Founders And Financials- Machine Learning Algorithms in Venture Capital

Abstract: The increased usage of machine learning (ML) algorithms in venture capital investment screening poses the question of how different characteristics influence predictions. The purpose of this study is to investigate how founder and financial characteristics influence ML predictions of the raising of more than one round for Swedish startups. A tuned random forest and logistic regression (logit model) are implemented on data for founder and financial characteristics for Swedish startups that have raised a minimum of one funding round. The target variable used for prediction is whether organizations have raised more than one round. SHAP values are used in an analysis of how different characteristics contribute to the ML predictions. For the random forest model, implemented on data from Sweden Tech Ecosystem and Serrano, financial characteristics impact the ML prediction more than founder characteristics. Further, a high share of female founders, a high distance from Stockholm and a lack of prior founder experience negatively contribute to the ML prediction. The importance of financials for the predictions is related to literature on founder replacement over time. The results for the founder characteristics are related to literature on geographical clustering in venture capital, gender bias among venture capitalists and the importance of entrepreneurial experience for future success. A positive contribution of prior founder experience to the prediction of a random forest model, implemented on data retrieved from EQT, is also found. The results inform a discussion on the effects of skewed data on ML predictions. Further, a discussion on how ML algorithms might institutionalize investor biases is conducted. If ML algorithms are not used constructively there is a risk that diversity becomes deprioritized in capital allocation.