wilmott.com

Given a time series (of a thinly traded futures market) fails the JB test miserably, can one take an alternative distibution? Or is the JB test too tight in an empirical framwork, and if so how much kurtosis is deemed to be acceptable?The aim is to examine the return distribution for the roll and have a statistical confidence of its cost. The problem relates to extending the curve by calculating the cumulative returns lost in an appropriate hedge.Thanks

It doesn't make much sense to speak of failing a Jarque-Bera test. Asset price return histories are almost never completely Normal. For a confidence interval on roll cost, non-Normality of the underlying asset returns is the least of your problems. At most, I would make an adjustment at the end of the analysis. I'd worry a lot more about the spread between the near-expiry price and the roll price, and its relation to other variables.

Thanks Aaron