How to read a backtest without fooling yourself

A backtest does not tell you what your strategy is going to return. It tells you what it would have returned, in a world slightly simpler than the real one. The gap between the two is where the money goes.

1. Are the costs counted?

A strategy that takes many short-duration positions can be profitable gross and losing net. Every round trip costs: the commission, but above all the gap between the theoretical price and the price actually obtained.

On a test that chains trades together, that friction changes the conclusion, not the decimal. If the tool does not ask you for your fees and your slippage before running the calculation, the result it shows is not usable.

2. How many rules did you try?

This is the question that hurts. If you test thirty parameter combinations and keep the best one, you have not found a strategy: you have found the noise that most resembles a strategy over that particular period.

The symptom is recognisable. The curve is beautiful, but it collapses as soon as you shift a threshold slightly: RSI below 10 works, below 12 no longer does. A robust rule survives a small move of its bounds. An overfitted rule does not.

The counter-test is simple: does the strategy hold over a period you did not use to build it? If you optimised on 2015 to 2021, look at 2022 to 2025 without touching anything.

3. What does the average result hide?

A positive average gain can hide anything. Two numbers matter more than it does:

• The worst drawdown. How far did the capital fall between its high and the trough that followed? It is the only number that predicts whether you will stick with the strategy. A drawdown you cannot bear will be abandoned at its worst moment, which is to say at the worst possible moment.
• The longest losing streak. How many losing trades in a row does the test contain? Multiply it by two: that is what you must be ready to absorb without changing method.

4. Is the universe tested an honest one?

A classic trap: testing a strategy on the stocks that exist today. The ones that went bankrupt or were delisted have vanished from the sample, and with them a good share of the losses. The test runs on a list of survivors, which flatters everything.

The second trap is quieter: the selection criteria. Filtering on a current capitalisation or volume amounts to knowing in 2016 which companies would be large in 2026. A filter has to apply with the data known at the time of the trade, not with today's.

What a good backtest actually allows

It does not predict. It eliminates. Its real value is to stop you from seriously trading an idea that did not hold, and that is well worth the time spent reading it properly.

Past performance is no guide to future results. That is not a legal formula for the bottom of the page: it is the only right way to read a backtest.