Differential Interpretations, Private Information and Trading Volume Around French Firms' Good News vs. Bad News Preliminary Announcements

ABSTRACT

This study provides new evidence that both differential interpretations and private information production spur trading volume for a sample of 144 preliminary earnings announcements in the French markets. After partitioning the sample into preliminary announcements that convey good news versus bad news, I find that good news stimulates more production of private information, whereas bad news leads to more differential interpretations. I further find that increased production of private information (but not differential interpretations) helps explain trading volume around good news preliminary earnings announcements. In contrast, differential interpretations (and not private information) help explain trading volume around bad news preliminary earnings announcements.

Acknowledgements

I thank Linda Bamber, Eugene Kandel, Orie Barron, Donal Byard and Neil Battacharya for their comments and suggestions. The paper has benefited from comments by participants at the American Accounting Association meeting (2005), the French Finance Association meeting (2002), the PACAP/FMA Finance Conference (2002), and ESA-ParisXII (2002). A previous draft of this paper was entitled: Differential Interpretations and Trading Volume Around Preannouncements. I also gratefully acknowledge the help of the two anonymous referees of the EAR. I am also grateful for ParisBourse SA, and JCFinance SA for their Database.

Notes

¹Differential interpretations among analysts stem from differences in their long-term beliefs, their valuation functions, their differential information processing ability, or even their irrationality.

²Chahine (2004) analyses the PA effect of 106 firms during the period 1998 to 2000. He shows that revised forecasts of earning-per-share (EPS_0,1) following PAs of bad news reflect 70.5% of the actual earnings (EPS_0,2), whereas the adjustment of earnings forecasts following PAs of good news reflect 41.4% of the actual earnings.

³Bris et al. (2004) and Charoenrook and Daouk (2004) present a comparative analyses of short-selling practices across the world. Both papers include a brief descriptive analysis of short-selling in France. They indicate that although certain restrictions exist in France, short selling occurs. Bris et al. (2004) distinguish four groups of countries according to whether short-selling is prohibited and or practiced. They include France and the US in the same group of countries.

⁴Soffer et al. (2000) count 1,161 preliminary announcements in the U.S. from September 1992 to December 1995.

⁵“Préliminaire” or “Provisoire” in French. These PAs have been reported in PRLine SA web site (www.prline.com), JCFinance database and other popular French financial press (Les Echos, La Tribune, etc.).

⁶For example, Assystem, a French engineering company, announced on January 16, 2002 sales revenues of 220 M[euro] and a preliminary operating margin rate of 8.5%. On April 2, 2002, it announced actual sales revenues of 218.2 M[euro] and an operating margin rate of 7.2%.

⁷The control sample of non-preannouncing firms matched on proximity to earnings announcement dates. The control sample of 144 firms is chosen over the 2350 firm-years from 1998 to 2003, for which I obtained annual earnings announcement dates, actual earnings, the analysts' forecasts and the number of financial analysts following these firms over the 3 months prior to actual earnings announcements. Firms in the control sample need to have at least 3 analysts' forecasts available in the JCFinance database. Selected firms are within the same industry using the Paris Stock Market industrial classification. They have a market capitalization within the range of + / − 20% of those of comparable studied firms at the time of PAs, and have an actual announcement within 20 calendar days of the announcement date of the studied firms.

⁸The average number of days between PA s and actual earnings announcements is 30 days.

⁹Further empirical investigations indicate that trading volume in windows including days − 4 through − 2, days − 1 through + 1, and days + 2 through + 4 is significantly higher than in the nonannouncement period. Although not tabulated here, empirical results indicate a higher abnormal trading volume around the PAs.

¹⁰There is no case where the abnormal volume is equal to zero.

¹¹As JCFinance database reports analysts' forecasts rounded to 6 digits, very few PAs cause 0% change in average forecasts.

¹²In order to understand the implications of the related measurement error, I first calculated the proportion of Case III revisions for the entire sample and distinguishes between good and bad news PAs (this indicates no significant difference in Case III for good and bad news PAs, 12.39% and 13% respectively). Second, I added the proportion of Case III and its interaction variable (good news) to primary and control variables, and I found positive but non significant coefficients for both the main effect (the proportion of Case III) and its interaction, and no significant changes in main inferences. This suggests that the limitation of DI related to Case III does not show any significant difference between good and bad news PAs, and is unlikely to affect the main inferences of the paper.

¹³It is equal to: ρ _j = 1-D _j /V _j , where ρ _j reflects the degree to which the analysts' beliefs covary relative to the overall level of uncertainty for firm j, or the ratio of common uncertainty to the overall (common and idiosyncratic) uncertainty.

¹⁴The Appendix explains assumptions necessary to interpret the Barron et al. (1998) correlation metric, ρ, as a measure of the percentage of common-to-total information contained in the mean forecast, and s as the precision of private information possessed by individual analysts.

¹⁵Prior research uses the standard deviation of analysts' forecasts (the dispersion in prior beliefs) as a proxy for differential precision of private pre-disclosure information (Chung and Jo, 1996). However, Barron et al. (1998) and Barron et al. (2002) demonstrate that the standard deviation of prior beliefs reflects not only differential precision of private pre-disclosure information, but also the uncertainty among analysts. This unwanted component creates measurement error in the “dispersion” metric. Controlling for dispersion (i.e., standard deviation of analysts' forecasts) in pre-PA window does not affect inferences for private information and DI.

¹⁶The elimination of the 29 firms included in the highest quintile of information commonality (17 firms with good news and 12 firms with bad news) reduces the median values of Δs for both PAs with good and bad news to 17.35% and 0%, respectively. These 29 firms are significantly smaller, less followed, and are more likely to have a higher cost of private information production. This indicates that firms with relatively little private information before the PA are more likely to experience an increase in the proportion of information that is private.

¹⁷Firm size is positively and significantly related to analysts following (a correlation coefficient of 61.21% at the 0.01% level). Controlling for the number of financial analysts in further empirical tests does not change the inferences.

¹⁸Voluntary disclosure of bad news can reduce negative surprise and litigation-related costs or reputational costs with financial analysts and institutional investors (Skinner, 1994). In fact, Baginski et al. (2002) compare the legal systems in the U.S. and Canadian markets. They show that Canadian managers, in a less litigious environment, issue more forecasts when earnings increase, whereas U.S. managers, in a more litigious environment, issue more forecasts in periods where earnings decrease.

¹⁹The Case IV inconsistencies of 3.21% in Kandel and Pearson (1995) is a weighted average calculated using inconsistencies in both explicit and implicit forecast revisions. Kandel and Pearson (1995) find 13.35% inconsistencies for 1,822 explicit forecast revisions and 0.86% inconsistencies for 7,864 implicit forecast revisions.

²⁰Additional analysis of the 31 firms followed by less than 9 analysts show that they have a smaller size and a larger forecast error after the post-preliminary announcement than the remaining sample (at the 10% and the 1% levels respectively).

²¹In line with Barron et al. (2002), scaling variance and squared error by the absolute value of actual earnings-per-share or stock price at fiscal year end does not affect my inferences about the production of private information.

²²Controlling for both book-to-market and the volatility of earnings over a three-year period prior to the PA date does not affect any of the study's inferences.

²³Further analysis includes the magnitude of the earnings forecast revisions to control for changes in aggregate beliefs. While Kandel and Pearson (1995) model the effect of differential interpretations of public announcements on the value of the firm, my empirical tests are based on how PAs affect analysts' forecasts of upcoming annual earnings. Controlling for the magnitude of the forecast revision (in addition to the magnitude of the price change) controls for items that affect upcoming earnings forecasts but not firm value. I defined the magnitude of the earnings forecast revisions as the natural logarithm of the absolute percentage change in the median forecast over the pre- and post-windows around the PA to adjust for skewed data. Including it as a control variable in empirical investigations does not show any significant effect on both trading volume and the main inferences of the study.