Daily Speculations 

The Chairman

7/8/05
Victor Niederhoffer reviews
Why Do Absolute Returns Predict Volatility So Well by
Lars Forsberg and Eric Ghysels
This paper concludes that absolute variation is a much better predictor of future variation in returns for stocks and also shows more persistence. A 20 year period encompassing all intervals from 30 minutes to two weeks is considered. The superior properties of absolute variation relative to standard deviation depend on its sampling distribution's being a function of the squares of returns rather than the fourth powers of returns, and on the absolute returns' invariance with respect to jumps in the process. In sample and out of sample data confirm these conclusions. The paper confirms the lack of ill wisdom of the emphasis I've placed on absolute variations for the last 30 years, with particular reference to the percentage of variation caught with an indicator, and the successive reductions in absolute variations that adding additional variables to a model versus the total variation about the mean creates. A discussion of how to compute R^{2} with absolute variation for commonly used predictors will be imminent.