model however, leverage loses its statistical significance due to the increased standard errors. The standard errors of LEV are 0.045 and 0.076 in the first and second model respectively. Leverage has a positive influence on CAR which is in contradiction with the earlier stated hypothesis. An explanation could be that financial markets perceive the probability of bankruptcy for firms with high leverage to be higher and incorporate this higher probability into security prices ahead of bankruptcy.

MEANREC1YR, Mean analyst recommendation, has a significant negative relation with CAR. It was however hypothesized that the relation was positive. This hypothesis assumes that analysts improve informational efficiency. The result however indicates that stocks for which analysts provide high sell recommendations, lose more value around bankruptcy. Meaning that analysts provide insightful forecasts, but markets do no process this information and incorporate it into stock prices. MEANREC1YR varies within the scale of 1-5, with one being a strong buy recommendation and 5 a strong sell recommendation. Every point increase, meaning a stronger sell recommendation, decreases abnormal returns with 29.67%. This seems to be an extremely impactful variable. MEANREC1YR however, varies between 2.54 and 3.83 with a standard deviation of 0.67 which slightly reduces its impact on CAR. A reason that market participants do not incorporate this information into stock prices could be the limited reach of recommendations listed on WRDS. Earnings Surprise has a significant and positive effect on CAR, meaning that stock that start accumulating losses at a relatively late point in time, eventually experience larger losses around the actual bankruptcy filing.

Table 12. displays the results of the regression analysis on the abnormal returns on day [+1]. On this day within the event window, the incremental effect of WSJ becomes apparent. The dummy variable WSJ increases abnormal returns with 16.30% on average and is significant at p<0.1. WSJ has a p-value of 0.068 which makes it nearly significant at p<0.05 as well. This is consistent with earlier findings that the difference in mean abnormal returns

(34)

for group WSJ and No WSJ is significant. LEV and MEANREC1YR display the same relation with abnormal returns as established in the regression on CAR [-5, +1]. PREDIS, the percentage share price decrease in the year prior to bankruptcy, negatively impact abnormal returns on day 1. The R-squared of the model is very low at 0.169. The model does however find significant variables. This means that there is an apparent relation and trend between the variables. However, the extend to which to model explains the variation in abnormal returns on day [+1], is limited.

Table 12. Cross-Sectional Regression Analysis Dependent Variable AR Day [+1]

VARIABLES DAY [+1] COV -0.0014 (-0.07) SIZE -0.0000 (-0.24) NUMREC1YR 0.0008 (0.42) MEANREC1YR -0.1181* (-1.81) PREDIS -0.2770* (-1.89) EARNSUR -13,287.3818 (-0.39) WSJ 0.1630* (1.86) LEV 0.0737** (2.59) Constant -0.1864 (-0.96) Observations 61 R-squared 0.169

(35)

5.3 Determinants of Accumulation of Losses.

Table 13. reports the results from the regression of BHAR on time and time squared. The coefficients for time and time squared are both negative. This indicates that the curve for BHAR is downward sloping and concave. On average, the firms in the sample accumulate losses faster as time progresses.

Table 13. Regression of BHAR on Time Variables

VARIABLES BHAR

Time -0.00240***

(-25.01)

Time Squared -2.03e-06***

(-5.25)

Constant -0.0261***

(-5.93)

Observations 26,488

R-squared 0.302

Table 14. presents the results of the regression analysis with CURV, the curvature of BHAR returns, as dependent variable. None of the variables have explanatory power with regards to the curvature of BHAR in the year prior to bankruptcy. This means that analyst and media activity does not influence the allocation and speed of accumulation of losses.

Informational efficiency and market reaction to bankruptcy announcement