Can machine learning methods predict beta?

It is common to estimate equity betas for private firms or non-traded assets through a comparable company analysis (CCA) of peer firms. Previous literature has questioned the accuracy of those estimates. We test if Machine Learning (ML) algorithms can provide superior forecasts. In out-of-sample tests from 1990 to 2021, we find that ML predictions reduce the mean absolute error by over 42% relative to the CCA. The improved accuracy of ML is most pronounced for smaller, younger firms with different capital structure from their peer group, suggesting potential large improvements are possible by applying ML methods to private firm valuation.