Analysis of Herding Behavior in Individual Investor Portfolios Using Machine Learning Algorithms

This paper examines the determinants of herding at both stock and individual investor levels and studies the portfolio performance of herd vs. non-herd portfolios using machine learning algorithms. The disposition effect and the attention effect seem to explain herding behavior at the stock level. At the individual investor level, the cumulative number of buys and portfolio values reduce the prediction of herding behavior, while high values of portfolio return lead to a small increase in herding. Individuals who herd do not outperform either market or non-herd portfolios, suggesting that herding is a behavioral bias. Thus, such behavior seems to destabilize stock markets, creating temporary discrepancies in stock prices followed by reversals back to fundamentals. The most predictive factor in the performance tests of individual portfolios is the market risk premium and using equally-weighted factors rather than value-weighted factors seems to provide more consistent results in the portfolio performance analyses