How Likely is the Investment of a Venture Capital Firm in a Latin American Country? Machine Learning Models Based on Experience, Distances and Network Features

How Likely is the Investment of a Venture Capital Firm in a Latin American Country? Machine Learning Models Based on Experience, Distances and Network Features

• Author
• Marcelo Guzella
• Co-authors
• Felipe Buchbinder
• Abstract

• The ability to accurately estimate whether a venture capital firm will or will not invest in a market benefits investors, companies and investment promotion agencies. In this analysis, we applied machine learning techniques to predict whether a VC firm will invest in a specific country in Latin America. Predictors included firm characteristics and past behavior, country macroeconomic situation, network-level variables, and institutional and geographic distances between the target market and the country where the firm is headquartered. The database encompasses more than 10 thousand funding rounds from 2002 to 2020. The gradient boosting algorithm presented the best predicting performance, in terms of the area under the curve that relates true positive with false positive rates. A classification using this technique has a precision 24 percentage points higher than the one of naively selecting firms that invested in the recent past, which translates into savings or gains in capital allocation or investment promotion processes.

• Keywords
• venture capital, machine learning, national distances, syndicated investments
• Subject Area
• Asset pricing, investments, and Derivatives

Back Download