Score contribution per author:
α: calibrated so average coauthorship-adjusted count equals average raw count
This paper proposes a new empirical testing method for detecting herding in stock markets. The traditional regression approach is extended to a vector autoregressive framework, in which the predictive power of squared index returns for the cross-sectional dispersion of equity returns is tested using a Granger causality test. Macroeconomic news announcements and the aggregate number of firm-level news items are treated as conditioning variables, while the average sentiment of firm-level news is treated as jointly determined. The testing algorithm allows the change points in the causal relationships between the cross-sectional dispersion of returns and squared index returns to be determined endogenously rather than being chosen arbitrarily a priori. Evidence of herding is detected in the constituent stocks of the Dow Jones Industrial Average at the onset of the subprime mortgage crisis, during the European debt and the U.S. debt-ceiling crises and the Chinese stock market crash of 2015. These results contrast with those obtained from the traditional methods where little evidence of herding is found in the US stock market.