Backtesting Without Lying to Yourself

The lies your backtest tells

1. Look-ahead bias — using information that wasn't available at the time.
2. Survivorship bias — the universe excludes delisted companies.
3. Overfitting — too many parameters tuned to noise.
4. Ignoring costs — slippage, commissions, and borrow fees are real.
5. Regime blindness — fitting one bull market.
6. Data-snooping — running 200 variants and reporting the best.
7. Optimistic execution — assuming you fill at the close.

A defensible workflow

Split your data into train / validation / test. Tune only on train. Select on validation. Report only once, on test. Walk-forward analysis is the gold standard.