How Informative is the Text of Securities Complaints?

Using the complaints from several thousand securities class actions filed from 1996 to 2019, this paper uses text analysis and machine learning to predict whether those lawsuits will settle or be dismissed. The strongest performing models, some of which incorporate non-text features, are able to predict the outcome in these cases at a rate of about 70 percent, which is a substantial improvement over baseline settlement rates. The models are also able to generate a probability that each case will settle and these estimates provide strong indications of future equity returns. A portfolio that goes long on the firms with consolidated cases that are most likely to be dismissed and short on those consolidated cases that are most likely to settle produces abnormal returns of over three percent in the ten-day window that follows the filing of the complaint. Beyond contributing to the asset pricing literature, the findings have implications for several other areas of research. That it is easier to predict the outcomes of first-filed complaints relative to consolidated complaints provides empirical support for the notion that there is still something of a race to the courthouse in securities litigation. In addition, the predictive ability of the text in complaints suggests that variables built on these measures may help to control for case quality in studies of business litigation. Finally, while these models perform reasonably well, there is substantial room for improvement. This observation implies that, at least for the time being, predictive analytics should act as a complement to, rather than substitute for, human legal judgment.

If you wish to attend, please register before October 21 by sending an email to: acle@uva.nl

Click here for a one-on-one meeting with Professor Badawi: link

For more information, visit ACLE webpage.