Computed on each out-of-sample test period and then average them? Just don't understand why I got unreasonably high Sharpe ratio after the Walk-Forward optimization?