Backtesting a Strategy: What to Actually Look For

Backtesting a strategy is a necessary step, but most beginners approach it wrongly, judging it only by the final profit. In reality a backtest calls for looking at something else entirely. Here is what to actually check in historical testing and which traps to avoid.

Why a backtest matters and its main trap

A backtest checks how a strategy would have behaved on historical data, giving a sense of its potential edge. But it has one main trap, overfitting: a system is easily fitted to a specific stretch of history so that it shows a brilliant result, having memorized random noise instead of real patterns, and then fails on the real market. So the goal of a backtest is not to get a pretty curve on the past but to assess whether the strategy has a real, durable edge. Looking only at the final profit means falling into the trap: a high profit on history guarantees nothing if it is the result of fitting.

What to look for beyond profit

In a backtest you should look not at the final profit but at the quality and durability of the result, through a set of metrics. Maximum drawdown: how deeply the account sank, which shows the real risk and the survivability of the system (a high profit with a monstrous drawdown is untradeable). The shape of the equity curve: smooth growth is better than a ragged one that hangs on a few lucky trades. The ratio of return to drawdown: how much profit per unit of risk. Expectancy and its components (win rate, average win and loss): whether there is a real statistical edge. Sample size: whether there are enough trades for the result to be significant rather than random. The maximum losing streak: what to prepare for psychologically. These metrics matter more than the final profit, because they show by what path and with what risk the result was obtained.

Traps of backtesting

Beyond overfitting, a backtest hides other traps. Ignoring costs: a test without realistic spread and slippage overstates the result, so a strategy profitable on paper can lose on the real market. A small sample: a result on a few dozen trades is random and insignificant. Look-ahead bias: accidentally using data in the test that would not yet have existed in reality. Survivorship bias and cherry-picking a convenient period: testing only on a stretch where the system works. Perfect execution: assuming every trade fills exactly at the price. All these traps make a backtest deceptively good. An honest backtest builds in realistic costs, a sufficient sample, and no peeking into the future.

How to test correctly

Correct historical testing is built on honesty and an assessment of robustness. Build in realistic costs (spread, slippage). Use a sufficient sample (dozens to hundreds of trades). Test on out-of-sample data and, better, walk-forward, on stretches the system did not see during tuning. Assess sensitivity to parameter changes (a robust system does not fall apart from small changes). Look at a set of metrics (drawdown, curve shape, expectancy, sample size), not just profit. Prefer simple, logical systems to complex fitted ones. Remember that a backtest is not a guarantee of the future but an assessment of the probable edge, and its goal is to weed out fitting and confirm a real durable pattern, not to decorate the past.

The practical takeaway

A backtest assesses a strategy's potential edge, but its main trap is overfitting: a system is easily fitted to the past, memorizing noise instead of patterns, so looking only at the final profit is a mistake (a high profit on history guarantees nothing if it is the result of fitting). Look in a backtest not at profit but at quality and durability: maximum drawdown (real risk and survivability), the shape of the equity curve (smooth is better than ragged), the ratio of return to drawdown, expectancy and its components (a real edge), sample size (significance), and the maximum losing streak (psychological readiness). Avoid the traps: ignoring costs (a test without spread and slippage is overstated), a small sample (a random result), look-ahead bias, cherry-picking a convenient period, and perfect execution. Test correctly: realistic costs, a sufficient sample, out-of-sample and walk-forward, sensitivity assessment, a set of metrics, and a preference for simple logical systems. Understanding that a backtest should reveal a durable edge and assess risk rather than final profit, and that the goal is to weed out fitting rather than decorate the past, helps you test strategies honestly and avoid building your trading on an illusory edge.

This material is for educational purposes and is not individual investment advice.