We demonstrate the value of machine learning in accounting through a detailed examination of litigation risk, an important and frequently used estimate in the literature. We evaluate a comprehensive set of twelve machine learning techniques and benchmark their performance against the logistic regression models in Kim and Skinner (2012). These models improve the prediction of litigation risk, with hourglass-shaped and convolutional neural networks the most effective. The improvements are substantial, and are driven by increased precision, the most salient attribute of litigation estimates in the accounting literature. We also produce firm-year litigation risk estimates for use in future research from a convolutional neural network model that uses recursive feature elimination on a pool of 68 possible parameters. Overall, our results suggest that the joint consideration of economically-meaningful predictors and machine learning techniques maximize the effectiveness of accounting estimates.
<aside> <img src="/icons/movie-camera_gray.svg" alt="/icons/movie-camera_gray.svg" width="40px" /> Media: Harvard Law School Forum on Corporate Governance
</aside>
<aside> <img src="/icons/groups_gray.svg" alt="/icons/groups_gray.svg" width="40px" /> Coauthored with Gene Moo Lee Jim Naughton Xin Zheng
</aside>
<aside> <img src="/icons/save_gray.svg" alt="/icons/save_gray.svg" width="40px" /> Download: WP
</aside>