Modeling Private Equity: A Machine-Learning Approach

Abstract

We introduce an estimator that measures time-varying factor exposures of individual private equity funds, with minimal assumptions about the fund return DGP. Simulations using varying assumptions about the autoregressive and moving-average properties of the DGP show that our estimator exhibits less bias and variance (thus, lower MSE) than competing estimators. Applying our model to a newly available commercial dataset,
PitchBook, we undercover new findings of economic importance: buyout managers have higher average skill levels than claimed by past studies; portfolios are marked with forward looking and lagged multiples of factors; and skill and systematic exposures vary significantly over time.

Keywords: Private Equity, Buyout Funds, Time-Series, Supervised Machine Learning, Jackknife, Cross-Validation, Factor Modeling, Long Horizons, Overlapping Observations

JEL Classification: C01, C22, G12, G17