15 ms·
Using abnormal returns to train a classifier, as suggested by olooney, is not appropriate in this study for two reasons. First, there is not enough observatio
by mazimi1 7y ago
Using abnormal returns to train a classifier, as suggested by
olooney, is not appropriate in this study for two reasons. First, there is not enough observations. The classifier labels sentences not the entire report. Each report contains more than 1000 sentences on average and there are 60,000 reports and 60,000 returns.
Second, the purpose of the study is to see if textual content (in this case sentiment)is related to return. Using returns to train a classifier would assume that sentiment is related to return, which is the question that paper tries to answer.
Also, the paper does control for the quantitative information in corporate reports. All the control variables in the regressions are from the reports.