I. INTRODUCCION
6. Marco Teórico
6.3 La eficiencia y eficacia empresarial en el mercado
6.3.1 La racionalidad en el uso de los recursos
6.3.1.2 La eficacia y la eficiencia
Hypothesis 2 predicts that investors use recommendation revisions to update their interpretation of past earnings surprises. It predicts a positive association between earnings surprises and abnormal returns around confirmatory revisions and a nega- tive association around contradictory revisions. To test these predictions, I estimate the following version of equation (2.12) using the 177, 792 firm-days with recom- mendation revisions following one of the sample’s earnings surprises:
ARt = α + β0Xt+ γ1C on f i r mt+ γ2C ont r ad i c tt+ γ3SU Eq+ (2.19)
+ γ4C on f i r mtSU Eq+ γ5C ont r ad i c ttSU Eq+ ²t.
(2.19) regresses abnormal returns surrounding a recommendation revision on day t on the last preceding earnings surprise interacted with dummies to indicate whether the revision on day t is confirmatory or contradictory. While individual earnings surprises can enter the estimation through different firm-day observations, each firm-day with at least one recommendation revision is included once at most. As some of the sample’s recommendation revisions are made concurrently with other recommendations for the same firm, I pool all recommendations released for a firm on the same day into one observation. For these observations C on f i r mtand
C ont r ad i c tt are 1 in case at least one analyst confirmed or contradicted the preced-
ing earnings surprise on this day, and 0 otherwise.
The focus in (2.19) is onγ4andγ5. Hypothesis 2 suggests that as investors in-
crease their initial estimate of the fraction of an earnings surprise that is permanent around confirmatory revisions, returns move into the direction of the earnings sur- prise (γ4> 0), and they decrease this estimate around contradictory revisions, result-
ing in a negative association between past surprise and abnormal returns (γ5< 0).
Recommendation revisions are likely to bring more to the market than only new information about past earnings surprises. I thus add a set of controls (Xt) to (2.19).
The controls include the change in recommendation ranks (4Rect) since Womack
(1996) shows that stock prices respond strongly to changes in recommendations. I calculate 4Rect as the average recommendation change across all analysts that
release a revision for the same firm on the same day t . Prior literature finds that recommendation revisions tend to have larger effects in worse information envi- ronments (e.g., Stickel, 1985). Xt therefore includes dummies for quarterly firm size
deciles and their interaction with 4Rect. It is possible that the response to recom-
the recommendation change (e.g., Jegadeesh et al., 2004; Barber, Lehavy, and True- man, 2010; Stickel, 1995). I add indicator variables for all 5 IBES recommendation ranks and their interaction with 4Rect as additional controls.14 Finally, I include
dummy variables for each quarter of the sample period and the 48 Fama-French in- dustries as well as their interactions with 4Rect to allow the impact of revisions to
vary in time and across industries.
Table 2.3. Returns around recommendation revisions and past earnings surprises
ARt (1) SU Eq 0.078 (0.079) C on f i r mt -0.003*** (0.000) C ont r ad i c tt -0.002*** (0.001) ConfirmtSUEq 0.805*** (0.115) ContradicttSUEq -0.538*** (0.121) C onst ant 0.007 (0.009) Controls Yes R-squared 0.117 Observations 177,792
Table 2.3 reports results of pooled OLS regressions that estimate equation (2.19) to assess whether investors reinterpret past earnings surprises using recommendation revisions. The left-hand variable is ARt, a firm’s buy-and-hold return in excess of the value-weighted CRSP market index from day −1
to +1 around the date of a recommendation revision (t).
The units of observation are firm-days with recommendation revisions. SU Eqis the last earnings sur-
prise preceding the recommendation revision. C on f i r mtis 1 if at least one of the recommendation
revisions on day t has the same sign as the last preceding earnings surprise, SU Eq, and 0 otherwise. C ont r ad i c tt is 1 if at least one of the recommendation revisions on day t and the last preceding
surprise have opposite signs, and 0 otherwise. Controls include the average change in IBES recom- mendation ranks associated with the recommendation revisions on t , size decile dummies, indicator variables for all 5 IBES recommendation ranks (rounded to the nearest integer in case I average over several recommendations on the same day), dummies for each quarter of the sample period, and dummies for 48 Fama-French industries. All controls are interacted with the change in recommen- dation ranks. For brevity, I again omit the coefficients of the control variables. They are available upon request. Standard errors are clustered by firm and by calendar date of the recommendation revision. ***, **, and * denote significance levels of 0.01, 0.05, and 0.1.
14 Since recommendation ranks are averaged across all recommendations on a given day, I round them to the nearest integer before generating rank-dummies.
Table 2.3 presents the results. I again restrict the presentation to the central coef- ficients.15 In line with the predictions of Hypothesis 2, abnormal returns are pos- itively associated with past surprises around confirmatory revisions (γ4 = 0.805)
and negatively associated around contradictory revisions (γ5= −0.538). Both coeffi-
cients are highly significant (p-values < 0.01).
The magnitudes of the coefficients imply an economically relevant adjustment in investors’ interpretation of past earnings. To see this, consider the predicted ef- fects of confirmatory and contradictory revisions on abnormal returns at the 1 per- cent and 99 percent quantiles of SU Eq (−0.019 and 0.015). For stocks with excep-
tionally positive surprises, the coefficient estimates predict an abnormal return of 1.21 percent around confirmatory revisions and an abnormal return of −0.81 per- cent around contradictory revisions. In contrast, for stocks with exceptionally nega- tive earnings surprises, the coefficients imply an abnormal return of −1.53 percent around confirmatory revisions and an abnormal return of 1.02 percent around con- tradictory revisions.
The results suggest that investors extract information about past earnings sur- prises from recommendation revisions and use this information to alter their own interpretation as predicted by Hypothesis 2. This reaffirms the findings concerning Hypothesis 1 and indicates that part of what revisions reveal about past earnings surprises reflects genuinely new information.