Limited Partners versus Unlimited Machines

We assemble a proprietary sample of 395 PPMs to analyze PE fund performance and fundraising success. We analyze hard and soft information contained in these documents using standard econometric methods and machine learning techniques. Our first set of results shows that the performance of private equity funds is unrelated to standard fund characteristics (e.g., past performance, prospectus length and complexity). Meanwhile, measures of fundraising success although correlated to most fund characteristics, are not related to future performance either. Our second set of findings shows that machine learning tools can use soft information to predict future performance. The spread between the funds within the top quartile and the lowest quartile of probability of success is large. The machine learning algorithm is better at selecting the right funds than at avoiding the bad ones.

http://pelaidbare.com/