Powers Correlation Analysis of Returns with a Non-stationary Zero-Process
研究发现，当零收益概率随时间变化时，经典的幂相关分析会错误判断波动率的长期记忆性；本文提出了对零收益概率变化和无条件方差非恒定均稳健的新诊断工具，并通过对智利股市和Facebook高频数据的分析，表明许多情况下的波动效应仅为短期。
Abstract The higher order dynamics of individual stocks is investigated. We show that classical powers correlation analysis can lead to a spurious assessment of the volatility persistence or long memory volatility effects, if the zero return probability is non-constant over time. In other words, classical tools are not able to distinguish between long-run volatility effects, such as IGARCH, and the case where the zero returns are not evenly distributed over time. As a remedy, new diagnostic tools are proposed that are robust to changes in the zero return probability. Since a time-varying zero return probability could potentially be accompanied by a non-constant unconditional variance, we also develop powers correlation analysis that is robust in such a case. In addition, the diagnostic tools we propose offer a rigorous analysis of the short-run volatility effects, while the use of the classical powers correlations lead to doubtful conclusions. Monte Carlo experiments, and the study of the absolute value correlation of daily returns taken from the Chilean financial market and the 1-min returns of Facebook stocks, suggest that the volatility effects are only short-run in many cases.