In hindsight, exceptional backtest performance is never that exceptional. A client once told me that he had never seen a backtest that didn’t work. He joked that the only people who can consistently generate Sharpe ratios of 3.0 or above were quants running backtests, and Bernie Madoff. Even when there’s no data mining, i.e. no explicit look-ahead bias, researchers benefit from years of published research on what works and what doesn’t, which is an implicit look-ahead bias.
Backtests can be useful, but use them carefully.
