Do investigated companies manipulate profitability data?

Transcription

1 Do investigated companies manipulate profitability data? Ludivine Garside (University of Bristol and CMPO) Paul Grout (University of Bristol and CMPO) Anna Zalewska (School of Management, University of Bath and CMPO) December 28 Abstract In this paper we consider whether companies manipulate their profitability data in response to regulatory investigations. In particular, we investigate whether companies reported profitability during an investigation of abuse of a monopoly position tends to be lower than pre-investigation profitability. First, in a theoretical model, we show that in equilibrium companies manipulate profitability data once an investigation starts. We then test this proposition on evidence from UK competition cases and find that there are significant differences in reported profitability during an investigation when compared to pre-investigation profit levels. JEL classification: G38, G39 CMPO, Department of Economics, University of Bristol, 2 Priory Road, Bristol, BS8 1TX, phone: +44 () , L.Garside@bristol.ac.uk CMPO, Department of Economics, University of Bristol, 2 Priory Road, Bristol, BS8 1TX, phone: +44 () , P.A.Grout@bristol.ac.uk Corresponding author: School of Management, University of Bath, Bath BA2 7AY, phone:+44 () , A.Zalewska@bath.ac.uk

2 1. Introduction In almost all jurisdictions there are competition law provisions of some form prohibiting the abuse of a monopoly (or dominant) position. The basic approach typically has three steps. First, a market is defined. Second, it is determined whether the company is dominant on that market. Finally, if the company is dominant, then evidence is gathered and assessed to decide whether the company abused a monopoly position or not. The final stage requires the collection and interpretation of relevant evidence, e.g., profitability, market shares, behaviour, contracts, etc. There is a clear incentive structure implicit in this process. Once a company knows that it is being investigated it has an incentive to try to change its behaviour and manipulate what evidence there is to put the company in a better light and reduce the chances of being found guilty. Unless the investigation process is without error, the incentive will exist for innocent and guilty companies. Of course, the investigating body will recognise the incentive effects and may choose to treat evidence of actions after the company knows it is being investigated differently from evidence of actions before the investigation is announced. This paper outlines a simple theoretical model of this effect and, with a data set of specific competition cases from the UK, uses evidence of predicted guilt and other data to test whether companies appear to engage in this behaviour. We find evidence that companies respond to investigation in this way. A problem with addressing case evidence is that it is hard to allocate information to pre- and during-investigation, particularly in a form that can be compared over time to see if there is a break in behaviour. In this paper we focus on profitability and consider if there is a difference in reported profitability during-investigation compared to pre-investigation. Compared to most other evidence, profitability has several advantages. It is quantifiable and is measured successively for discrete periods (hence a time path can be followed with pre- and during-investigation periods relatively well). Furthermore, profitability is generally thought to be a measure that has scope for manipulation especially where there is some scope to change it during a short period. The paper uses a data set of all the companies investigated by the UK Competition Commission (CC) for suspected abuse of monopoly power from 197 till This data set is probably the only data set able to conduct an investigation of this type for several reasons. First, the UK is almost the only jurisdiction to collect and publish sufficient profitability data on a regular basis. This is recognised in Office 1 Formally the organisation was referred to as the Monopolies and Mergers Commission (MMC) throughout almost all this period but for simplicity, throughout the paper, the terminology Competition Commission will be referring either to the UK Competition Commission (as it is now called) or the MMC (as it was called before the 1998 Competition Act). 1

3 of Fair Trading Publications: The UK seems to be one of the few jurisdictions where the usefulness of profitability assessment has been explicitly recognized, and where it is regularly applied in investigations. 2 Second, the legal process was virtually identical throughout the period we consider, allowing for some degree of legal consistency. Third, during this period the test was whether the monopoly action under investigation operated against the public interest (i.e., adverse finding, which can loosely be thought of as guilty), which gives an ambiguous outcome, making it easier to predict the probability of guilt from pre-investigation evidence. Finally, after each case the CC must produce a detailed summary of the case and the decision (amounting to several hundred pages in some cases), which provides a good summary of information (although we only use hard numerical characteristics in this investigation). The profitability figures in the report have been collected by the CC staff and provide a careful measure of profitability in the relevant market, and as such are far more reliable and informative than accounting measures. There is some related empirical literature on the economic causes of verdicts in competition law, which we note here. The closest is Davies et al. (1998, 1999) who draw on UK data to investigate causes of decisions and analyse 73 cases investigated by the CC (a subset of our data) between 1973 and 1995, finding market share of the largest firm in a case to be a major factor in the outcome. Their data set of cases overlaps with the one used in the current paper but their paper is not concerned with the question of how companies respond to the incentives in a legal investigation. Lauk (22) applies a case approach to 196 observations on both monopoly and merger cases investigated by the German Federal Cartel Office between 1985 and 2. Finally, Neven and Röller (2) discuss the tenets underlying a competition policy investigation, albeit as a premise to a theoretical discussion of jurisdictional conflict. None of these papers look at profitability data and so are only tangentially related to our paper. Indeed the role of profitability, particularly accounting profitability has been the centre of considerable debate in economics for many years. 3 The structure of the paper is as follows. Section 2 provides a semi-formal model of the process we consider. Here the government agency, usually the competition authority, is aware that a company is able to adjust profitability data during the investigation period and the government agency adjusts its interpretation of this evidence, compared to pre-investigation profitability data, accordingly. The company 2 Office of Fair Trading, Assessing Profitability in Competition Policy Analysis (July 23). 3 See Fisher and McGowan (1983) and the ensuing literature. Grout and Zalewska (27) includes a summary of the issues. 2

4 knows that the government assumes that the company will reduce profitability during investigation and so has to reduce reported profitability to prevent the government agency concluding that profits are increasing (potentially indicating that the company is more likely to be taking advantage of a monopoly position). The model has an equilibrium where the extent that the government agency predicts that the company will try to manipulate profit is exactly equal to the level of manipulation. So the whole process is self-defeating in one sense but is the natural outcome of the investigative process. The actual profit in any period depends on a stochastic process so observed outcomes are random but a conclusion of the model is that, other things equal, the expected pre-investigation profit will be greater than the expected value of reported profit during the investigation period. Section 3 of the paper outlines the data we use to test this prediction. Section 4 provides the empirical results. We assume that profit in any year depends on the profit of the previous year and various other factors. We then ask if the profit level is also sensitive as to whether the reported measure comes from within the investigation period or is pre-investigation. We show that the during-investigation dummy is negative and significant. Section 5 discusses these results. 2. Theoretical model In this section we provide a semi-formal discussion of a model. The model has two periods, period and period 1. At the start of period 1 the company is unexpectedly investigated for potential abuse of a monopoly position. During the investigation the government agency collects data on profitability and other relevant information about the case. We denote the latter by Z. The company cannot influence Z or profit reported in period. However, the company is able to implement some unobservable effort which is costly but reduces the observed profit in period 1 by an amount e. There is an underlying level of profitability for the company,, and the probability that the government agency will find the company guilty depends on the government agency s estimate of the underlying level of profitability and Z. A lower estimate of reduces the probability of being found guilty. Let p denote the probability of the company being found guilty. The cost to the company of implementing effort to reduce reported profit by e is c(e). This is a differentiable increasing convex function of e with derivative of zero at e =. Furthermore, we make the realistic assumption that there is a limit as to how far the company can manipulate the profit. Specifically, there is a bound e on e such that the 3

5 cost of effort approaches infinity as e approaches e from below. Let the reported profit in period 1 be denoted by 1, and the level of profit if there is no manipulation be denoted by 1n. Hence the reported period 1 profit is equal to: = e 1 1n. We assume that profit generated by the company in period is a random variable 2 normally distributed with mean, and variance σ. Therefore, the pre-investigation profit,, can be expressed as: = + ε, (1) where is the true underlying profit level and ε is a normally distributed random 2 variable, i.e., ε ~ N (, σ ). 4 If there is no manipulation of the profit, i.e., e =, then we assume that profit in period 1 depends on the true underlying profit and the level of profit that is drawn in period zero. Specifically, we assume: α. where the parameter [, 1] 1 n = α + (1 α ) + ε, (2) Assume that the company has to choose e at the start of period 1, i.e., before the company knows the exact value of 1n. That is, when the choice of e is made the company knows that 1n will be determined by Eq. (2) but does not know the draw from the distribution. The government agency knows that a company will manipulate the profit in period 1, but does not know by how much. The government agency holds a belief of how much effort the company makes to reduce its reported profit in period 1. This belief is denoted by ê. Given ê the company will choose a level of manipulation, e, i.e., e is a function of ê. We define an equilibrium e E, e E ) fixed point ( e = eˆ ) of this function, i.e., if the government agency believes that the (ˆ as a 4 Here we are analysing the position of a single firm under investigation. If we imagine that the firm is picked for investigation from a potential pool of firms then this relationship between pre-investigation profit and underlying profit implicitly assumes that there is no sample selection such that firms that have an abnormally high pre-investigation profit are more likely to be chosen. This is a reasonably good assumption in practice since an abuse generally only involves a small part of a company s activities and the profitability of this part of the company only becomes known upon detailed investigation. 4

6 effort level is ê E and the company knows this belief then the company will indeed choose effort equal to ê E. We assume that the behavioural function for the government agency that determines the probability of the company to be found guilty, p, is of the form ( ˆ Z) p = f,, (3) where Z denotes all other relevant observable characteristics of the case and denotes the government agency s estimate of ˆ. To ensure the second order conditions are satisfied the partial function of f with respect to ˆ is assumed to be an unbounded convex function. Given that the government agency s belief of how much the company manipulates its profit in period 1 is ê then, if the reported profit in period 1 is 1, the government agency s belief of the non-manipulated profit in period 1 is 1+ ê. Given this belief Eq. (2) provides a period 1 estimate of, i.e., 1 ( 1 α) α + eˆ ε. α This is also a period estimate of (from Eq. (1)), i.e., it is equal to ε. Therefore, using Bayes theorem, we have: (1 ) ˆ 1 α + e ˆ = γ + (1 γ ), (4) α 2 where γ = 1 (1+α ). Note, that γ depends only on α because of the simple process that determines and pre-investigation period profit level. 1n. The government agency always puts more weight on the On the assumption that the company is profit maximising, the company will seek to minimise the sum of the expected cost of being found guilty, i.e., the expected value of the fine, pf, plus the cost of manipulation, c(e). More precisely, for given 1n, the company wants to solve: min e ( pf c( e) ) +, 5

7 or, using Eqs. (3) and (4): + ˆ 1 (1 α) e min f γ + (1 γ ), Z F+ c( e). (5) e α Therefore, for given 1n the minimisation problem becomes: + ˆ 1n e (1 α) e min f γ + (1 γ ), Z F+ c( e), e α However, the company does not know Therefore, the minimisation problem is: 1n at the time of determination of e. Let + ˆ 1n e (1 α) e min E f γ + (1 γ ), Z F+ c( e). (6) e α e be the argmin of Eq. (6). e) e ( ˆ is a non-decreasing function. Fig. 1 shows the function for given parameters values and given realisation of. Note that e is always strictly less than e. So e is less than ê at ê = e. Furthermore, because the derivative of c(e) is zero at e =, e will be greater than at ê =. Continuity implies that there must be at least one fixed point. Hence we have equilibrium. In equilibrium: e eˆ >. E = E That is, the government agency believes that the company manipulates profit by an amount e ˆ >, the company knows that this is what the government agency believes E and it exactly chooses e E = eˆ E >, i.e., exactly fulfils the government agency s expectations. It follows that in equilibrium: and E ( ) = E( e 1 ) = α + (1 α ) E( ) e =., Hence, E ) > E( ). ( 1 6

8 Figure 1. Function e ( eˆ ) for e ˆ [, e]. e e e E êe e ê 3. Data In Garside, Grout & Zalewska (27) we examine whether and how the individual attributes of the investigators work alongside more traditional economic variables to predict verdicts in antitrust cases. We find that the more experienced the chairman of the investigating panel is the more likely a firm is to be found guilty. We use the core model from this paper to produce the predicted adverse findings used in this study. The data subset used in this research targets only the 4 cases where profitability figures are disclosed in the publicly available CC case reports. In line with existing empirical competition studies (Davies et al. 1999, Lauk 22), the characteristics of the company with the largest market share on the reference market are used to define the case. 3.1 Profitability data and Within-investigation period As part of its activities, the Office of Fair Trading (OFT) identifies market situations potentially harmful to competition or to the public interest. Faced with the possibility of undesirable likely economic consequences, the Director-General of Fair Trading makes a reference to the CC, specifying which actions deserve further scrutiny. 5 The date when the reference is made marks the beginning of the CC investigation. In our dataset, flagging the year when the investigation starts allows us to separate reported 5 In general, the OFT considers that the likely effect of a dominant undertaking s conduct on customers and on the process of competition is more important to the determination of an abuse than the specific form of the conduct in question. OFT, December 24, Abuse of a dominant position Understanding Competition Law, OFT 42, p.18. 7

9 annual ROCE observations that are pre-investigation (hence are free from possibility of manipulation of the underlying profitability) and those where the firm could potentially choose a profitability-reducing effort level. Figure 2 shows a clear decline in profitability as the investigation starts. This is in contrast to the increases in previous periods. Interestingly those cases that are subsequently not found guilty have a larger drop than cases that are subsequently found guilty. The profitability figures appear to drop significantly for the few cases that have two or three withininvestigation profit observations. However, we cannot make this inference because there are almost no cases in this position so the averages for these few cases cannot be meaningfully compared to the average profitability at t and (t-1) since the latter two periods are averaged over virtually all the cases in the sample. Figure 2: Aggregate mean ROCE in 6-year window dataset (39 firms), breakdown by actual verdict 7% Aggregate mean ROCE (%) 6% 5% 4% 3% 2% 1% % t-3 t-2 t-1 t t+1 t+2 Time line (CC investigation starts in year t) Adverse findings by CC No adverse finding by CC The provision of pre- and within-investigation data is quite varied. Most companies report one within period profit level, although a few have two and even three. Similarly, most companies provide no more than two or three years of preinvestigation data. At the other extreme, one company has 14 years of preinvestigation data. While it is tempting to use longer histories where available, adding such long runs of data risks biasing the sample by over-representing a few companies and includes data that is likely to have limited relevance to the play of incentives as the company starts being investigated. 8

10 Clearly, very recent pre-investigation profitability data is most appropriate for our analysis. We therefore cut our data set in several ways. In most regressions we limit ourselves to very recent data (using three pre-investigation observations, i.e., in total no more than the six most recent profit observations per firm). We also consider up to seven observations per firm and also use all the available data. As is shown in Section 4, results are not sensitive to whether six or seven years are used as the ceiling on observations per firm but are slightly weaker when every profitability figure is included no matter how far the start of the series is from the investigation period. For the majority of regressions we use the six-year window since this gives us a workable data set where firms are relatively equally represented and all observations are reasonably close to the investigation period. Following the theoretic model of Section 2 we assume that current-period profitability is affected by profitability in the previous period (and other factors). So we work with profitability transitions between two consecutive periods (although as a robustness check we take the difference between current and lagged profitability as the dependent variable for some regressions). Table 1 shows the gradual attrition of new transition data entering the dataset as windows grow larger than 7 years. Window size Table 1 Transition frequencies, breakdown by window size Increment by adding one year to window Number of transitions in window Total number of transitions in window 2-year year year year year year year year year year year year year year year year

11 We do not correct profitability numbers for differences in company risk. There are two reasons for ignoring this. One is that the CC do not calculate company risk measures (e.g., Capital Asset Pricing Model). Hence practically there is no reason to suppose it affects their decisions and legally the CC should not use evidence to come to a decision that they do not report. The second is that the spread and scale of ROCE numbers is so large that adjustment for a company s risk changes over time (using any sensible range for the equity risk premium) will have negligible impact on the relative differences in ROCE between periods Predicted Findings We assume that the prediction of guilt based on data available immediately preinvestigation contains much of the information that may affect changes in profitability. We use alternative predictions of adverse findings, produced on the 431 firm-level dataset used in Garside, Grout and Zalewska (27). The main prediction is obtained from the core model in this paper. It is a probit regression with actual adverse findings (1 if yes, otherwise) as the dependent variable, and using chairman experience, gender ratio of the panel, company market share, repeated investigation dummy, dummy for multiple types of anti-competitive conducts featured in the case, and a political climate variable. A second prediction of adverse findings is also made, ignoring experience of the chairman and gender variables. A third prediction includes a snapshot of profitability for the year before the investigation starts, and therefore covers a smaller dataset (119 firms). Unless otherwise indicated, the terminology predicted findings refers to our first prediction. Figure 3 illustrates the relationship between higher mean profitability per firm and higher predicted probability of adverse findings, while also capturing our profitability sample outlier (whose effect is discussed in the next section). This shows a weak negative relationship between ROCE and predicted findings. Further details of the data set are given in Tables 2 and 3. 6 Grout and Zalewska (26) find significant beta changes due to anticipated major policy changes. However the effects on beta are less than.5 so it is reasonable to suppose that the changes in company betas during the years where profitability data is used are likely to be less than.5. 1

12 Figure 3: Scatter diagram of predicted and actual findings against mean ROCE per firm in 6-year window dataset (39 firms) 1.2 Probability of Adverse findings % % 5% 1% 15% 2% 25% Mean ROCE per firm (%) Actual Findings Predicted Findings Table 2: Correlation matrix of key variables in 6-year window dataset (N = 121) Predicted finding Actual adverse finding Within-investigation Lagged profitability Profitability Profitability 1 Lagged profitability Within-investigation Actual Adverse finding Predicted finding

13 Table 3 Mean and range of key variables in the 6-year window dataset (N = 121) Variable Mean Range Difference in profitability 2.2% % 146.8% Profitability 42.2% -33.3% 315.1% Lagged profitability 4.% -75.5% 315.1% Within-investigation.37 1 Predicted finding Actual adverse finding Regression Results Most regressions take profitability as the dependent variable. This is regressed on one-period lagged profitability, a measure of growth in the economy (which could cause profitability to rise), actual or predicted adverse findings, the withininvestigation dummy and further case characteristics. While the various cases are independent from one another, the profitability transitions within a given case are not. If we do not take account of this correlation we under-estimate standard errors. Hence, we cluster our data by case to obtain robust standard error estimates Main regressions Table A1 shows our first eight regressions. Specification A1(I) uses (i) one-period lagged returns on capital employed, (ii) macroeconomic growth defined as annual growth percentage in UK gross domestic product at constant 2 prices, (iii) predicted adverse findings obtained as discussed in sub-section 3.2 and (iv) withininvestigation variable taking the value of one for financial years during the CC investigation and zero otherwise. Lagged profitability is statistically significant at 1%. The CC investigation having started has a negative impact on current profitability, with 5% significance level. These results are robust to increasing the window size up to seven years (Table A1(II)). When we use all of the historical profitability data available (Table A1(III)), coefficients signs remain unaffected, magnitudes are slightly reduced and significance of the within-investigation dummy is diluted to 1%. 12

14 Macro-economic growth remains insignificant throughout, and our next step is to omit it from the regression in our two windows of interest (Table A1(IV-V)). Other results are largely unaffected. One firm exhibits much larger than average profitability levels for the two years preceding the investigation. 7 When we dummy for this specific case (Table A1(VI)), earlier results still stand and are therefore not driven by this outlier. Further, we impose a unit coefficient on lagged profitability by running our model on the difference in profitability between the current and lagged period. For obvious reasons the overall fit of the model falls (.7 adjusted R-squared), however the within-investigation variable remains negative and significant at 5% in the six-year window (Table A1(VII)). Its significance drops to 1% in the seven-year window (Table A1(VIII)) Evidence on further case characteristics Next, for robustness, we take into account a variety of case characteristics using the six-year window. Our attention first turns to the type(s) of monopoly action(s) under investigation, which the OFT stipulates at the start of the investigation in its reference to the CC. We use categories of anti-competitive actions, taking value one if the action is displayed in the case and zero otherwise, as defined in Garside, Grout and Zalewska (27). We also use an alternative specification with more condensed categories, whereby all pricing practices are regrouped and the various exclusivity restraints are treated as one. Our key results are robust to both specifications with extensive or condensed categories of actions (Table A2(I-II)). The within-investigation variable stays negative and significant at 5%. This also holds if we switch the dependent variable to the difference in profitability (Table A2(V-VI)). We also consider the total number of firms investigated within the case (Table A2(III)). We also consider market share inequality, defined as the ratio of the largest market share to the combined second and third largest market shares on the reference market (Table A2(IV)). Results are unaffected. 7 Bryant & May Ltd, investigation report Cm 1854, Matches and Disposable Lighters, published

15 It is possible that there are firm specific effects that we have not explicitly modelled. Equations VII and VIII in Tables A1 and equations V to VI in Table A2 would not suffer from any such problems since a firm specific effect would cancel as a result of taking the difference in profitability as the dependent variable. However, Equations I to VI in Table A1 and equations I to IV in Table A2 could be affected by such a problem and if this is the correct specification then it is important to investigate whether eliminating any firm specific effect removes our result. To check this we difference the regressions, i.e., take the equation for t-1 away from the equation for t, and estimate the difference equation. All variables drop out except for the difference in lagged profitability, the difference in within investigation dummy and the difference in macroeconomic growth. The difference in within-investigation dummy variable is tantamount to a start of investigation variable taking the value of one for the first financial year during the CC investigation and zero otherwise. We regress these on the six-year (Table A2 equation VII) and seven-year windows (Table A2 equation VIII). Taking differences reduces the number of observations that we can use and we are pushing the data so to speak, nevertheless, the within investigation dummy has the correct sign and is significant at 1%. in both regressions, suggesting that either there are no firm specific effects or if there are firm specific effects that they are not significant for our the results Different specifications of the adverse findings variable Our third area of scrutiny is to vary the definition of the adverse findings variable. We make use of the two alternative predictions described in sub-section 3.2 (Table A3(II-III)). We also specify an estimation using the actual findings (Table A3(I)). The sign and magnitude of the coefficients on the predicted adverse finding variable varies from one specification to the next, and are usually insignificant or only very narrowly significant at 1%. Hence we also regress our model excluding any predicted or actual findings (Table A3(IV)). Throughout the four changes in specification, coefficients and significance levels of lagged profitability and of the within-investigation variable remain stable. Finally, we apply the same four changes in specification while using the difference in profitability as the dependent variable (Table A3(V-VIII)). Within-investigation is negative and significant at 5%. 14

16 5. Conclusions In this paper we consider whether companies manipulate their profitability data in response to regulatory investigations. In particular, we investigate whether companies reported profitability during an investigation of abuse of a monopoly position tends to be lower than pre-investigation profitability. Given that the government agency knows that the company tries to reduce reported profit and the company knows that the government agency knows that the company tries to reduce reported profit, then it is useful to model the process to show that the equilibrium will indeed involve expected falls in profitability data. We show that such an equilibrium exists. We then test this proposition on evidence from UK competition cases and find that there are significant differences in reported profitability during an investigation when compared to pre-investigation profit levels. 15

17 References Clarke, Roger; Stephen Davies and Nigel Driffield,1998, Monopoly Policy in the UK: Assessing the Evidence, Edward Elgar Publishing: Cheltenham, UK. Davies, Stephen W.; Nigel L. Driffield and Roger Clarke, 1999, Monopoly in the UK: What determines whether the MMC finds against the investigated firms?, Journal of Industrial Economics 47(3), Franklin M. Fisher, and John J. McGowan, 1983, On the Misuse of Accounting Rate of Return to Inter Monopoly Profits, American Economic Review 73, Garside, Ludivine; Paul A. Grout and Anna Zalewska, 27, Does Within-tenure Experience Make You Tougher? Evidence from Competition Law, mimeo, University of Bristol. Paul A. Grout and Anna Zalewska, 26, The Impact of Regulation on Market Risk, Journal of Financial Economics 8, Paul A. Grout and Anna Zalewska, 27, Measuring the Rate of Return for Competition Law, forthcoming in Journal of Competition Law and Economics. Lauk, Martina, 22, Econometric analysis of the Decisions of the German Cartel Office, EARIE 22 (plenary session), Madrid, Spain. Neven, Damien J. and Lars-Hendrick Röller, 2, The allocation of jurisdiction in international antitrust, European Economic Review 44(4-6),

18 Appendix Table A1. Basic Model Dependent Variable Profitability Difference in profitability I II III IV V VI VII VIII Constant Lagged profitability Macro growth Predicted finding Within-investigation.185 (.134).854 (.61).13 (.13) (.164) -.14 (.65).182 (.122).839 (.52).17 (.12) (.154) (.58).137 (.95).837 (.47).14 (.9) -.13 (.121) -.11 (.51).26 (.152).852 (.64) -.17 (.149) (.68).29 (.142).834 (.56) -.13 (.144) (.63).129 (.12).711 (.187).17 (.16).1 (.141) (.48).67 (.384).119 (.84).14 (.15) (.126) -.15 (.7) Outlier case dummy Observations Adjusted R (.88).19 (.14) (.132) (.65) 17

19 Table A2. Evidence on case characteristics Dependent Variable Profitability Difference in profitability I II III IV V VI VII VIII Constant Lagged profitability Macro growth Predicted finding Within-investigation Monopoly pricing Discriminatory pricing Collusive pricing Predatory pricing Pricing practice Tie-in Sales Vertical integration Exclusive distribution Exclusive purchasing Exclusive restraint Number of firms investigated Market share inequality Lagged difference in profitability.32 (.17).821 (.52).13 (.11) -.29 (.22) (.7) -.13 (.123) (.11) -.18 (.119).4 (.87) (.93) -.21 (.113) (.65) (.94).289 (.161).836 (.5).13 (.11) -.52 (.156) (.69) (.81).4 (.14) (.1) (.87).223 (.158).844 (.66).13 (.13) (.178) (.65) (.129).181 (.128).831 (.87).12 (.13) (.182) -.14 (.64).14 (.118).196 (.112).16 (.14) -.3 (.125) (.78) (.93) (.99) (.13).36 (.74) (.76).33 (.16) -.82 (.58) -.11 (.78).187 (.112).14 (.13) -.44 (.1) (.77) -.96 (.69).49 (.12) (.88) -.95 (.72).54 (.3).315 (.149).5 (.27).329 (.137) 18

20 Difference in withininvestigation dummy Difference in macro growth (.9).11 (.6) Observations Adjusted R (.88).11 (.5) 19

21 Table A3. Different specifications of the adverse finding variable. Dependent Variable Profitability Difference in profitability I II III IV V VI VII VIII Constant Lagged profitability Macro growth Actual adverse finding Adverse finding predicted without information about the chairman Adverse finding predicted with past profitability.9 (.47).847 (.48).12 (.12).34 (.45).356 (.171).84 (.58).18 (.13) -.48 (.218).124 (.72).861 (.49).13 (.13).15 (.44).857 (.54).12 (.12).51 (.4).13 (.13) -.3 (.44).256 (.119).19 (.15) (.173).91 (.58).15 (.15) -.38 (.77) -.78 (.81) Within-investigation (.65) (.62) (.66) (.65) (.7) (.68) (.7) Observations Adjusted R (.32).13 (.13) -.15 (.7) 2