Human Interactions and Financial Investment: A Video-Based Approach

Abstract

We quantify human interactions in three-V dimensions---visual, vocal, and verbal, using machine learning (ML)-based algorithms with videos as data input. We apply the method to videos of startups pitching investors for funding. We find that venture investors are more likely to invest in startups showing more positivity (i.e., happy, warm, passionate), even though those startups underperform conditional on funding. Investors do not seem to correctly predict startup quality using interactions. Instead, we document interaction-induced biases---using video analysis and an experiment, we show that interaction-induced biases can be explained by a taste-based channel (18 percent) and inaccurate beliefs (82 percent).

Keywords: Behavioral Economics, Social Interactions, Video Data, Machine Learning, Entrepreneurship

JEL Classification: D91, G41, C55, G24