Characterizing heteroskedasticity
研究了金融时间序列中波动率聚集（异方差性）的滞后相关性衰减形状，发现对数衰减优于指数和幂律衰减，并分析了新兴市场时间序列的记忆较短。
Volatility clustering, or heteroskedasticity, is an important feature of all financial time series. In particular, the lagged correlation for the volatility is slowly decreasing with increasing lags. This paper characterizes its decay. First, Monte Carlo simulations are used to select the best volatility and correlation estimators for this task. Second, the empirical lagged correlations are studied over a set of 225 daily time series, and for the DJIA with a sample size of one century. The results strongly favor a log-decay shape, while an exponential and power law decay do not describe the data well. The implications for the description of financial time series by processes are important, as these findings exclude hyperbolic decay, but favor volatility cascade and multi-component ARCH processes. Third, the analysis of the decay coefficient shows that time series related to emerging countries have a shorter memory, in agreement with an analysis of the Hurst exponents published recently.