I start by investigating whether there is asymmetry even well before the news being publicly released.
Asymmetric average returns generated by good versus bad news
Following Kothari, Shu, and Wysocki (2009), I run the following baseline regression:
whereRit is the quarterly excess return over Fama and French (1993) market, size, book-to-market factors and Carhart (2001) momentum factor, or the residuals of running a Carhart four-factor model in quarter t for firm i. N egit is a dummy variable for firm iin quarter t that equals one for negative earnings surprises, and zero otherwise. I also conduct the F-test to examine whether stock market reactions to negative versus good earnings surprises are the same. Specifically, we test if 2*|α0|=|β0|. Since the magnitude for stock returns to negative earnings surprises is |α +β| with the β being negative, so testing 2*|α0|=|β0| is equivalent to test
|α+β|=|α|.
Table 4.2 presents the results. FordE/P, positive earnings surprises generate average stock return of +3.1% around the fiscal quarter the earnings cover. Negative surprises capture average return of −4.2% (= 0.031−0.073). The F-test confirms that the market’s reaction to earnings decreases is significantly at 1% level larger in magnitude than the reaction to earnings increases, with a difference of −1.1%. When I divide the firms into quintiles according to ascendingρ, the higher average returns for negative news remain for each of the quintile. The estimates of the model are presented in panel A columns 3-7. A1/D5 means the lowest quintile for ambiguity also the fifth quintile for DoO, and so forth. Then, I plot the difference between the average negative returns by negative news and average positive returns by positive news. Figure 4.1 shows that the differential returns exhibit a ”yes” tick shape with respect toρ, with the lowest point in either second or third quintile. The results are the same for all three earnings surprises measures.
The market reaction to earnings decreases might be more pronounced be- cause (1) the amount of total news revealed is greater for bad news disclosures and/or (2) the information content per unit of news is greater because investors are more surprised by the bad news disclosures. I firstly examine the former that how information amount contributes to the asymmetric reaction to bad and good news. I use the measures for managers’ incentives to release bad news. First, Reg FD
arguably has limited the ability of managers to informally leak news either good or bad to experts and institutional investors prior to the announcements of quarterly earnings. This implies that firms before the passage of Reg FD are more incentivized to release (bad) news. Second, to mitigate litigation risk, firms are more likely to disclose bad news before it becomes public. Third, firms with high information asymmetry tend to release earlier all kinds of news to avoid market penalties from investors. Fourth, reputation risks limit the ability of managers to withhold bad news particularly when the firms are in financial distress. I examine the effects of those factors using the following regression:
Rit = α0+β0N egit+β1RegF Dit+β2RegF Dit∗N egit+β3LitRiskit (4.2) + β4LitRiskit∗N egit+β5Inf oAsymmit+β6Inf oAsymmit∗N egit + β7F inDistressit+β7F inDistressit∗N egit+it,
whereRegF Dis a dummy variable equal to one if the announcement occurs before the passage of Regulation FD in October 2000, and zero otherwise. LitRisk is a dummy variable that equals one if the firm has less than median litigation risk cal- culated using Rogers and Stocken (2005) predictive regression, and zero otherwise.
Inf oAsymm is a dummy variable that equals one if the firm is above the median
value of a single information asymmetry factor, and zero otherwise. The informa- tion asymmetry factor is derived from a factor analysis based on the information asymmetry proxies: market-to-book ratio, stock volatility, high-tech firms, financial leverage, and regulatory status. F inDistress is a dummy variable that equals one if the firm’s Z-score (Zmijewski, 1984) financial distress rank is in the top decile of all firms in a given year, and zero otherwise. I present the F-tests of whether the estimated intercept coefficient is equal in magnitude to coefficient for the dummy variable N eg. Note that this is different from the F-test for equation 4.1, which is to examine the asymmetric reaction (in magnitude) to good versus bad earnings surprises. Here, I test whether the identified asymmetry could be explained entirely
by the information amount of news proxied by the managerial incentives to release or withheld bad news. If the explanation is in full, then|α0+β0|should be statis- tically indifferent from zero, in other words |α0| should be statistically indifferent from|β0|. Thus, I test|α0|=|β0|.
Panel A Column 2 of table 4.3 reports the estimates of regression model (4.3) for earnings surprise measure dE/P. RegF D∗ N eg, Inf oAsymm ∗N eg,
and LitRisk∗N eg have the anticipated negative signs and are significant at 1%
level. F inDistress∗N eg is indistinguishable from zero. The coefficient for N eg
drops to -0.043 from -0.073 of the baseline regression. The rest is captured by the interaction terms mentioned above. There is−1.2% of average return for negative news unaccounted for and significant at 1% level shown by the F-test at the bottom of column 2. This return is not significantly different from the −1.1% differential return recorded in table 4.2 (not tabulated).
Secondly, I examine the explanatory power of the informational content of the news per se by including the size of the news as well as its interaction with the directional dummy variable in the baseline equation 4.1:
Rit=α0+β0N egit+δ1sueit+δ2sueit∗N egit+it, (4.3) wheresueit is the standardised unexpected earnings and refers to one of the three earnings surprises measures. Panel A of table 4.4 presents the regression results for model (4.3) withsue=dE/P. As can been seen, the average negative return gen- erated by negative earnings surprises is around−2.8%, computed as the sum of the estimated intercept and N egit coefficients. This is indistinguishable in magnitude from the average positive returns by positive news 2.9%: the p-value for the F-test of their equal magnitude stands at 0.217. Here, the F-test is the same as in equation 4.1. The interaction variable sueit∗N egit is positive and highly significant at 1% level. It shows that for a given percentage change in earnings, investors’ reaction to earnings decrease is much more pronounced than to earnings increases. This implies
that investors are more surprised by bad news. In other words, the informational content of per unit of bad news is higher than that of good news. The F-test shows that this surprise explains the differential average firm-level stock returns of bad versus good news in its entirety. The results are similar for each of quintiles.
Figure 4.2 plots the differential returns of each quintile after controlling the informational contents of news per se. The ”yes” tick shape disappears and more importantly those differential returns are insignificant at 5% level. Figure 4.3 plots the coefficients for interaction termsueit∗N egit, which generally matches the ”yes” tick shape. It further enhances the explanation that it is the larger information content of bad news per se explains the differential average returns generated by bad versus good news.
Finally, I include both the information content per unit of news and the fthe proxies for managerial incentives as controls and run the following regression:
Rit = α+β0N egit+δ1sueit+δ2sueit∗N egit+β1RegF Dit (4.4)
+ β2RegF Dit∗N egit+β3LitRiskit+β4LitRiskit∗N egit+β5Inf oAsymmit + β6Inf oAsymmit∗N egit+β7F inDistressit+β7F inDistressit∗N egit+it,
Panel A of table 4.5 reports the estimates of the model (4.5) for dE/P.
sueit∗N eg is significantly positive at 1% level for all firms and for each quintile. For the regression with all firms, the reaction to per unit of bad news is more than 4.5 times (i.e. 00..491108) of that to per unit of good news measured by the coefficient of
sueit. pvalueof 0.299 for the F-test confirms that all the differential average return of
−1.1% of the baseline model comes from the differential reactions to the information contents of bad versus good news. Figure 4.4 plots the differential earnings response coefficients for bad versus good news. The ”yes” tick shape remains.