How to read a backtest report without fooling yourself

Every backtest looks great. That is the point — whoever published it chose a period, a universe, and a set of parameters that make it look great. Your job as a reader is to figure out what they did not tell you, and whether those omissions change the answer.

The six numbers you need

1. 1Sharpe ratio — but only alongside the sample size. Sharpe 2.0 on 6 months of data is noise; Sharpe 1.2 on 10 years is a real strategy.
2. 2Max drawdown — the emotional test. Can you actually sit through a -20% drawdown for 8 months? Most people say yes and act otherwise.
3. 3Win rate — a sanity check. Strategies with < 45% win rate are mostly tail-risk trades; > 65% usually means you are picking up pennies in front of a truck.
4. 4Trade count — too few and your Sharpe is noisy, too many and your costs will eat the edge. 100-500 trades/year is the sweet spot for most retail-scale strategies.
5. 5Turnover — the hidden cost multiplier. 600% annualised turnover at 5bp round-trip is 30% of your gross return gone.
6. 6Days in market — if the strategy is flat 80% of the time, you need to compare it to cash+risk-premium, not to buy-and-hold.

What they are probably hiding

Three common omissions kill most "great" backtests when you trade them live. Survivorship bias — the index you tested includes only companies that exist today. Look-ahead bias — some fundamentals data was not actually available on the date your backtest used it. Cost underestimation — quoted spreads are not fill prices, especially for small caps or off-hours.

A worked example

Let us apply this to a low-turnover strategy — the Dividend Aristocrats template below. It only rebalances annually, so cost sensitivity is low. Sharpe 1.05 over 1080 days is credible. Max drawdown -7.9% is actually emotionally tradeable. Win rate 58.8% and only 24 trades total means it is a genuine slow buy-and-hold quality sleeve, not a high-frequency illusion.

The question to always ask

"Under what regime would this strategy have performed poorly, and is that regime more or less likely over the next 3 years?" Most backtests implicitly assume the next 3 years look like the last 5. If you can articulate a plausible regime where the logic breaks, discount the headline Sharpe accordingly.