2.4 Relation between presentation and discussion period lengths
2.4.2 The effect of abnormal presentation length on discussion length
0 1
DISC PRES
LENGTH E E ABLNGTH H (2)
We expect the coefficient on E1 to be negative – hence managers present relatively little information in the presentation, analysts will probe more in the discussion.
Note, we do not include any of the determinants of presentation length in this analysis because the residual from the first-stage regression, ABLNGTHPRES, is by definition orthogonal to these other variables and including them in the regression has no effect on E1.
Our second hypothesis posits a stronger relation between ABLNGTHPRES and LENGTHDISCin situations where managers provide abnormally low levels of information relative to situations where managers provide abnormally high levels of information because it is possible that managers are providing a high volume of wrong information. To test this hypothesis, we allow the coefficient on ABLNGTHPRES to vary for positive and negative values. Specifically, we define a dummy variable (POS) that is equal to one if ABLNGTHPRES is positive and zero otherwise. We then interact this variable with ABLNGTHPRES:
H E
E E
E PRES u PRES
DISC ABLNGTH POS POS ABLNGTH
LENGTH 0 1 2 3 (3)
We expect the coefficient onE3to be positive if abnormally long presentations do not lead to shorter discussion periods as much as abnormally short presentations lead to longer discussions.
Hypothesis three predicts that the relation between abnormal presentation length and discussion length is stronger for firms with greater analyst following. We define a dummy variable (HI_ANAL) equal to 1 if analyst following (AF) is above the median for the entire sample (AFt 6) and zero otherwise. We then this variable with our variables from model (3):
0 1 2 3
LENGTH ABLNGTH POS POS ABLNGTH
HI ANAL HI ANAL ABLNGTH HI ANAL POS
HI ANAL POS ABLNGTH
If firms with greater analyst following are more likely to probe managers when they fail to disclose sufficient information in the presentation portion of the call, then the coefficient on E5 should be negative. For equations two, three and four, we winsorise the top and bottom 1% to reduce the impact of outliers on our inferences and, as before, we include firm-fixed effects.
The results of equation two, reported in the first set of columns of Table 2.4, indicates support for hypothesis one. Discussion periods are longer, on average, when managers’
presentations are abnormally shorter (t-statistic = –2.53). The effect, however, is not substantial. An increase in presentation length of 745 words (the average within-firm inter-quartile range for presentation length) results in a decrease in discussion length of 46 words. However, part of the reason for this low effect is the fact that positive residuals — indicating abnormally long presentations — do not necessarily result in shorter discussion periods (i.e., the relation is not symmetric).
The results of equation three, presented in the next set of columns, demonstrate this effect.
The coefficient on ABLNGTHPRES in this equation represents the relation between abnormal presentation length and discussion length for firms with abnormally short calls and the coefficient here is –0.4094 (t-statistic = 6.66). A reduction of 745 words in the presentation would result in an increase of 305 words in the discussion. The coefficient on the interaction term, POS×ABLNGTHPRES, represents the incremental effect for positive residuals. The coefficient is significantly positive (t-stat = 6.60) and, in fact, an F-test on the sum of the two coefficients (ABLNGTHPRESand POS×ABLNGTHPRES) indicates that the relation between abnormal presentation length and discussion length is statistically greater than zero for the positive residual group (F-stat = 11.26, p = 0.0008). It appears that abnormally long presentations actually lead to longer discussion periods. However, the magnitude of the effect is significantly less than it is for the negative residual group (the sum of the coefficients = 0.1886 vs. –0.4094 for the negative residual group).
Nevertheless, it appears that when managers present for a longer duration than expected, they are not necessarily presenting information that analysts want.34
Finally, the results of equation four are presented in the last three columns of Table 2.4.
The results are consistent with hypothesis three. The coefficient on HI_ANAL×ABLNGTHPRES is significantly negative (t-stat = 2.64), indicating that the negative relation between abnormal presentation length and discussion length is stronger for firms with high analyst following.
As an alternative specification, we also ran equations two, three and four using the number of questions asked by analysts (#QUESTION) as our dependent variable (untabulated).
The results are inferentially similar (all variables of interest are significant at similar probability levels). The one exception is that the F-test on the sum of the coefficients ABLNGTHPRESand POS×ABLNGTHPRESis not significant, indicating there is no relation between discussion length and abnormal presentation length when presentations are abnormally long (vs. the positive relation reported using LENGTHDISC as the dependent variable). However, the coefficient on POS×ABLNGTHPRESis still significantly positive, consistent with hypothesis two.
Overall, the evidence is consistent with analysts playing a significant role in determining the information that is disclosed during conference calls and hence, in the incremental informative nature of conference calls. Moreover, the results suggest that managers (on average) respond to analysts’ inquiries by providing more information, rather than refusing to answer their questions. These conclusions, however, assume that longer calls are a reasonable proxy for more information — an issue we address in the next section.
34This result also addresses the concern that our finding is the result of some conference calls being constrained to a certain time limit (i.e. when managers’ presentations are abnormally long, discussion periods will, by necessity, be cut short due to time constraints). It does not appear to be the case that abnormally long presentations lead to short discussion periods. Rather, the overall negative relation between abnormal presentation length and discussion length is driven by abnormally short presentations being followed by longer discussions (which is less likely the result of time limits). As additional evidence that our findings are not the result of time constraints, we also regressed discussion length on raw presentation length (not abnormal presentation length) and we do not find a statistically significant negative coefficient on presentation length (results untabulated). In other words, discussion lengths are longer not just when presentation are shorter but when presentations are unexpectedly shorter given the circumstances of the firm (e.g., the firms reports bad news, surprising earnings, special items, etc.).