Abstract: The structural changes which may result in spurious long memory always occur in crude oil market, and then they usually lead to the biased estimation of parameters. As a result, the features of structural changes and long memory have become the key to the rational modeling and accurate forecasting of crude oil price volatility. However, the existing models often only consider a certain factor, or only consider the long memory or short memory in the volatility, which may lead to the inaccurate forecasting of crude oil price volatility. In this situation, this paper aims to investigate whether the volatility forecasting models considering structural change and long memory have better forecasting performance on crude oil price volatility than traditional models, and whether the mixed memory GARCH model considering different memory and volatility level appears effective in depicting the characteristics of structural changes and long memory in crude oil price volatility. Therefore, both of the characteristics are focused on, and the GARCH-type models incorporating structural break points and the MMGARCH model are used to estimate and forecast crude oil price volatility. The empirical results prove the existence of structural change and long memory characteristics in crude oil market volatility, and indicate that the models which incorporate the two characteristics usually yield superior fitting and forecasting performance to standard GARCH-type models. In particular, the MMGARCH model outperforms other competitive models on forecasting crude oil price volatility, which indicates that the MMGARCH model can dynamically depict the volatility level and memory of the process, and then capture the structural changes and long memory simultaneously. Therefore, the MMGARCH model can be considered a helpful alternative to make accurate crude oil price volatility forecasting.

Key words: crude oil market, volatility forecasting, structural changes, long memory