Abstract: High-frequency financial data analysis has received more and more attention. Stationary of continuously rising and falling returns and durations from one-minute intraday high frequency SSE Composite Index and SZSE Component Index was analyzed and their distributions were fitted by exponential distribution, Gamma distribution and Weibull distribution with bad results. Influence factors of continuously rising and falling returns were studied based on quantile Granger causality test. The findings show that the possibility of a big rise followed by a big fall is high, but continuously rising extreme return is not affected by previous continuously falling return. The longer the durations of continuously rising or falling returns, the smaller the risk of continuously falling return is. The longer the duration of the last continuously rising process, the lower the extreme return of the next continuously rising process. Continuously falling duration has no effect on previous continuously rising extreme return. Finally, the prediction of conditional VaR for continuously falling return shows that the quantile regression model has good power to predict conditional VaR.

Key words: high frequency data, continuously rising (falling) return, quantile Granger causality testing, conditional VaR