Feeds:
Posts
Ding-Granger-Engle 1993 (ding-granger-engle-1993) had shown a log-linear model that we apply to around 2000 stocks and show that the fits to actual volatility autocorrelation is good with average $R^2 > 9.8$.  Using this model we estimate the number of days before autocorrelations drop below $\exp(-2.0) \sim 0.134$ and find the average to be above 1800 days.  This is a fairly direct test of universality for long memory in volatility.