A new approach to detect long memory by fractional integration or short memory by structural break

Abstract

Long memory in test statistics can either originate from fractional integration or be spuriously induced by a short memory process with a structural break. This research estimated and detected long memory from the two causes by simulations and empirical analysis. The simulation results showed that fractional integration and structural break processes could demonstrate long memory properties. The 2ELW estimator was stable for fractional integration but not stable for time series with structural breaks. The modified W statistic based on 2ELW was efficient in discriminating fractional integration and structural breaks. Moreover, we found that six volatility time series of stock indexes and individual stocks in the Chinese market experience significant long memory and structural breaks, and the fractional differencing parameter is overestimated without controlling structural breaks.

Keywords

• long memory
• fractional integration
• structural break
• volatility
• stock market