We propose using the U-statistic as a new measure of market integration and construct the null hypothesis to test herding. We derive asymptotic distribution of the U-statistics under four different market conditions and examine herding in NYSE and NASDAQ. Not based on a specific theoretical asset pricing model, we define the herding as a severe dynamic comovement of stocks which is a function of mean market return. Our empirical results show that there exists more frequent herding in NYSE than in NASDAQ while NASDAQ is more integrated than NYSE during tranquil period, whereas different results are observed during crisis periods.​

Keywords: overall market herding; network structure approach; U-statistic