SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. I. Hypotheses and Implications
  4. II. Data and Empirical Results
  5. 1. Univariate Analysis
  6. 2. OLS Regressions
  7. III. Robustness
  8. IV. Conclusion
  9. References

This paper examines how the announcement of an accusation of fraudulent financial misrepresentation affects industry rivals of the accused firm. Consistent with the importance of the industry competition effect, we find that rivals in less competitive industries benefit from the event. However, in competitive industries, the information spillover effect dominates the competition effect, resulting in negative returns to rival shareholders following the event. The spillover effect increases in importance with the severity of the accusation and is more important for opaque rivals and for rivals that had positive stock price reactions to past positive earnings surprises of the accused firm.

Recent interest surrounding the revelations of corporate financial misrepresentation has centered on the firms accused of such activities. Palmrose, Richardson, and Scholz (2004), Karpoff, Lee, and Martin (2008a, b), and Gande and Lewis (2009) document significant declines in shareholder value around the first revelation of accusations of financial misrepresentation and earnings restatements. While the shareholders of a firm accused of financial misrepresentation suffer significant financial loss, this disclosure may hold economic implications for other “related” firms as well. The main goal of this paper is to investigate how the value of rival firms is affected by such revelations. For example, the following is a quote from a March 2002 Associated Press report concerning the exposure of wrongdoing at Global Crossing Corporation:

“Global Crossing's sudden implosion has done more than sting shareholders, creditors and employees. It's burdened the suffering telecom industry with new worries about solvency and accounting irregularities while prompting some investors to flee the sector. ‘Global Crossing was a wake-up call for a lot of people that didn't realize how bad the telecom sector had become,’ … . ”

An accusation of financial misrepresentation in one firm may have negative implications for shareholders of rival firms. Therefore, one hypothesis supposes that information contained in an accusation of financial misrepresentation at one firm affects the information used to value rival firms (the information spillover hypothesis). Another hypothesis posits that rival firms benefit from the accusation of financial misrepresentation of a firm, possibly by attracting its customers (the industry competition hypothesis). As a result, it is not obvious a priori that an accusation of financial misrepresentation is detrimental to all rival firms. This study aims to measure the importance of these two hypotheses in the context of financial misrepresentation deemed to have the deliberate intent of fraud.

We begin our analysis with a sample of 1,001 enforcement actions for financial misrepresentations from April 1976 to January 2010, an extended version of the sample used in Karpoff et al. (2008a, b).1 We restrict this sample to the subgroup of actions that are flagged by Karpoff et al. (2008a, b) for fraud charges. Our final sample (after several additional restrictions due to data availability) includes 444 enforcement actions. For these 444 cases, we find that firm value declines by 19.7%, on average, in the three days around the enforcement action begin date. In the same time period, we find that rival firm values decline by 0.54%, on average.2

Although the average drop in rival value suggests that the information spillover effect dominates the industry competition effect, this average number hides significant variation at the industry and intraindustry level. Figure 1 illustrates the distribution of rival cumulative abnormal announcement returns (CAR) for the three-day window around the event. Rival firms are grouped by industry concentration deciles based on the Herfindahl Index. This figure reports a significant variation in rival CAR within each group. For all rivals, the variability in CAR is 0.091. The within-industry variation is 0.087, while the across-industry variation is only 0.0029. We explore both sources of variation, industry- and firm-level, to determine which rival firms react more positively (negatively) to the event.

image

Figure 1. CAR Rival Dispersion by Herfindahl Deciles

The cumulative abnormal returns (CAR) of the rival firms are illustrated by Herfindahl deciles. CAR rival is computed as the market-adjusted cumulative return over the three days around the announcement of the accusation of financial misrepresentation of a competitor. The market return is the CRSP value-weighted index. Rivals are selected using the historical four-digit SIC in Compustat. The Herfindahl index is computed based on sales of companies listed in Compustat. Herfindahl deciles are formed based on one Herfindahl index per four-digit SIC industry. The lowest Herfindahl decile corresponds to the most competitive industries.

Download figure to PowerPoint

To guide our empirical analysis, we develop a set of implications based on the idea that the relative magnitude of the competition effect and the information spillover effect depend upon specific firm and industry characteristics. We then test these implications by analyzing the cross-sectional variation in rival firms’ abnormal returns around the event date.

We find that rivals in less competitive industries experience higher CAR around the event. CAR is also higher for rivals of accused firms that have both large sales and are in the least competitive industries. These results are consistent with the industry competition hypothesis, which predicts that rivals of a troubled firm may gain value by attracting clients of the accused firm.

We construct a rival-level proxy for the level of competition with the accused firm (this proxy measures the stock price reactions of the rival to the earnings announcements by the accused firm, prior to the event). We find that this variable is negatively correlated with rival CAR. A rival that experienced stock price drops when the accused firm reported higher than expected earnings in the past, now experiences an increase in value on the event date. We interpret these results as consistent with the industry competition hypothesis. For example, when WorldCom manipulated its cost numbers, it reported earnings growth that was difficult for its main rival, AT&T, to match.3 The market believed, at the time, that AT&T was losing the fight against WorldCom. Once WorldCom's false earnings statements were exposed, the market realized that AT&T was actually doing well in holding off its competitors.

In addition, we find that rival CARs are lower when the CAR of the accused firm is more negative, consistent with the information spillover hypothesis. This effect, however, is reversed for rivals in the least competitive industries, which is further indication that the industry competition effect dominates the information spillover effect in concentrated industries.

Finally, we find that rival CARs are more negative for rivals that exhibit higher levels of opacity (as measured, for example, through bid-ask spreads, intangibles-to-asset ratios, and analyst forecast dispersion). When the market has less information about a rival firm, it assumes that the lowered earnings reported by the accused firm also reflect lowered earnings for rival firms. Hence, these rivals see a larger drop in value following the event date.

To highlight the economic impact of industry competition versus information spillover effects, we form two subsamples of firms that are predicted to be more affected by either the industry competition effect or by the information spillover effect. For the subsample of firms predicted to be most affected by the industry competition effect (rivals in the least competitive industries, rivals of large accused firms with very negative event date CARs, least opaque rivals, and most opaque accused firms), we find that the average three-day CAR is 3.2%. For the subsample of rival firms predicted to be most affected by the information spillover effect (rivals in the most competitive industries, large accused firms with very negative CARs, most opaque rivals, and least opaque accused firms), we find that the average three-day CAR is –1.5%.

In summary, our findings indicate the importance of both the competition and information spillover effects in determining rival firm response to financial misrepresentation. The results highlight the firm- and industry-level characteristics that distinguish firms who are more sensitive to the information spillover effect from firms that are more sensitive to the industry competition effect. Thus, this paper provides a more complete picture of the overall impact that cases of fraudulent financial misrepresentation have had on investors, extending recent work by Karpoff et al. (2008a) and Gande and Lewis (2009) that focus on the value loss to shareholders of the accused firm.4

This paper relates to two strands of literature. One strand analyzes the importance of information spillover and industry competition effects, while the second focuses on financial misrepresentation. The first strand of literature primarily focuses on analyzing the abnormal returns of portfolios of rival companies following various events.5 Most closely related to our paper are Lang and Stulz (1992) who investigate the impact of bankruptcy on rivals, Laux, Starks, and Yoon (1998) who analyze rival reactions to dividend revisions, and Gleason, Jenkins, and Johnson (2008) who investigate the effect of earnings restatements on industry rivals. Lang and Stulz (1992) find that rival returns are negative, on average, following a bankruptcy event in their industry; however, for a portfolio of rivals in highly concentrated industries with low leverage, the returns are positive.6Laux et al. (1998) find that firm-level characteristics are important in explaining cross-sectional variation in rival reaction to dividend revisions beyond industry-level (or portfolio-level) characteristics. Gleason et al. (2008) find that nonrestating firms that have the same external auditor also incur a stock price drop if one of their competitors in the same industry restates earnings. Finally, Chen, Ho, and Shih (2007) find that rivals react negatively when other firms announce corporate capital investments.

The second stream of related literature investigates the effects of financial misrepresentation. Palmrose et al. (2004) document an average abnormal announcement return of –9% in response to restatements of annual or quarterly financial statements. Karpoff et al. (2008a) find an average loss in market value of 38% when news of financial misconduct breaks. They indicate that a significant part of the loss relates to reputation costs, costs that are not explained by future direct penalties levied on the firm. Gande and Lewis (2009) find a significant drop in share price of 4.5% upon the announcement of shareholder-initiated class action lawsuits. All three studies suggest that firms caught in or accused of financial misrepresentation, fraud, or misstatements typically experience a substantial drop in firm value.

We add to these two strands of the literature in the following ways. First, our analysis documents the simultaneous importance of both the competition and the information spillover effects. While Lang and Stulz (1992) report some aspects of these effects in the case of bankruptcy announcements, we document their importance in the context of intentional financial misrepresentation. In fact, Gleason et al. (2008) suggest that investigating competitive effects from accounting-related issues is an avenue for future research.7

Additionally, our study adds to the literature on the competitive and the information spillover effects by analyzing the importance of firm-level characteristics in addition to industry-level ones. This is another way in which we extend the analysis of such studies as Lang and Stulz (1992) and Laux et al. (1998).

Thus, we explore the impact on rivals when accused of financial misrepresentation and conduct the analysis on a firm-level basis. This analysis explores the more subtle aspects of the information spillover and competition effects in the context of intentional financial misrepresentation.

The rest of this paper is organized as follows. In Section I, we derive the main empirical implications, which are then tested in Section II. Section III provides several robustness tests, while Section IV sets forth our conclusions.

I. Hypotheses and Implications

  1. Top of page
  2. Abstract
  3. I. Hypotheses and Implications
  4. II. Data and Empirical Results
  5. 1. Univariate Analysis
  6. 2. OLS Regressions
  7. III. Robustness
  8. IV. Conclusion
  9. References

We hypothesize that the public announcement of an investigation of financial reporting violations that are deemed to be fraudulent will have two effects on the stock prices of rival firms. The industry competition hypothesis suggests that a rival firm will benefit from the investigation in the form of reduced competition from the investigated firm. This implies a positive abnormal stock return following the announcement. The information spillover hypothesis suggests that a rival firm suffers from the announcement since it reveals information that the rival firm's value may also be inflated. This implies a negative abnormal stock return following the announcement. The potential existence of these two effects leads to the following implications:

Implication 1: The industry competition hypothesis predicts that greater industry competition is associated with lower (more negative) abnormal returns to the industry rivals of the investigated firm. The information spillover hypothesis predicts that the level of industry competition will have no bearing on the abnormal returns of rival firms.

In more competitive industries, firms are less able to extract surplus from consumers. In such industries, a loss of one competitor will have a smaller impact on the profits of rival firms. One empirical proxy for measuring industry competition is the Herfindahl Index calculated based on sales of firms in the same four-digit SIC code. A higher index number represents a less competitive industry. Another empirical proxy is the Parrino (1997) Homogeneity Index that measures the similarity of firms within industries.

Implication 2: The industry competition hypothesis predicts that an event that is more detrimental to the accused firm will be associated with higher (more positive) rival abnormal returns. In contrast, the information spillover hypothesis predicts that an event that is more detrimental to the accused firm will be associated with lower (more negative) rival abnormal returns.

When the (negative) magnitude of the event is large, the resulting impact on the rival will be large, as well. If the competition effect dominates, then the benefit to a rival firm will be magnified if the event has severe implications for the accused firm. If the information spillover effect dominates, then the cost to the rival firm and the drop in its value will be magnified with the severity of the event. Implication 2 is a test of which of the two effects is stronger. One empirical proxy for the importance of the event is the CAR of the accused firm around the event date. A second measure of the importance of the event to rivals is the revenues of the accused firm. An additional measure of the importance of the event to a rival is the earnings response coefficient (ERC). This variable is a rival-specific measure. The ERC measures the rival's average stock market reaction to the accused firm's earnings announcements in the years preceding the event.

Implication 3: The industry competition hypothesis predicts that greater opacity in a rival firm has no bearing on its abnormal returns. The information spillover hypothesis predicts that greater rival opacity is associated with lower (more negative) rival abnormal returns.

The market uses information generated by the event to update its estimate of each rival's market value. For more opaque rivals and for rivals whose valuation is more uncertain (e.g., rivals with more growth opportunities), the market will put more weight on new (negative) information. Some measures of firm-level opaqueness include rival bid-ask spreads, the proportion of assets that are intangible, and rival-analyst forecast dispersion.

Implication 4: The industry competition hypothesis predicts that greater opacity in an accused firm has no bearing on rival abnormal returns. The information spillover hypothesis predicts that greater opacity in an accused firm is associated with higher (less negative) rival abnormal returns.

When the accused firm is opaque, the announcement will tend to have a smaller impact on rivals. The assumption here is that information about an accused firm that is more opaque is less likely to be used in updating rival firm value. Table I summarizes the four implications.

Table I.  Implications
  Industry Competition Hypothesis Information Spillover Hypothesis
  1. a

    The table summarizes the implications of the industry competition and information spillover hypotheses. The first column also lists proxies used to test the implications. Columns 2 and 3 report the expected sign of the relation between the proxy and the cumulative abnormal return (CAR) of the rival firm.

Imp1: Product Market Competition Lower competition, higher (more positive) CARNo effect
 Herfindahl index+0
 Parrino Homogeneity index+0
Imp2: Importance of Event Higher (more positive) reaction of rivals – especially in concentrated industriesMore negative reaction of rivals (lower announcement returns)
 Announcement return of accused firm
 Sales of accused firm++
 Earnings Response Coefficient
Imp3: Opacity of Rival Firm No effectMore negative reaction of rivals (lower announcement returns)
 Bid-ask spread0
 MB0
 Intangibles/ assets0
 Analyst forecast dispersion0
Imp4: Opacity of Accused Firm No effectLess negative reaction of rivals (higher announcement returns)
 Bid-ask spread0+
 MB0+
 Intangibles/assets0+
 Analyst forecast dispersion0+

II. Data and Empirical Results

  1. Top of page
  2. Abstract
  3. I. Hypotheses and Implications
  4. II. Data and Empirical Results
  5. 1. Univariate Analysis
  6. 2. OLS Regressions
  7. III. Robustness
  8. IV. Conclusion
  9. References

In this section, we describe our data, formulate tests, and discuss the empirical results.

A. Data

Sample selection begins with a comprehensive list of hand-collected Securities and Exchange Commission (SEC) and Department of Justice (DOJ) enforcement actions for financial misrepresentation from April 1976-January 2010 (1,001 events).8 We impose several filters on these data. First, we restrict the sample to firms with a Compustat and CRSP identifier (848 remaining observations). Additionally, we consider only actions that include charges of fraud under the 1933 Securities Act or the 1934 Securities Exchange Act (646 observations remaining). Therefore, we focus on charges for intentional financial misrepresentation rather than charges for accidental accounting or auditing errors. Finally, we require that all firms have return data around the enforcement begin date (hereafter the event date), reported sales in Compustat for the two years preceding the event date, and have a historical SIC code (reported since 1987). Following Karpoff et al. (2008a, b), an enforcement action is defined as “the full chain of public releases that relate to a specific company where books are suspect.” Most enforcement actions originate from public announcements (e.g., self-disclosures, restatements, management departures) or from investigations initiated by other agencies (e.g., Department of Defense, Environmental Protection Agency). These restrictions leave a final sample of 444 events.

Table II presents the distribution of these events by industry. We follow Lang and Stulz (1992) in identifying industry rivals of the accused firm based on historical four-digit SIC codes. For the purpose of describing the sample, however, we present the industry distribution based on two-digit SIC codes. Panel A of Table II reports that the sample of accused firms spans a large number of industries and varies significantly in terms of reported sales. A large portion of the accused firms (65%) are in manufacturing and services. Panel B of Table II presents the average number of rivals for each industry across all events. This table also describes the minimum, maximum, and median number of rival firms. The maximum number of rivals varies from 13 in the less competitive industries to 469 in the most competitive ones.

Table II.  Distribution of the Sample by Industry
Panel A. Distribution of Firm Accused of Financial Misrepresentation (Event Firms)
2-digit SIC Bracket Industry Sized Based Deciles Total
1 2 3 4 5 6 7 8 9 10
01–09Mining111  3 1  7
10–14Construction  121112  8
15–17Manufacturing13 17 22 17 16 18 24 12 27 16 182  
20–39Transportationa3654 4151736 
40–49Wholesale trade124254 42125 
50–51Retail trade254218144334 
52–59Finance, Insurance125336869245 
60–67Services614 912 710 513 19 11 106  
70–89Others      1   1
 Total27475142335441476240444  
Panel B. Distribution of Rivals
2-Digit SIC Bracket Industry No. Rivals per Event
Mean Median Min Max
  1. a

    Panel A presents the number of accused firms in our sample per two-digit SIC and by size deciles. Size deciles are created based on sales using all firms in Compustat by year and historical two-digits SIC codes. Panel B reports the average (median, minimum, and maximum) number of rivals per event in a given two-digit SIC. There are a total of 28,907 rival firms in our sample.

01–09Mining58204156
10–14Construction13121 21
15–17Manufacturing38231171
20–39Transportationa25202 91
40–49Wholesale trade13102 43
50–51Retail trade23152100
52–59Finance, Insurance101 312469
60–67Services141 771397
70–89Others131313 13
aIncludes communications and utilities.

Table III presents the distribution over time for our sample of accused firms. The number of accused firms varies from 3 in 2008 to a high of 59 in 2002 (following the collapse of Enron). The average number of accused firms per year is 21.

Table III.  Distribution of Accusations of Financial Misrepresentation by Year
Year No. Firms (Events) Year No. Firms (Events)
  1. a

    The table reports the number of accused firms (events) by year. The year refers to the year of the enforcement start date. There are a total of 444 events in our sample.

198812200036
1989 9200136
1990 9200259
199116200322
199222200420
199318200524
199428200631
1995162007 7
1996132008 3
199720  
199825  
199918  
  Total444

B. Variables

The dependent variable in our analysis is the rival firm three-day abnormal stock return following the announcement of an accusation of fraudulent financial misrepresentation by a competitor. We calculate the cumulative abnormal return (CAR) by taking the difference between the firm's stock return and the value-weighted CRSP index (Brown and Warner, 1985).

To test Implication 1, the effect of the level of product market competition on rival CAR, we compute the Herfindahl Index to measure the competitiveness of the industry. The Herfindahl Index is calculated based on sales of firms in the same historical four-digit SIC code using all firms in Compustat. A higher index number represents a less competitive industry. We also compute the Herfindahl Index based on segment sales. This index is calculated based on sales of the firm's largest business segment. Firms not identified in the segment-level files are considered single segment firms and are assigned that firm's primary four-digit SIC code.

To measure industry homogeneity, we compute a homogeneity index as suggested by Parrino (1997). This proxy measures the partial correlation between monthly common stock returns and an industry equal-weighted index. A higher value for this index is associated with a higher degree of product homogeneity within the industry.

We use the accused firm's CAR and sales, as well as the ERC between each accused firm and its rivals, as proxies for the importance of the event. Firm sales are extracted from Compustat. We compute the CAR for the accused firm as the difference between the firm's return and the value-weighted CRSP index using a three-day window around the event. The ERC is derived by computing the stock price reaction of each rival to past announcements of earnings surprises of the accused firm. For this purpose, we employ the method used in the accounting literature in Foster (1981) and Freeman and Tse (1992), among others, which investigate intraindustry information transfers.9

The ERC is calculated based on quarterly information about earnings surprises of the accused firm during the three-year period preceding the event. We use the following regression specification:

  • image

where CARijt is the cumulative abnormal return of rival i for the five days around the earnings announcement of accused firm j in quarter t, and UEjt is the unexpected earnings announcement (earnings surprise) of accused firm j calculated in quarter t. The earnings surprise, UEjt, is computed as the difference between the analyst consensus quarterly earnings estimate and the realized earnings as reported in IBES. The slope coefficient, bij, is the ERC estimated for each rival i of accused firm j.

This variable conveys information about the extent to which rival firm i competes with accused firm j. A positive ERC indicates that, on average, past positive earnings surprises of the accused firm led to positive stock returns at the rival firm (low competition). A negative ERC implies that, on average, past positive earnings surprises of the accused firm led to negative stock returns to the rival (high competition). To test Implication 2, we use the property that a higher absolute value of ERC indicates a higher importance of the event for the rival firm.

Implications 3 and 4 require proxies for the opaqueness of the rival, as well as for the accused firms. To measure firm opaqueness we compute: 1) the quoted bid-ask spread, 2) the market-to-book ratio (MB) as the ratio between total debt plus market value of equity and book assets, 3) the ratio of intangibles to total assets, and 4) analyst fiscal year-end earnings per share (EPS) forecast dispersion six months prior to the event. In all regressions, we include as control variables, book leverage, computed as the book value of long-term debt divided by the book value of assets, and sales of the rival.

C. Results

Before analyzing the cross-sectional determinants of the stock price response of rival firms to the announcement of an accusation of financial misrepresentation, we first examine several univariate statistics.

1. Univariate Analysis

  1. Top of page
  2. Abstract
  3. I. Hypotheses and Implications
  4. II. Data and Empirical Results
  5. 1. Univariate Analysis
  6. 2. OLS Regressions
  7. III. Robustness
  8. IV. Conclusion
  9. References

Figure 2 and Table IV confirm that accused firms typically suffer a negative CAR of 19.7% in the period –1 to +1 days surrounding the event date. As the figure indicates, the event date is preceded by a significant negative stock return suggesting that there is some information leakage prior to the event date. For this reason, we use a three-day window (–1,+1) in the multivariate analysis.

image

Figure 2. Cumulative Return for Firms Accused of Financial Misrepresentation

The average cumulative abnormal return (CAR) of the accused firms are shown from Day –5 to Day +5, where Day 0 is the enforcement begin date (the event date or announcement date). The sample consists of 444 firms accused of financial misrepresentation where the accusation includes charges of fraud under the 1933 Securities Act or the 1934 Securities Exchange Act. CAR is computed as the market-adjusted cumulative return. The market return is the CRSP value-weighted index.

Download figure to PowerPoint

Table IV.  Sample Characteristics: Accused Firms and Their Rivals
Panel A. Accused Firms (AF)
  Mean Median STD Min Max N
CAR (–1,+1)−19.70%−11.68%24.87%−101.42%16.35%444
CAR (−2,+1)−20.32%−12.49%28.27%−104.56%21.96%444
CAR (−2,+0)−18.77%−11.68%25.61%−90.64%13.57%444
CAR (−1,0)−18.15%−10.62%21.94%−84.45%14.46%444
CAR (0,0)−17.56%−10.25%19.65%−70.10%10.09%432
SALES AF ($mil)1,8851329,0110.00051,056444
MK SHARE AF0.0550.0070.1270.0000.725444
BID ASK SPREAD AF2.65%1.38%4.51%0.00%16.07%408
MB AF2.8371.7314.5110.72116.790431
INTANGIBLES AF0.1410.0660.1860.0000.811385
FORECAST DISP AF0.8670.02011.9400.0004.590267
HERFINDAHL0.1830.1420.1480.0180.647444
PARRINO INDEX0.5140.4890.1100.3630.710444
DURABLE INDUSTRY0.4780.0000.5010.0001.000157
Panel B. Rival Firms
  1. a

    This table reports summary statistics for the sample of firms accused of financial misrepresentation (Panel A) and for their rivals (Panel B). Mean (MEAN), median (MEDIAN), standard deviation (STD), minimum (MIN), maximum (MAX), and the number of observations (N) of each variable are presented. The cumulative abnormal return (CAR) is the sum of the market-adjusted return using the CRSP value-weighted index as a benchmark over the bracketed window (in days). SALES is from Compustat measured at the fiscal year-end prior to the enforcement begin date. MK SHARE is computed as sales of the rival or accused firm (AF) divided by the sum of sales of all firms in Compustat in the same historical four-digit SIC. The BID ASK SPREAD is calculated as (closing ask-closing bid) divided by closing midpoint. For each firm, we average the bid ask spread over one year based on daily spreads. The market-to-book ratio (MB) is (total debt + market value of equity) divided by book value of assets. INTANGIBLES is the ratio of intangibles to total assets. FORECAST DISP is the analyst earnings forecast dispersion based in IBES. More than one analyst with an earnings forecast is required. We use the fiscal year-end earnings forecast closest, but prior to the enforcement begin date. HERFINDAHL is based on sales in Compustat and is computed by historical four-digit SIC codes. The PARRINO INDEX is based on Parrino (1997) and computed as the partial correlation between monthly common stock returns and an industry equally weighted index. A larger index value is associated with a higher degree of product homogeneity with the industry. The DURABLE INDUSTRY dummy is equal to one if the four-digit SIC industry is classified as delivering durable goods according to the definition of Yogo (2006). Panel B also provides the earnings response coefficient (ERC) and leverage of the rival. ERC is computed based on a regression estimating the sensitivity of the rival's abnormal announcement return to the accused firm's earnings surprises during the three years prior to the enforcement begin date using quarterly earnings surprise data from IBES. Book leverage (BLEV) is the ratio of long-term debt divided by the book value of assets. With the exception of the cumulative abnormal return (CAR), all variables are calculated for the year preceding the event date.

CAR (−1,+1)−0.54%−0.57%9.08%−25.57%27.11%28,906
CAR (−2,+1)−0.66%−0.68%10.39%−28.62%30.77%28,907
CAR (−2,+0)−0.47%−0.61%9.26%−24.90%27.11%28,907
CAR (−1,0)−0.35%−0.48%7.75%−20.77%22.72%28,906
CAR (0,0)−0.23%−0.27%5.82%−15.27%17.26%28,903
SALES Rival ($mil)652643,9080.00112,02328,888
MK SHARE Rival0.010.000.040.000.2028,888
ERC−0.01−0.023.89−11.0813.176,845
BID ASK SPREAD Rival2.83%1.70%3.78%0.05%17.12%28,354
MB Rival3.061.754.610.5322.0928,066
INTANGIBLES Rival0.090.020.150.000.6924,876
FORECAST DISP Rival0.150.021.510.002.0514,077
BLEV Rival0.210.081.95−0.091.5928,686

In Table IV Panel A, we see that accused firms have average (median) sales of $1,885 ($132) million in the year prior to the event, and an average (median) market share of 5% (0.7%). On average, the accused firms are followed by eight analysts with an EPS forecast dispersion of 0.87.

In Panel B of Table IV, we report statistics for rival firms. There are a maximum of 28,907 rivals in our sample. This number varies in our multivariate analysis depending upon the availability of rival information on multiple dimensions. We reach the smallest number of rivals (6,845) when we require information on the ERC.

Average rival CAR in the three-day window around the event date is –0.54%. Thus, generally, across all events and all rivals, the information spillover effect dominates the competition effect. However, there is great variability in these CARs with a maximum of 27.11% and a minimum of –25.57%. Rival firms have average (median) sales of $652 ($64) million in the year preceding the event, which is smaller than the sales of the accused firms. Rivals have an average of 3.8 analysts who follow them with an average EPS forecast dispersion of 0.15. Finally, average ERC is –0.01 suggesting that positive (negative) earnings surprises at the accused firm have led, in the past, to negative (positive) abnormal returns for rivals. More importantly, to test Implication 2, there is substantial variation in ERC with a standard deviation of 3.89.

2. OLS Regressions

  1. Top of page
  2. Abstract
  3. I. Hypotheses and Implications
  4. II. Data and Empirical Results
  5. 1. Univariate Analysis
  6. 2. OLS Regressions
  7. III. Robustness
  8. IV. Conclusion
  9. References

To test the implications, we begin with an OLS regression analysis with rival firm CAR as the dependent variable. This approach allows us to test the cross-sectional predictions based on rival-level characteristics, while simultaneously controlling for event-level variables, such as the CAR of the accused firm, which have the same value within events. Standard errors are heteroskedasticity-robust and corrected for clustering at the event level. In the robustness tests below, we find the inferences to be qualitatively the same when we run between effects regressions where each observation represents the average of the variable within an event. Thus, inferences about event-level variables in the rival-level OLS regressions are consistent with inferences from the event-level regressions.

Results from the OLS regressions are presented in Table V, Panels A and B. In all of these regressions, the dependent variable is the three-day rival CAR around the event.

Table V.  Multivariate Analysis
Panel A. Alternative Proxies for the Competitive Effect
Dependent Variable CAR Rival [–1,+1]
(1) (2) (3) (4) (5) (6)
Imp1: Product Market Competition
 HERFINDAHL0.023  0.0580.0220.020
 (0.04)  (0.04)(0.06)(0.10)
 HERFINDAHL SEGMENTS 0.015    
  (0.06)    
 PARRINO INDEX  0.039   
   (0.04)   
 DURABLE INDUSTRY   0.008  
    (0.08)  
Imp2: Importance of Event
 CAR AF0.0150.0190.0150.0120.0160.016
 (0.01)(0.00)(0.01)(0.30)(0.00)(0.00)
 CAR AF*HIGH HERF    –0.057 
     (0.00) 
 SALES AF–0.0000.000–0.000–0.002–0.000–0.000
 (0.91)(0.89)(0.59)(0.49)(0.87)(0.89)
 SALES AF*HIGH HERF     0.095
      (0.06)
Control Variables
 SALES RIVAL−0.002−0.003−0.003−0.004−0.002−0.002
 (0.01)(0.02)(0.01)(0.04)(0.01)(0.01)
 BLEV RIVAL−0.002−0.002−0.002−0.004−0.002−0.002
 (0.14)(0.22)(0.13)(0.20)(0.14)(0.14)
 Constant−0.0020.001−0.017−0.009−0.001−0.001
 (0.48)(0.69)(0.06)(0.15)(0.61)(0.72)
Events428428428157428428
Observations27,65527,65527,6555,28727,65527,655
R 2 0.0100.000.010.0200.0100.010
Adj. R20.00510.00470.00600.01790.00570.0056
F-stat3.623.5333.9182.695.604.09
Prob > F0.00330.00380.00170.01650.00000.0005
Panel B. Alternative Proxies for Rival and Accused Firm Opaqueness
Dependent Variable CAR Rival [−1,+1]
(7) (8) (9) (10) (11)
  1. a

    The table reports the OLS estimation results with the rival firm market adjusted return (CAR) cumulated over the three days around the enforcement begin date [–1,+1] as the dependent variable. Variables are either computed for the rival or the accused firm (AF). HERFINDAHL is based on sales in Compustat and is computed by historical four-digit SIC codes. HIGH HERF is a dummy equal to one if the rival firm is in an industry in the tenth decile of the Herfindahl industry distribution in a given year (most concentrated industries). HERFINDAHL SEGMENTS is based on sales reported in the Compustat segment database. The PARRINO INDEX is based on Parrino (1997) and computed as the partial correlation between monthly common stock returns and an industry equally weighted index. A larger index value is associated with a higher degree of product homogeneity with the industry. The DURABLE INDUSTRY dummy is equal to one if the four-digit SIC industry is classified as delivering durable goods according to the definition of Yogo (2006). SALES is from Compustat measured at the fiscal year end prior to the enforcement begin date. BLEV is the ratio of long-term debt divided by book value of assets. ERC is computed based on a regression estimating the sensitivity of the rival's abnormal announcement return to the accused firm's earnings surprises during the three years prior to the enforcement begin date using quarterly earnings surprise data from IBES. The BID ASK SPREAD is calculated as (closing ask-closing bid) divided by closing midpoint. For each firm we average the bid ask spread over one year based on daily spreads. The market-to-book ratio (MB) is (total debt + market value of equity) divided by book value of assets. INTANGIBLES is the ratio of intangibles to total assets. FORECAST DISP is the analyst earnings forecast dispersion based in IBES. More than one analyst is required with an earnings forecast. We use the fiscal year-end earnings forecast closest, but prior to the enforcement begin date. With the exception of the cumulative abnormal return (CAR), all variables are calculated for the year preceding the event date. p Values, based on standard errors that are heteroskedasticity robust and clustered at the event level, are reported in parentheses below the coefficient estimates. F-statistics of each model are reported including its p value.

Imp1: Product Market Competition
 HERFINDAHL0.0230.0180.0200.0330.003
 (0.12)(0.10)(0.08)(0.01)(0.87)
Imp2: Importance of Event      
 CAR AF0.0080.0150.0140.0150.022
 (0.47)(0.01)(0.01)(0.02)(0.01)
 ERC−0.001    
 (0.05)    
Imp3: Opacity of Rival Firm
 BID ASK SPREAD RIVAL −0.064   
  (0.07)   
 MB RIVAL  −0.000  
   (0.07)  
 INTANGIBLES RIVAL   −0.019 
    (0.03) 
 FORECAST DISP RIVAL    −0.000
     (0.78)
Imp4: Opacity of Accused Firm
 BID ASK SPREAD AF 0.023   
  (0.62)   
 MB AF  −0.000  
   (0.94)  
 INTANGIBLES AF   −0.002 
    (0.83) 
 FORECAST DISP AF    0.001
     (0.00)
Control Variables
 SALES AF−0.000−0.000−0.001−0.0000.001
 (0.82)(0.90)(0.51)(0.99)(0.22)
 SALES RIVAL−0.004−0.002−0.002−0.002−0.002
 (0.07)(0.06)(0.03)(0.01)(0.19)
 BLEV RIVAL−0.002−0.001−0.002−0.0020.000
 (0.16)(0.39)(0.16)(0.32)(0.94)
 Constant−0.003−0.003−0.001−0.001−0.001
 (0.44)(0.30)(0.83)(0.84)(0.80)
Events139392416369242
Observations6,60326,84126,50519,85610,675
R 2 0.0100.0000.0000.0100.010
Adj. R20.00830.0050.0040.0080.006
F-stat1.982.362.313.3212.05
Prob > F0.07280.02290.02520.00200.0000

A. Testing Implication 1: Product Market Competition

In Table V, Panel A, Model 1, we find a positive association between the Herfindahl Index and rival CAR. This is consistent with the competition hypothesis, which predicts that lower industry competition is associated with higher rival abnormal returns (Implication 1). In Model 2, we compute the Herfindahl Index based on segment sales and once again find a positive association between the Herfindahl Index and rival CAR. In Model 3, we determine that the Parrino Homogeneity Index is significantly positively correlated with rival CAR. Thus, rivals in industries that are more homogenous experience higher (more positive) abnormal returns around the event. This finding is consistent with the industry competition hypothesis, which predicts that rivals in industries with more homogenous products find it easier to acquire the customers of the accused firm or that customers find it easier to switch suppliers (Implication 1).

To further test whether customers care about financial misrepresentation, a prediction of the competition hypothesis, we follow the approach of Titman (1984) who finds that firms with customers who care about the firm's long-term survival because of switching costs (e.g., truck manufacturing) should have low leverage to reduce the chance of financial distress. Thus, we expect that customers are more likely to select a rival firm rather than the accused firm in industries that sell durable goods. We use the definition of durable goods industries of Yogo (2006) and find a positive correlation between rival CARs and the dummy variable indicating durable goods industries in Model 4. This finding is consistent with the importance of the industry competition effect in determining rival stock price reactions.

The results from the Herfindahl Index, the Parrino Index, and the durable goods industries are all consistent with Implication 1 of the industry competition hypothesis.

B. Testing Implication 2: Importance of the Event

Implication 2 concerns the importance of the event. The competition hypothesis predicts that a more important event is associated with a higher (more positive) rival CAR since the rival is expected to attract customers from the accused firm. Alternatively, the information spillover hypothesis predicts that more important events are associated with lower (more negative) rival CARs.

In all of the models, we find a positive association between the CAR of the accused firm, as a proxy for the importance of the event, and the CAR of the rival firm. This finding is consistent with the interpretation that the information spillover effect typically dominates the competition effect. However, given that we find higher (positive) rival CARs in more concentrated industries, consistent with Implication 1, we test whether a more important event in concentrated industries also leads to a more positive rival CAR, and whether a more important event in competitive industries leads to a more negative rival CAR. Thus, to further test Implication 2, we construct an interaction variable between the CAR (and sales) of the accused firm and the level of industry competition, HIGH HERF, that takes a value of one for rivals in industries in the highest Herfindahl decile (most concentrated), and zero otherwise. In Model 5, we find a significantly positive coefficient on the accused firm's CAR, while the interaction with HIGH HERF is significantly negative. The coefficient on the interaction variable is –0.057, while the coefficient on the accused firm's CAR is 0.016. Thus, rival firms in very concentrated industries experience a significantly higher (positive) CAR when the event is worse for the accused firm. This is consistent with the interpretation that the information spillover effect dominates the industry competition effect in competitive industries, while the reverse is true in less competitive industries.

The results are weaker when we use the accused firm's sales as a proxy for the importance of the event. In Model 6, we find a marginally significant positive coefficient on the interaction term between the sales of the accused firm and HIGH HERF indicating that rivals of a larger accused firm benefit more from the event, but only in very concentrated industries. However, we find no significant coefficient on the sales of the accused firm. Thus, rivals in more competitive industries do not seem to suffer more if the accused firm had larger reported sales.

While sales and CAR of the accused firm are event-level variables (i.e., they are all the same value for each rival of an accused firm), ERC is a proxy that differs for each rival. A positive (negative) ERC indicates that the rival's stock price reaction is positive (negative) when the accused firm reports better than expected earnings (in the three years prior to the event). A zero ERC indicates no correlation between earnings surprises in the past and rival firm value. Implication 2 predicts a negative (positive) correlation between ERC and rival CAR under the industry competition (information spillover) hypothesis. If, in the past, a rival's stock price went down upon positive earnings surprises in the accused firm, this variable suggests competition between the rival and accused firms for the same customers. In addition, the more negative the ERC, the more important the event is expected to be for the rival.

In Regression 7, we find a significantly negative coefficient on ERC. This finding is consistent with the competition hypothesis. The more negative the ERC, the more positive the rival reaction to the accusation of financial misrepresentation of the competitor. Interestingly, after introducing ERC, the coefficients on the Herfindahl Index and on the CAR of the accused firm lose significance. This is due to the fact that ERC is both a proxy for the competition between the accused firm and its rivals and a proxy for the extent to which news about the accused firm is important to the value of the rival.

The analysis of the importance of the event for the rival firms indicates that more important financial misrepresentations have, on average, a larger effect on rivals, consistent with Implication 2. Whether rival firm stock prices are affected positively or negatively, however, depends upon the competitiveness of the industry, consistent with Implication 1.

C. Testing Implication 3: Rival Firm Opacity

Implications 3 and 4 concern the opacity of the rival and accused firms, respectively. The industry competition hypothesis posits no correlation between our proxies for opaqueness and the rival firm CARs. Meanwhile, the information spillover hypothesis posits more negative rival CARs for more opaque rival firms and less negative CARs for rivals of more opaque accused firms.

Table V, Models 8–11, demonstrate a negative correlation between rival CAR and rival bid-ask spread, rival market-to-book ratio, rival intangibles-to-assets ratio, and rival analyst forecast dispersion. While the first three coefficients are significant, the last is not. One reason for this lack of significance may lie in the fact that our sample size drops from approximately 26,000 observations to about 10,000 as we require at least two analysts to be following the firm to compute forecast dispersion. Using the number of analysts (not shown), we find a significant negative coefficient. These coefficients are consistent with the information spillover hypothesis, which predicts that stock prices of more opaque rival firms should suffer more when it is more likely that the accused firm's financial information is used in setting rival stock price.

D. Testing Implication 4: Accused Firm Opacity

We find very little evidence of a significant relationship between proxies for the opaqueness of the accused firm and rival CAR. Models 8–11 in Table V display insignificant coefficients on the accused firm's bid-ask spread, market-to-book ratio, and intangibles-to-assets ratio. Only the accused firm's analyst forecast dispersion is significantly positively related to rival CAR. Thus, we conclude that the data are not consistent with the predictions of the information spillover hypothesis with regard to the importance of the opaqueness of the accused firm (Implication 4). Consequently, our results suggest that rival firm reactions are dependent upon opaqueness at the rival level, consistent with the information spillover effect, but not at the accused firm level, inconsistent with the information spillover effect of Implication 4.

Finally, note that Lang and Stulz (1992) find that leverage is a key determinant of rival firm portfolio reactions to a bankruptcy announcement by a competitor. We control for leverage (book leverage) in all regressions, but find no significant effect. This finding highlights the difference between the bankruptcy announcements these authors study and our event.

E. Economic Significance

We find that, on average, the information spillover effect dominates given that the average rival CAR is negative. However, in more concentrated industries, more homogenous industries, and durable goods industries, the competition effect dominates. In these industries, rival CARs are higher (positive).

To highlight the economic impact, we compute the aggregate dollar value impact on rivals of firms that are accused of financial misrepresentation. Figure 3 illustrates the stock market value change (in dollars) over the three days around the event date aggregated over all rivals in a given Herfindahl decile. In competitive industries (Herfindahl Deciles 1–3), the aggregate market value loss among rivals is about $295 billion, substantially more than the aggregate $80 billion loss for all accused firms in these industries. In contrast, rival firms in more concentrated industries (Deciles 8–10) gained about $690 million in market capitalization, while accused firms lost about $39 billion in these industries. Analysis of the wealth effects on rivals highlights the differing externalities caused by cases of financial misrepresentation in competitive and concentrated industries.

image

Figure 3. Aggregate Market Value Changes by Herfindahl Deciles

Aggregate stock market value (MV) changes (for rivals and accused firms) in the three days surrounding the enforcement begin date are reported by Herfindahl deciles. The stock market value change is computed as the change in market value from Day –1 to Day +1 from the enforcement begin date. All changes are added up across firms in the same Herfindahl decile, differentiating between the accused firms and their rivals. Rivals are selected using the historical four-digit SIC in Compustat. The Herfindahl index is computed based on sales of companies listed in Compustat, at the four-digit SIC level. Herfindahl deciles are formed based on one Herfindahl index per four-digit SIC industry. The lowest Herfindahl decile corresponds to the most competitive industries.

Download figure to PowerPoint

Our analysis also documents the importance of rival firm-level characteristics as determinants of rival firm stock market reaction to an accusation of financial misrepresentation by the competitor. To illustrate the additional importance of these characteristics, we form two subsamples of firms that are predicted to be more affected either by the industry competition effect or by the information spillover effect, respectively. For the subsample of firms that are predicted to be most affected by the competition effect (rivals in the least competitive industries, rivals of large accused firms with very negative event date CARs, least opaque rivals, and most opaque accused firms), we find that the average three-day rival CAR is 3.2%.10 For the subsample of rival firms predicted to be most affected by the information spillover effect (rivals in the most competitive industries, large accused firms with very negative CAR, most opaque rival, and least opaque accused firms), we find that average rival CAR is –1.5%.

Thus, we conclude that industry- and firm-level characteristics both play an important role in understanding rival firm shareholder reaction to the event, with some rivals significantly gaining and others losing. We identify the reasons behind the differing reactions to accusations of financial misrepresentation by a competitor.

III. Robustness

  1. Top of page
  2. Abstract
  3. I. Hypotheses and Implications
  4. II. Data and Empirical Results
  5. 1. Univariate Analysis
  6. 2. OLS Regressions
  7. III. Robustness
  8. IV. Conclusion
  9. References

A. Between Effects Regressions

An additional approach to testing Implications 1, 2, and 4 is to analyze the response of rivals at the event level. Using a between effects regression framework (Greene, 1997) allows us to test for event-level implications while suppressing all rival firm variation within an event. The only variation here is in average rival characteristics across events. The “between” estimator has the advantage of removing any unstructured cluster effects. Therefore, it can increase the efficiency of least squares estimation.

The dependent variable is the average rival CAR of a given event. All independent variables are event-level averages as well. Note that variables such as the accused firm's CAR remain unaffected by the averaging. However, given that each event now has only one observation (i.e., the number of observations is equal to the number of events), the problem of correlations in computing the standard errors is solved. However, testing Implication 3 (on the opaqueness of the rivals) is now a cross event test rather than a within and across event test.

Table VI reports the regression results. The sequence of the regressions is identical to that in Table V and the main conclusions drawn from the OLS regressions remain robust to this change in regression technique. In Model 1, we find that rival firm average CAR is positively related to the degree of industry competition. This is true for the various approaches used to compute industry competition (Herfindahl Index, Herfindahl segments, Parrino Index, and durable goods industries), and is consistent with Implication 1 for the competition hypothesis.

Table VI.  Between Effects Regressions
Panel A. Alternative Proxies for the Competitive Effect
Dependent Variable CAR Rival [−1,+1]
(1) (2) (3) (4) (5) (6)
Imp1: Product Market Competition
 HERFINDAHL0.026  0.0570.0240.023
 (0.01)  (0.01)(0.01)(0.02)
 HERFINDAHL SEGMENTS 0.016    
  (0.06)    
 PARRINO INDEX  0.044   
   (0.00)   
 DURABLE INDUSTRY   0.009  
    (0.09)  
Imp2: Importance of Event
 CAR AF0.0150.0190.0150.0120.0160.016
 (0.01)(0.00)(0.01)(0.22)(0.00)(0.00)
 CAR AF*HIGH HERF    −0.055 
     (0.11) 
 SALES AF0.000−0.000−0.000−0.0010.0000.000
 (0.91)(0.96)(0.97)(0.79)(0.96)(0.94)
 SALES AF*HIGH HERF     0.091
      (0.10)
Control Variables
 SALES RIVAL−0.003−0.000−0.004−0.008−0.003−0.003
 (0.59)(0.95)(0.49)(0.33)(0.66)(0.63)
 BLEV RIVAL−0.009−0.008−0.0100.002−0.009−0.009
 (0.03)(0.08)(0.02)(0.81)(0.03)(0.03)
 Constant0.0000.0020.006−0.0100.0010.001
 (0.91)(0.41)(0.66)(0.10)(0.79)(0.72)
Events428428428157428428
Observations27,65527,65527,6555,28727,65527,655
R 2 0.040.040.050.100.050.05
Adj. R20.03250.02960.04140.06100.03490.0342
F 3.8733.6014.6862.6883.5763.517
Prob > F0.00190.00340.00040.01660.00180.0021
Panel B. Alternative Proxies for Rival and Accused Firm Opaqueness
Dependent Variable CAR Rival [−1,+1]
(7) (8) (9) (10) (11)
  1. a

    The table presents the parameter estimates of the between effects (events) regressions using weighted least squares to adjust for unbalanced data. The rival firm market-adjusted return (CAR) cumulated over the three days around the enforcement begin date [–1,+1] is the dependent variable. Variables are either computed for the rival or the accused firm (AF). HERFINDAHL is based on sales in Compustat and is computed by historical four-digit SIC codes. HIGH HERF is a dummy equal to one if the rival firm is in an industry in the tenth decile of the Herfindahl industry distribution in a given year (most concentrated industries). HERFINDAHL SEGMENTS is based on sales reported in the Compustat segment database. The PARRINO INDEX is based on Parrino (1997) and computed as the partial correlation between monthly common stock returns and an industry equally weighted index. A larger index value is associated with a higher degree of product homogeneity with the industry. The DURABLE INDUSTRY dummy is equal to one if the four-digit SIC industry is classified as delivering durable goods according to the definition of Yogo (2006). SALES is from Compustat measured at the fiscal year-end prior to the enforcement begin date. BLEV is the ratio of long-term debt divided by the book value of assets. ERC is computed based on a regression estimating the sensitivity of the rival's abnormal announcement return to the accused firm's earnings surprises during the three years prior to the enforcement begin date using quarterly earnings surprise data from IBES. The BID ASK SPREAD is calculated as (closing ask-closing bid) divided by closing midpoint. For each firm, we average the bid ask spread over one year based on daily spreads. The market-to-book ratio (MB) is (total debt + market value of equity) divided by the book value of assets. INTANGIBLES is the ratio of intangibles to total assets. FORECAST DISP is the analyst earnings forecast dispersion based in IBES. More than one analyst is required with an earnings forecast. We use the fiscal year-end earnings forecast closest, but prior to the enforcement begin date. With the exception of the cumulative abnormal return (CAR), all variables are calculated for the year preceding the event date. p Values, based on standard errors that are heteroskedasticity robust, are reported in parentheses below the coefficient estimates. F-statistics of each model are reported including its p value.

Imp1: Product Market Competition
 HERFINDAHL0.0140.0170.0200.0330.010
 (0.29)(0.10)(0.04)(0.00)(0.53)
Imp2: Importance of Event      
 CAR AF0.0260.0170.0140.0170.022
 (0.00)(0.00)(0.01)(0.01)(0.01)
 ERC0.000    
 (0.17)    
Imp3: Opacity of Rival Firm
 BID ASK SPREAD RIVAL −0.236   
  (0.00)   
 MB RIVAL  −0.003  
   (0.04)  
 INTANGIBLES RIVAL   −0.023 
    (0.21) 
 FORECAST DISP RIVAL    −0.001
     (0.92)
Imp4: Opacity of Accused Firm
 BID ASK SPREAD AF 0.100   
  (0.05)   
 MB AF  0.000  
   (0.85)  
 INTANGIBLES AF   0.001 
    (0.95) 
 FORECAST DISP AF    0.000
     (0.08)
Control Variables
 SALES AF0.000−0.000−0.0010.0000.001
 (0.78)(0.83)(0.71)(1.00)(0.68)
 SALES RIVAL−0.0040.004−0.005−0.001−0.004
 (0.22)(0.60)(0.43)(0.84)(0.68)
 BLEV RIVAL−0.001−0.006−0.010−0.005−0.002
 (0.15)(0.10)(0.01)(0.25)(0.82)
 Constant0.001−0.0050.0070.001−0.000
 (0.63)(0.12)(0.10)(0.77)(0.99)
 Events139392416369242
 Observations6,60326,84126,50519,85610,675
R 2 0.010.060.050.050.04
Adj. R20.0060.0420.0350.0340.016
F 7.3513.4263.1432.8411.556
Prob > F0.00010.00140.00300.00680.1490

Implication 2 predicts that the more important the event is, the higher (lower) the rival CAR under the competition (information spillover) hypothesis. Consistent with the OLS results in Table V, we find a significant positive coefficient on the accused firm's CAR and a negative coefficient on the interaction term between the accused firm's CAR and HIGH HERF in Model 5 of Table VI. While the interaction term has a p value of only 0.11, the size of the coefficient is virtually the same as in the OLS regression indicating that firms in very concentrated industries experience a higher (positive) average rival CAR the worse the accused firm's reaction to the event. The findings using the accused firm's sales as a proxy for the importance of the event are again weaker. Only the interaction term between the accused firm's sales and HIGH HERF is significant, consistent with the importance of the competition effect (Model 6).

One major difference from the OLS regression is that we find an insignificant coefficient on ERC in Model 7. However, this is less surprising given that the between effect regression takes the average across all rivals. Thus, the Herfindahl Index and the accused firm's CAR seem to be stronger event-level proxies for the level of competition than average ERC.

Implication 3 relates to the opacity of the rival firm. The information spillover hypothesis predicts that more rival opacity leads to lower (more negative) rival CAR. In Models 8-11 of Table VI, we find that proxies for average rival opaqueness are negatively related to average rival CAR. However, only the coefficients on the bid-ask spread and the market-to-book ratio are statistically significant. These findings also suggest that average rival firm opaqueness is related to rival stock price reaction, consistent with the predictions of the information spillover hypothesis and Implication 3; however, it seems that within event variation of opaqueness is an important driver of rival firm CAR.

The between effects framework provides additional support for Implication 4 relative to the OLS regressions. Implication 4 predicts a positive correlation between rival CAR and proxies for the opacity of the accused firm. Both the bid-ask spread and the analyst forecast dispersion of the accused firm are positively related to the average rival CAR, although the coefficients are only marginally significant.

B. Industry-Adjusted Regressions

One remaining question is whether the OLS results are driven predominantly by average rival characteristics across events (industries) or individual rival firm differences within an event. To address this question, Table VII provides regressions with industry-adjusted variables. The industry adjustment is performed using all Compustat firms’ median value within a four-digit SIC by year of a given variable. The dependent variable is rival firm CAR. The results from the industry-adjusted regressions are qualitatively the same as those from the OLS regressions. Thus, we conclude that differences between rival firms within an event play a significant role in explaining the rival firm announcement reaction above and beyond industry-level differences (Lang and Stulz, 1992).

Table VII.  Industry Adjusted Regressions
Panel A. Alternative Proxies for the Competitive Effect
Dependent Variable CAR Rival [–1,+1]
(1) (2) (3) (4) (5) (6)
Imp1: Product Market Competition
 HERFINDAHL0.020  0.0550.0180.016
 (0.09)  (0.05)(0.12)(0.19)
 HERFINDHAL SEGMENTS 0.014    
  (0.06)    
 PARRINO INDEX  0.038   
   (0.05)   
 DURABLE INDUSTRY   0.009  
    (0.06)  
Imp2: Importance of Event
 CAR AF0.0160.0200.0160.0110.0170.017
 (0.01)(0.00)(0.01)(0.34)(0.00)(0.00)
 CAR AF*HIGH HERF    –0.060 
     (0.00) 
 SALES AF (Iadj)−0.0000.000−0.000−0.000−0.000−0.000
 (0.82)(0.96)(0.52)(0.46)(0.78)(0.79)
 SALES AF*HIGH HERF     0.104
      (0.04)
Control Variables
 SALES RIVAL (Iadj)−0.000−0.000−0.000−0.000−0.000−0.000
 (0.01)(0.02)(0.00)(0.03)(0.01)(0.01)
 BLEV RIVAL (Iadj)−0.001−0.001−0.001−0.004−0.001−0.001
 (0.23)(0.30)(0.23)(0.22)(0.23)(0.23)
 Constant−0.0010.000−0.017−0.010−0.001−0.001
 (0.55)(0.87)(0.06)(0.09)(0.69)(0.82)
 Events428428428157428428
 Observations27,65527,65527,6555,28727,65527,655
R 2 0.000.000.010.020.010.01
Adj. R20.00450.00450.00580.01720.00510.0051
F-stat3.40803.56703.84102.57504.99104.0090
Prob > F0.00500.00360.0020.02090.00010.0006
Panel B. Alternative Proxies for Rival and Accused Firm Opaqueness
Dependent Variable CAR Rival [−1,+1]
(7) (8) (9) (10) (11)
  1. a

    The table reports the OLS estimation results with the rival firm market adjusted return (CAR) cumulated over the three days around the enforcement begin date [–1,+1] as the dependent variable. Variables are either computed for the rival or the accused firm (AF). (Iadj) following the variable name indicates that the variable has been industry-adjusted by the industry median. HERFINDAHL is based on sales in Compustat and is computed by historical four-digit SIC codes. HIGH HERF is a dummy equal to one if the rival firm is in an industry in the tenth decile of the Herfindahl industry distribution in a given year (most concentrated industries). HERFINDAHL SEGMENTS is based on sales reported in the Compustat segment database. The PARRINO INDEX is based on Parrino (1997) and computed as the partial correlation between monthly common stock returns and an industry equally weighted index. A larger index value is associated with a higher degree of product homogeneity with the industry. The DURABLE INDUSTRY dummy is equal to one if the four-digit SIC industry is classified as delivering durable goods according to the definition of Yogo (2006). SALES is from Compustat measured at the fiscal year-end prior to the enforcement begin date. BLEV is the ratio of long-term debt divided by the book value of assets. ERC is computed based on a regression estimating the sensitivity of the rival's abnormal announcement return to the accused firm's earnings surprises during the three years prior to the enforcement begin date using quarterly earnings surprise data from IBES. The BID ASK SPREAD is calculated as (closing ask-closing bid) divided by closing midpoint. For each firm, we average the bid ask spread over one year based on daily spreads. The market-to-book ratio (MB) is (total debt + market value of equity) divided by the book value of assets. INTANGIBLES is the ratio of intangibles to total assets. FORECAST DISP is the analyst earnings forecast dispersion based in IBES. More than one analyst is required with an earnings forecast. We use the fiscal year-end earnings forecast closest, but prior to the enforcement begin date. With the exception of the cumulative abnormal return (CAR), all variables are calculated for the year preceding the event date. p Values, based on standard errors that are heteroskedasticity robust and clustered at the event level, are reported in parentheses below the coefficient estimates. F-statistics of each model are reported including its p value.

Imp1: Product Market Competition
 HERFINDAHL0.0190.0140.0180.029−0.003
 (0.21)(0.23)(0.15)(0.03)(0.89)
Imp2: Importance of Event
 CAR AF0.0120.0180.0160.0160.025
 (0.31)(0.00)(0.01)(0.01)(0.01)
 ERC−0.001    
 (0.06)    
Imp3: Opacity of Rival Firm
 BID ASKED SPREAD RIVAL(Iadj) −0.053   
  (0.08)   
 MB RIVAL (Iadj)  −0.000  
   (0.10)  
 INTANGIBLES RIVAL (Iadj)   −0.011 
    (0.09) 
 FORECASTS DISP RIVAL (Iadj)    −0.001
     (0.51)
Imp4: Opacity of Accused Firm
 BID ASK SPREAD AF (Iadj) 0.110   
  (0.06)   
 MB AF (Iadj)  0.000  
   (0.95)  
 INTANGIBLES AF (Iadj)   0.000 
    (0.97) 
 FORECASTS DISP AF (Iadj)    0.000
     (0.00)
Control Variables
 SALES AF (Iadj)−0.000−0.000−0.000−0.0000.000
 (0.74)(0.67)(0.47)(0.89)(0.20)
 SALES RIVAL (Iadj)−0.000−0.000−0.000−0.000−0.000
 (0.06)(0.06)(0.03)(0.01)(0.20)
 BLEV RIVAL (Iadj)−0.002−0.001−0.002−0.0010.000
 (0.29)(0.58)(0.24)(0.40)(0.74)
 Constant−0.002−0.001−0.001−0.002−0.000
 (0.64)(0.75)(0.64)(0.44)(1.00)
 Events139392416368240
 Observations6,60326,84126,50519,85610,675
R 2 0.010.010.000.010.01
Adj. R20.0070.0050.0040.0080.008
F-stat1.7742.4592.2342.77518.710
Prob > F0.1090.01770.03070.00800.0000

Yet another way to test for the importance of rival firm characteristics is to run event fixed effects regressions. These regressions include only rival firm variables. We find that the inferences drawn from Table VII hold true with event fixed effects regressions (not shown).

C. Changes in Market Share

In the next robustness test, we examine the change in the realized market share of rival firms. The industry competition hypothesis implies that the accusation of financial misrepresentation by one firm can change the rival firm's market share. In competitive industries, this change is likely to be smaller than in concentrated industries. We calculate the change in the rival's market share as the difference between the rival's market share in the four quarters preceding and the four quarters following the event. Total market size is based on firm sales information from Compustat. Results are qualitatively similar when we use the sales of the largest segment of the firm (not shown).

In Models 1–4 of Table VIII, we report the results of OLS regressions where the dependent variable is the change in rival firm market share. In Model 1, we find a positive and significant coefficient on industry concentration (Herfindahl Index). This finding is consistent with the interpretation that firms in concentrated industries are better able to extract market share from the accused firm. The table also indicates that the larger the lagged market share of the accused firm, the larger the market share gain of the rival firm. Additionally, we find that our proxies for rival firm opaqueness (high bid-ask spread, high analyst forecast dispersion, and high MB) are negatively related to rival firm change in market share. This is consistent with the interpretation that information effects related to the accusation of financial misrepresentation have real consequences (Durnev and Mangen, 2009). Overall, these results are consistent with the interpretation that event date rival CARs, at least in part, reflect future changes in the market share of rival firms.

Table VIII.  Other Robustness Checks
Dependent Variable (1) ΔRival Market Share (2) ΔRival Market Share (3) ΔRival Market Share (4) ΔRival Market Share (5) CAR [–1,+1] (eliminate clustered industry events) (6) CAR [–2,0] (different window)
  1. a

    Columns (1)–(4) of the table report the OLS estimation results with the rival firm change in market share (MK SHARE) as the dependent variable. The change in market share (Δ MK SHARE) is calculated as the difference between the average market share in the four quarters preceding the event and the average market share in the four quarters following the event. In Column (5), the dependent variable is the cumulative abnormal return (CAR) in the event window [–1,+1]. We eliminate events that occur in the same industry within the next 30 days. CAR is the sum of the market-adjusted return in the respective event window. In Column (6), we run the full sample OLS regression for a different window of [–2,0]. Variables are either computed for the rival or the accused firm (AF). HERFINDAHL is based on sales in Compustat and is computed by historical four-digit SIC codes. The BID ASK SPREAD is calculated as (closing ask-closing bid) divided by closing midpoint. For each firm, we average the bid ask spread over one year based on daily spreads. The market-to-book ratio (MB) is (total debt + market value of equity) divided by the book value of assets. INTANGIBLES is the ratio of intangibles to total assets. FORECAST DISP is the analyst earnings forecast dispersion based in IBES. More than one analyst is required with an earnings forecast. We use the fiscal year-end earnings forecast closest, but prior to the enforcement begin date. SALES is from Compustat measured at the fiscal year-end prior to the enforcement begin date. BLEV is the ratio of long-term debt divided by the book value of assets. With the exception of the cumulative abnormal return (CAR), all variables are calculated for the year preceding the event date. p Values, based on standard errors that are heteroskedasticity robust and clustered at the event level, are reported in parentheses below the coefficient estimates. F-statistics of each model are reported including its p value.

Imp1: Product Market Competition
 HERFINDAHL0.0120.0120.0120.0190.0180.019
 (0.00)(0.00)(0.00)(0.00)(0.00)(0.03)
 MK SHARE Rival (LAG)−0.085−0.072−0.070−0.127  
 (0.00)(0.00)(0.00)(0.00)  
 MK SHARE AF (LAG)0.0320.0300.0310.039  
 (0.01)(0.02)(0.01)(0.00)  
Imp2: Importance of Event
 CAR AF−0.001−0.001−0.001−0.0010.0210.015
 (0.65)(0.54)(0.43)(0.09)(0.00)(0.01)
Imp3: Opacity of Rival Firm
 BID ASK SPREAD RIVAL−0.012   −0.044−0.065
 (0.04)   (0.03)(0.06)
 MB RIVAL −0.001    
  (0.00)    
 INTANGIBLES RIVAL  0.003   
   (0.00)   
 FORECAST DISP RIVAL   −0.001  
    (0.01)  
Imp4: Opacity of Accused Firm
 BID ASK SPREAD AF−0.008   0.0640.025
 (0.17)   (0.34)(0.59)
 MB AF 0.000    
  (0.86)    
 INTANGIBLES AF  −0.001   
   (0.26)   
 FORECAST DISP AF   0.001  
    (0.00)  
Control Variables
 SALES AF    −0.001−0.001
     (0.72)(0.88)
 SALES RIVAL    −0.003−0.002
     (0.06)(0.06)
 BLEV RIVAL−0.000−0.000−0.000−0.000−0.001−0.001
 (0.26)(0.16)(0.07)(0.74)(0.39)(0.39)
 Constant0.000−0.000−0.000−0.0000.002−0.002
 (0.66)(0.30)(0.38)(0.39)(0.56)(0.34)
Events391415427240368391
Observations25,28724,96026,07010,07922,46825,287
R 2 0.030.030.030.080.010.01
Adj. R20.030.030.030.080.0060.005
F-stat5.7396.3167.37811.4202.3382.370
Prob > F0.0000.0000.0000.0000.02530.0229

D. Repeated Events within an Industry

Table II reports some industry concentration among accused firms. In particular, there is a large number of events in the software industry (SIC code 7372), as well as in several other industries to a lesser extent. To test that our results are not driven by a particular industry and are not clustered in time, we run our basic multivariate specification (Model 8, Table V) on the subsample of rivals associated with all financial misrepresentation events that are not preceded by another event in the same industry within a period 30 days prior to the event date (this restriction eliminates most multiple industry events, leaving us with 364 events with available data). Model 5 in Table VIII reports the results from this regression and determines that they remain significant and with the same signs as in Table V. Thus, our main findings are not driven by any particular industry. In further tests, we find that using only the first accusation of financial misrepresentation in an industry (in our database) leaves the inferences unchanged (not shown). This is important, as Gande and Lewis (2009) find that the announcement return of an accused firm does not reflect the full extent of the event if the firm is in an industry in which previous events have occurred as the stock price has already reacted to some extent to the previous accusations of financial misrepresentation of a rival.

E. Different Event Windows

We conduct our main analysis using a three-day event window [–1, +1] to compute the abnormal announcement returns for the accused firm and its rival firms. We re-run our regressions using a [–2, +0] event window and obtain qualitatively similar results, as shown in Model 6 of Table VIII.

For our main measure of rival CAR, we use market-adjusted daily abnormal returns with the CRSP value-weighted index as a benchmark. We find that the results are qualitatively similar (not reported) using the Fama and French (1993, 1992) three-factor model instead of the CRSP index as a benchmark.

F. Financial Misrepresentation versus Bankruptcy

Lang and Stulz (1992) investigate rival reactions to bankruptcy announcements. If fraudulent financial misrepresentation were to lead to bankruptcy announcements (quickly), rival reactions to the announcement of accusation of financial misrepresentation might be driven by the expectation of the accused firm going bankrupt. To obtain a clearer picture of what happens to our sample of firms accused of financial misrepresentation in subsequent years, we combine the delisting codes in CRSP with the follow-up information about these events provided in the KLM database. We find that 72% of our sample firms are still active two years after public accusation of financial representation, 18% are in bankruptcy, 5% are acquired, 2.7% are privatized, 1.6% are delisted for various reasons, and 0.7% are liquidated. Reasons for delisting include delinquency in filing, nonpayment of fees, not meeting minimum exchange standards (float, minimum stock price, information filed), or at company request. From this analysis, we conclude that the event of accusation of fraudulent financial misrepresentation does not imply future bankruptcy, but simply represents a discrete corporate event. As mentioned above, unlike Lang and Stulz (1992), we do not find that rival leverage plays a significant role in determining rival CAR. Thus, financial misrepresentation is likely to propagate through channels other than bankruptcy news.

IV. Conclusion

  1. Top of page
  2. Abstract
  3. I. Hypotheses and Implications
  4. II. Data and Empirical Results
  5. 1. Univariate Analysis
  6. 2. OLS Regressions
  7. III. Robustness
  8. IV. Conclusion
  9. References

This paper examines how the announcement of an accusation of fraudulent financial misrepresentation affects industry rivals of the accused firm. Our results highlight the fact that while rival firms generally tend to suffer from the announcement of accusation of financial misrepresentation, there is significant cross-sectional variation. We find that rival firm characteristics, as well as industry characteristics, explain this cross-sectional variation. In addition, we find that both the industry competition hypothesis and the information spillover hypothesis lead to predictions that are consistent with the data.

This paper helps shed light on the broader costs and benefits of financial misrepresentation, extending the work of Karpoff et al. (2008a) and Gande and Lewis (2009) that each analyze the costs to the accused firm of financial misrepresentation. The analysis presented here is intended to further the understanding of the broader impact of financial misrepresentation, while at the same time providing additional evidence regarding the importance of the industry competition effect and the information spillover effect.

Footnotes
  • 1

    We thank Karpoff, Lee, and Martin for providing us with an updated version of their data.

  • 2

    A rival firm is defined based on historical four-digit SIC codes, extracted from Compustat.

  • 3

    The New York Times article “Trying to Catch Worldcom's Mirage” (6/30/2002) reflects this idea: “ … AT&T's shares had been hammered by investors concerned about the company's cable strategy and because Wall Street did not believe that AT&T's core operations were being run as well as, say, WorldCom's” and “Several top executives said last week that competing against WorldCom for the attention of investors and Wall Street analysts in recent years was essentially like running track against an athlete who is later discovered to be using steroids.”

  • 4

    Research on corporate fraud takes many different perspectives. Karpoff, Lee, and Vendrzyk (1999) study firms indicted for procurement fraud in government contracts. Goldman and Slezak (2006) theoretically analyze optimal executive compensation and its effect on fraud, while Burns and Kedia (2006) and Bergstresser and Philippon (2006) empirically demonstrate that earnings management and financial misreporting are motivated by stock-based compensation. Johnson, Ryan, and Tian (2009) confirm a positive association between stock-based compensation and the fraudulent manipulation of accounting statements, while Erickson, Hanlon, and Maydew (2006) argue that there is no such link. Miller (2006), Dyck, Morse, and Zingales (2009), and Yu and Yu (2011) examine various aspects relating to fraud detection. Finally, Gande and Lewis (2009) find that firms facing shareholder class action lawsuits suffer a loss in value before their actual indictment.

  • 5

    Eckbo (1983), Eckbo and Wier (1985), and Shahrur (2005) examine the impact of mergers on industry rivals. Olsen and Dietrich (1985) look at retail sales announcements. Docking, Hirschey, and Jones (1997) examine bank loan loss announcements. Laux et al. (1998) study rival stock price reactions to dividend revisions. Slovin, Sushka, and Poloncheck (1999) examine competitive effects and contagion at commercial banks. Bittlingmayer and Hazlett (2000) look at the impact of antitrust rulings against Microsoft. Chen, Ho, and Ik (2005) investigate rival responses to new product announcements. Hung-Chia, Reed, and Rocholl (2010) consider rival responses to the announcement of an IPO by a large entrant. Zhang (2010) examines the effect on industry competitors when firms emerge from bankruptcy. Akhigbe, Borde, and Whyte (2000) find positive information spillover effects for rivals of M&A targets at the time of the takeover announcement.

  • 6

    It is also possible that fraud can have an effect on firms along the supply chain. While we do not study such firms, Hertzel et al. (2008) look at the impact of bankruptcy on firms along the supply chain and Shahrur (2005) studies the effects of horizontal mergers on customers and suppliers.

  • 7

    Several other studies focus on the event of earnings announcements and earnings restatements that are closer to the type of event we describe, but that are not detrimental to the firm's ability to continue in business and to compete in the market. These studies include, for example, Han and Wild (1990), Pownall and Waymire (1989), and Ramnath (2002).

  • 8

    The data are from the Karpoff et al. (2008a, b) database, which has been graciously provided to us by the authors (KLM database hereafter). It consists of all enforcement actions initiated by the SEC and DOJ from April 1976 to January 2010 for violation of one or more of three provisions of the Securities and Exchange Act of 1934, as amended by the Foreign Corrupt Practices Act of 1977: 1) 15 U.S.C. §§ 78m(b)(2)(A) that requires firms to keep and maintain books and records that accurately reflect all transactions; 2) 15 U.S.C. §§ 78m(b)(2)(B), which requires firms to devise and maintain a system of internal accounting controls; and 3) 15 U.S.C. §§ 78m(b)(5), which establishes that no person shall knowingly circumvent or knowingly fail to implement a system of internal accounting controls or knowingly falsify any book, record, or account.

  • 9

    This is slightly different from the standard ERC used in the literature that focuses on the firm's own ERC (e.g., Collins and Kothari, 1989; Kothari and Sloan, 1992).

  • 10

    A rival firm is selected to be part of the subgroup based on the following selection process. For each criterion (e.g., competitiveness of industry), we place the rival firms into five buckets. Rivals in the most competitive industries go into Bucket 1. Those in the most concentrated industries go into Bucket 5. We apply the same assignment process to the other four criteria. We then select rival firms that have a combined score of 23, 24, or 25 for the industry competition subsample (5, 6, or 7 for the information spillover subsample). Note that the score of the bucket depends upon the hypothesis. We have a sample of 451 rival firms that should be most affected by the industry competition effect, and a sample of 231 firms most affected by the information spillover effect.

References

  1. Top of page
  2. Abstract
  3. I. Hypotheses and Implications
  4. II. Data and Empirical Results
  5. 1. Univariate Analysis
  6. 2. OLS Regressions
  7. III. Robustness
  8. IV. Conclusion
  9. References