How to tell whether a trading strategy actually works
Updated November 9, 2026 · SignalPoint
A strategy that made money is not the same as a strategy that works. Over a few dozen trades, luck alone can produce a convincing profit, and luck can just as easily produce a loss from a sound idea. The question to ask of a trade history is not "did it make money?" but "how sure can I be that the average trade is better than zero?"
Here are four checks that answer that honestly. None of them need anything more than the list of trades your strategy actually made.
1. Put an interval around the average trade
Take the net return of every closed trade (after fees) and compute the average. Then ask how much that average could move if you had been dealt a different, equally likely set of trades.
A simple way is the bootstrap: resample your trades with replacement thousands of times, recompute the average each time, and look at the middle 90% of those averages. If that 90% interval includes zero, your data cannot distinguish the strategy from no edge at all, however good the total looks.
Example: 150 trades averaging +0.3% with a typical trade swing of ±4% gives a 90% interval of roughly −0.24% to +0.84%. The profit is real money, but the evidence for an edge is weak.
2. Check whether you have enough trades
The width of that interval shrinks only with the square root of the number of trades. Quadrupling your sample halves the uncertainty. A useful question is: what is the smallest average trade this sample could reliably detect? If your trades swing ±4% and you have 100 of them, an edge smaller than about 1% per trade is essentially invisible. Many real strategies have edges far smaller than that.
Below about 30 closed trades, most statistics are too unstable to say much at all.
3. Adjust for how many versions you tried
If you tested twenty variations (different moving-average lengths, stop levels, filters) and kept the best one, that best one looks good partly because you picked it. This is the multiple-testing problem, and it is the most common way backtests mislead.
The honest fix is to count the variants you tried and make the bar higher accordingly. The simplest adjustment (Bonferroni) multiplies your p-value by the number of variants. A result with p = 0.04 after trying ten versions is not significant at all.
4. See what carries the result
Remove your five best trades and recompute the profit. If it turns into a loss, the strategy's result depends on a handful of outcomes you may not see again. Do the same by symbol: if one ticker produced most of the profit, you may have found one good trade rather than a repeatable method.
And then: costs
Even an edge that survives all four checks has to survive realistic trading costs. That's a topic of its own. See how fees and slippage affect strategy profitability.
What these checks don't tell you
They don't predict the future, and they can't detect problems the trade file doesn't show, such as a backtest that accidentally used information not available at the time. A strategy that passes is not obviously noise. That's a meaningful bar, but not a guarantee.
SignalPoint Strategy Audit runs these checks on a trade-history CSV and labels anything the data can't support as "not run". It is a statistical review of historical trades, not investment advice. Past results do not predict future results.