A Tale of Two Zoos: Machine Learning Insights on Retail Investors

This paper employs various machine learning models to analyze how stock characteristics (the "factor zoo") and behavioral biases (the "bias zoo") affect the returns of millions of retail investors in India. We observe that Neural Networks outperform other machine learning and OLS models in uniquely predicting both good and bad out-of-sample performance. Moreover, behavioral biases exert a more significant influence than holding-weighted firm characteristics. Among all predictors, (under)diversification, portfolio turnover, and momentum are the leading factors to influence overall retail returns. Additionally, turnover, the disposition effect, and diversification emerge as the three most important factors in predicting the returns for newly initiated trading.