experienced abnormally high volume. To better distinguish our findings from previous
results, in Table 7.6 (panel C) we sort stocks on both abnormal trading volume and
returns. For our single-dimensional sorts, alphas for our buy-minus-sell portfolios are
negative for volume deciles 9 and 10 and return deciles 1, 2, 9, and 10. Therefore, for
both dimensions, we combine deciles 1 and 2, 3 through 8, and 9 and 10. This gives us a
manageable nine partitions.
Consider stocks with high abnormal trading volume but unremarkable returns
(i.e., panel C, thecolumn for volume deciles 9–10and therows for return deciles 3–8).
For both the portfolios of stocks purchased and stocks sold by individual investors,
there are reliably positive alphas (0.477,t¼1.69, and 0.499,t¼1.76). This is consistent
with Gervais et al.’s finding that the persistent outperformance of stocks with abnor-
mally high trading volume is most pronounced for stocks that did not experience
extreme contemporaneous returns.
Now consider stocks with extreme returns and high trading volume (i.e., panel C, the
column for volume deciles 9–10and therows for return deciles 1–2 and 9–10). Conrad et
al. find that for weekly formation periods, mean reversion is most pronounced for stocks
with high trading volume. Our analysis weakly confirms their finding. For the purchase
portfolio, high-volume high-return stocks experience significant subsequent reversion
( ¼0.833, t¼1.89), but high-volume low-return stocks do not. For the sales
portfolio, the alpha for high-volume high-return stocks is consistent with reversion
but not statistically significant.^26
While our double-sort buy-and-sell portfolio results confirm Gervais et al. and are
roughly consistent with Conrad et al., our double-sort buy-minus-sell portfolio results
confirm the effects of attention-driven buying in high-attention partitions. The high-
attention partitions are the intersections of high abnormal volume (i.e., panel C, the
column for volume deciles 9–10) with extreme return (i.e., panel C, therows for return
deciles 1–2 and 9–10). For these high-attention partitions, we get strong confirmation
that stocks bought by individual investors underperform those sold. For the high-
volume, low-return partition ¼0.705 (t¼3.56) and for the high-volume, high-
return partition ¼0.607 (t¼3.21).
7.7 Conclusion
We propose an alternative model of decision making in which agents faced with many
alternatives consider primarily those alternatives that have attention-attracting
qualities. Preferences come into play only after attention has limited the choice set.
206 News and abnormal returns
(^26) Several factors may account for discrepancies between Conrad et al.’s and our results. We examine a different time period
than they, we focus on stocks heavily trading by individual investors, we sort stocks on previous-day return rather than
contemporaneous weekly returns, and we use a slightly different measure of abnormal trading volume. To further determine
whether short-term mean reversion can explain our return results, we re-estimate abnormal returns (alphas) using a five-factor
model, where the fifth factor is a short-term mean reversion factor. The short-term mean reversion factor is constructed by
calculated weekly return (Wednesday to Wednesday). The long portfolio invests in a value-weighted portfolio of losers (stocks
with negative weekly returns); the short portfolio invests in a value-weighted portfolio of winners. We value-weight the returns
to mitigate the microstructure issues (i.e., bid–ask bounce) that arise when constructing a short-term mean reversion factor.
The holding period for each investment is one week. The daily returns on the long and short portfolios are compounded to
yield a monthly return. The monthly factor is the difference between the monthly returns of the loser and winner portfolio.
Consistent with prior research, this monthly factor is reliably positive (1.50%,t¼9.60). Though the factor loadings on the
short-term mean reversion factor are positive in all but the lowest abnormal volume decile, the high-abnormal volume sort
decile alpha for the five-factor model is0.636 (t¼3.37) and very close to the four-factor alpha of0.690 (t¼3.83) for
that decile reported in Table 7.6 (panel A).