Can strategy text
predict fund returns?
A walkthrough of the private-equity study and a practical guide to testing its approach with XGBoost.
01
The study
Limited Partners versus Unlimited Machines: Artificial Intelligence and the Performance of Private Equity Funds.
395fund prospectuses in the final sample
72funds in the 2014–2016 test cohort
2,774words in the average strategy section
The central question
The authors ask whether a fund’s written investment strategy helps predict its later returns, beyond what investors can learn from its track record and basic characteristics.
They find that strategy text helps rank funds by their likelihood of beating a peer benchmark. To test that result with XGBoost, we need to reproduce how the authors assembled the documents, defined success, and separated training funds from test funds.