What you will learn
- Interpret core performance metrics and their blind spots.
- Apply robustness checks, including outlier removal and sensitivity tests.
Core report metrics
A backtest report typically shows annualized return, volatility, Sharpe ratio, maximum drawdown, trade count, turnover and exposure. Each describes one aspect, and none is sufficient alone.
The Sharpe ratio divides mean excess return by volatility and treats upside and downside swings alike. The Sortino ratio uses only downside deviation. Maximum drawdown reflects one episode, and comparison with a benchmark shows whether the strategy added anything beyond simple market exposure.
Robustness checks
Vary parameters near the chosen values, split the sample into subperiods, test related markets, raise costs, and delay execution by a bar. A robust result bends gently under these changes; a fragile one breaks.
Also remove the best few trades or the best period and recompute. If profit depends on a handful of outliers, the apparent edge may be a few lucky events rather than a repeatable process.
Honest reporting
Every variant tested raises the chance that the best one looks good by luck. Report how many variants were tried, state the data period, universe, costs and assumptions, and prefer medians across settings to the single best result.
Short samples are noisy. Five annual observations cannot pin down a Sharpe ratio precisely, so treat a backtest as a hypothesis check under stated assumptions, not a forecast of future performance.
Worked example: A Sharpe ratio from five years
Illustrative annual excess returns are 12%, 4%, -2%, 10% and 6%. The mean is 6.0% and the squared deviations sum to 120, so the sample standard deviation is √(120 ÷ 4) = 5.48% and Sharpe is 6.0 ÷ 5.48 ≈ 1.1. Dropping the best year leaves a mean of 4.5% and squared deviations of 75, so the deviation is √(75 ÷ 3) = 5.0% and Sharpe is 0.9.
Check your understanding
A report shows 40,000 of total profit over 100 trades, but the five best trades earned 45,000 in total. What is the result without those five, and what does it suggest?
Show answer and explanation
Without the best five, the other 95 trades produced 40,000 - 45,000 = -5,000, a net loss. All of the profit came from five trades, 5% of the sample, so evidence of a repeatable edge is weak. A larger sample or a different market would be needed first.