Yury,

I agree a confidence interval for Sharpe ratio is better than a single number. But how should we estimate the confidence interval? Do we calculate a Sharpe ratio on a trade by trade basis in order to generate a random sample from which we calculate the confidence interval? But trades don't have the same duration. So should we instead estimate the Sharpe ratio of a strategy on periods of fixed length, say periods of two months, and therefore generate a new sample (smaller than the first)? Should we use periods as long as a year? I guess there is a tradeoff here. We want the sample to be large enough, but the Sharpe ratio estimates are better over longer periods. Unless we annualize the two month estimates. Can we do that?

I have another question: how should we take into account the time when a strategy is out of the market in our Sharpe ratio estimation? Should we consider the periods when a strategy is out of the market as periods with zero returns and include the periods in our calculation or should we simply forget the idle periods and instead use proper multiplicator when annualizing the Sharpe ratio?

Thanks!