In the rolling backtest, is there any overlap between an optimization interval and the subsequent interval during which the optimized weights are used to calculate performance? I'm a bit concerned, because total returns are just TOO good for re-optimization done every data interval. It almost seems that the optimization process uses the next period's data. I can furnish an example showing almost 60% annualized return over the past 7 years using an ETF portfolio containing only SPY, EFA, IEF, TLT, DBC and VNQ with weekly optimization. I know trading costs are not included, but is the ideal result too good to be true?
