ForgeXBacktesting
How many trades do you need before a backtest means anything
Ten trades settle nothing. Here is why small samples mislead, and how to tell a real result from a lucky one.
This is the most common question in backtesting and it has an uncomfortable answer. You need far more trades than feels necessary, and the number of trades most people stop at is the number where a coin flip still looks like a strategy.
Why small samples lie
Flip a fair coin ten times and getting seven heads is not unusual at all. It happens about once in every eight attempts. Nobody would call that a biased coin.
Now imagine those ten flips were trades. Seven winners out of ten feels like a strategy. It looks like a 70 percent win rate. You would trade it.
Nothing changed except the label. A short run of luck looks identical to a real edge, and there is no way to tell them apart by staring at the results harder. The only cure is more trades.
The numbers people use
FTMO are direct about it. "To be really sure our trading system is stable and robust, we need a sample size of at least 100-200 trades" [1].
They also name the mistake precisely. "Some traders might test the first ten trades, and if they see their strategy works, they decide it is just enough and give up on further backtesting. This is definitely not a good approach as we don't have a robust data sample" [1].
As a rough guide. Under 30 trades, you have an anecdote. Between 30 and 100, you have a hint worth following. Above 100, you have something you can start to believe. Above 300, the number is fairly stable.
It depends on how big your edge is
A big edge shows up in fewer trades. A small one needs many more.
If your strategy wins 70 percent of the time, that stands out from random quickly. If it wins 53 percent of the time, which is a perfectly good edge in the right structure, you may need several hundred trades before you can tell it apart from 50 percent.
This is why traders with small, reliable edges need bigger samples than traders chasing rare setups. It is also why a rare setup you have only seen twelve times is untested no matter how good those twelve looked.
A test you can run yourself
Take your list of trades. Shuffle the order and add them up. Do it again. Do it a hundred times.
If almost every shuffle ends in profit, the result is coming from the trades themselves. If plenty of shuffles end in a loss, your total was carried by the order things happened in, which will not repeat.
This is a resampling test, and it is the honest version of asking whether your edge is real. On ForgeX, Vektor runs this automatically and reports the answer as a percentage.
What to do when you do not have enough
Keep testing. It is the only real answer, and it is slow on purpose.
In the meantime, trade smaller than you want to. Position size is where uncertainty should show up. If you are not sure the edge is real, that belongs in your size, not in your confidence.
Sources
- FTMO Academy, "How to Backtest Trading Strategies". https://academy.ftmo.com/lesson/how-to-backtest-trading-strategies/