Portfolio Diversification Dynamics as a Measure of Market Sentiment

Abstract

We build an original market sentiment index based solely on changes over time in the number of different stocks held by individual investors. No prices, returns or trading volumes enter the definition. As a consequence, our index is not contaminated by liquidity concerns present in measures based on buy-sell imbalances. Using the trades and portfolios of a large sample of ninety thousand investors over a eight-year period, we show that our index outperforms other usual indices (based on surveys, macro-economic variables or buy-sell imbalances) in predicting returns on long-short size-based portfolios. Moreover, a simple dynamic strategy driven by our sentiment index delivers a Sharpe ratio higher than that of ninety nine percent of random dynamic strategies and much higher than the Sharpe ratio of a buy-and-hold strategy. The predictive performance of our market sentiment index survives several robustness checks, including restrictions on the minimum portfolio value of individual investors and constraints on the number of states of the Markov chain used to model the time-series of diversification levels.