This study aims to propose a new perspective on identifying herd behavior at the market-wide level by using panel data analysis on 56 countries around the globe.
Traditional herding detection models at the market-wide level are based on time series analysis of return variations and market returns to capture trading behavior within an individual country. The authors augment four widely used herding detection models, those are the models of Chang et al. (2000), Yao et al. (2014) and Bui et al. (2017), by using panel data analyses with fixed effects for the entire 56 countries.
The authors confirm the existence of herd behavior in all 56 countries and within 33 developing markets. Interestingly, anti-herd behavior is present among 23 developed markets. Herd behavior is more dominant during down and crisis periods, whereas anti-herd behavior disappears. Severe herd behavior is the most prominent during the crisis period. In addition, the results demonstrate economic significance. An increase of one standard deviation of the return squared (as a proxy for herding) decreases return dispersion by 0.0466, accounting for 6.8544% per annum relative to the cross-sectional absolute deviation. The economic significance is more amplified to 17.9172% per annum and 1,590.15% per annum over the negative market return and the financial turmoil periods, respectively. All of these support herd behavior in global equity markets.
Investors and fund managers could design profitable investment strategies based on these premises. Policymakers and regulators could monitor investment performance and improve market efficiency.
First, the authors propose a different approach to investigating herd behavior by using panel data on international equity markets rather than focusing on time series data from a single country, as prior research has done. This fills the gap left by prior mixed findings on the existence of herd behavior at the individual country level. Second, the authors compare the efficacy of the aforementioned four well-known herding detection models found in empirical research and find that each model possesses relatively similar power to detect herd behavior in international equity markets.
