Machine Learning About Venture Capital Choices

Abstract

We study early-stage venture capitalists' (VCs) decisions through the lens of a predictive model of venture success. Using French administrative data on VC-backed and non-VC-backed companies, we find that VCs invest in some companies that perform predictably poorly and pass on others that perform predictably well. VCs tend to select entrepreneurs whose features are representative of success – such as being male, graduates of elite schools, and based in Paris. Although entrepreneurs with these characteristics exhibit higher success rates, VCs exaggerate the importance of these features relative to their impact on performance, contributing to the narrowness of the VC industry.

Keywords: venture capital, machine learning, entrepreneurship, capital allocation, stereotypes, representativeness

JEL Classification: G11, G24,G41, M13, D83, D8