Somebody told you 30 trades is enough. Somebody else told you that you need 10,000 or the whole thing is worthless. Both numbers are pulled from thin air, and neither one asks the only question that matters: enough to prove what ? A backtest is not a certificate. It is a sample, and a sample is only useful if it is large enough to separate your edge from luck and varied enough to have seen a few different markets. Those are two different problems, and most traders solve neither. The honest answer is: it depends on your win rate, your reward-to-risk, and how much variety your sample has. But you can get to a real number without a statistics degree, and this post walks you through the reasoning so you can stop guessing. What sample size actually buys you The whole point of counting trades is to protect yourself from randomness. Flip a fair coin ten times and you can easily get seven heads. That does not mean the coin is biased. A strategy that wins 6 of its first 10 trades tells you almost nothing, because a 50% strategy would produce that result constantly. The more trades you log, the harder it is for a mediocre system to fake a good-looking result. Two things drive how many trades you need: Your win rate. A system that wins 30% of the time needs more trades to prove itself than one that wins 60%, because low-win-rate systems are lumpy. You can lose eight in a row and still have a profitable edge, but eight in a row inside a 40-trade sample looks like a disaster. How big your edge is. A strategy that clears a strong, consistent profit shows up fast. A wafer-thin edge (a small positive expectancy) hides inside noise for a long time, and you need a lot of trades before it separates from zero. This is why a fixed rule like '30 trades' is useless. Thirty trades might genuinely settle the case for a high-frequency scalp with an obvious edge, and be completely inadequate for a swing setup that trades twice a month with a coin-flip win rate. Rough numbers you can actually use Here is a practical ladder rather than a single magic figure. Under 30 trades: an anecdote You have a story, not evidence. This range is fine for a first look, to check that your rules are even followable and that the setup appears often enough to bother with. Do not size real risk off it. 30 to 100 trades: a hint Now you can see the shape of the thing. Is it roughly a trend follower with a low win rate and big winners, or a mean-reversion system with lots of small wins? You can start estimating expectancy and spotting whether the profit came from the whole sample or from two lucky months. Treat any edge here as provisional. 100 to 300 trades: a working sample This is the range where most retail strategies start to mean something. At 100-plus trades a low-win-rate system has had room to string together its losing runs, and a thin edge has had a chance to either show up or evaporate. If your strategy is still profitable across 200 trades that were spread over different market conditions, you have something worth forward testing. 300-plus trades: comfortable More is better for confidence, with one caveat covered below. Beyond a few hundred trades you are mostly firming up numbers you already believe, not discovering whether the edge exists. If you want a single sentence to remember: 100 trades is a floor for most strategies, 200 to 300 is where you can start trusting the numbers, and anything under 30 is a conversation starter, not a conclusion. Variety beats raw count Here is the part the trade-count debate almost always misses. Two hundred trades all taken from a single trending year in EURUSD is a worse sample than 120 trades spread across a trend, a range, and a high-volatility news period. Quantity without variety just measures how well your rules fit one market. A strategy that only ever saw 2021 has never met a choppy, mean-reverting environment, and it will find one in live trading. When you build your sample, deliberately drag it across: An uptrend, a downtrend, and a sideways range A calm period and a volatile one Different seasons, since some setups behave differently around seasonal patterns and month-end flows If your edge survives all of those, the raw trade count matters less. If it only worked in one regime, no number of trades from that regime will save you. The trap of a huge, over-optimised sample There is a failure mode on the other end too. You can keep tweaking a strategy against a big historical sample until it fits that exact history beautifully and predicts nothing about the future. That is overfitting , and a large trade count does not protect you from it. It can even make it worse, because a system with 40 tuned parameters can memorise 1,000 trades far more easily than a simple two-rule system can. The defence is not more data. It is fewer moving parts. Every extra rule, filter, or indicator you add to make the backtest look better is a place where you might be fitting noise. A blunt strategy that works okay across many conditions beats a finely tuned one that works brilliantly across the exact history you tuned it on. Count trades, not just time People conflate 'a year of data' with 'enough trades', and they are not the same. A setup that fires three times a month gives you 36 trades a year. To reach a working sample of 150 to 200 trades you would need four or five years of history, and that is before you account for regime variety. A scalping approach might hit 200 trades in a fortnight. So think in trades, then check the calendar afterwards to make sure those trades are not all bunched into one market mood. Both numbers have to pass. How to actually build the sample The slow, honest way is to click through historical candles bar by bar, mark your entries and exits by your written rules, and log every result including the ones you would rather forget. It is tedious, and that is exactly why it works. You cannot cherry-pick outcomes you have not seen yet if you are stepping forward one candle at a time. If you want a walk-through of the method, there is a full guide on backtesting without code . This is the part TradeSave+ is built for. The backtesting tool lets you step through historical price one candle at a time, place trades by your rules, and it records the outcomes into a journal automatically, so by the time you have your 150 or 200 trades you already have the win rate, expectancy, and equity curve sitting in front of you rather than in a spreadsheet you have to nurse by hand. Logging the sample and analysing it are the same action. Reading the sample once you have it A big enough sample answers three questions. Look at all three, not just the profit at the bottom. Did the edge come from the whole sample or a few outliers? Pull up the equity curve. A steady climb is believable. A flat line with two huge jumps means your 'edge' is really two trades, and 200 becomes 2 in a hurry. What is the worst losing run? Your longest string of losses in the backtest is the minimum you should expect live, and live will usually be worse. If a 12-trade losing streak in your history would have made you quit, the strategy is untradeable for you regardless of its average. Is the expectancy comfortably above zero, or barely? A thin positive edge across 300 trades can still be noise. A clear one across 150 is more convincing. This is where the debate between R-multiple and win rate earns its keep, because a low win rate with big winners is fine, as long as you have enough trades to prove the big winners are reliable and not flukes. The step nobody wants to hear Even a clean 200-trade backtest is a hypothesis, not a verdict. History is the one dataset where you already know the answer, so it flatters every strategy a little. The only test that counts double is putting the system in front of price it has never seen. Forward testing on a demo or in small size is slower and less satisfying than scrolling through years of candles in an afternoon, and it is the single most honest thing you can do. There is more on why in this piece on forward testing versus backtesting . So the number you are looking for is not really a number. Aim for at least 100 trades, prefer 200 to 300, spread them across different market conditions, keep the strategy simple enough that it cannot memorise the past, and then earn your confidence forward. A backtest of the right size does not tell you the strategy will work. It tells you the strategy is worth risking real attention on, which is all a backtest was ever able to do.
How Many Trades You Need to Backtest a Strategy (the honest number)
The '30 trades and you're done' rule is wrong in both directions. Here is how to size a backtest sample that actually tells you something.
Somebody told you 30 trades is enough. Somebody else told you that you need 10,000 or the whole thing is worthless. Both numbers are pulled from thin air, and neither one asks the only question that matters: enough to prove what ? A backtest is not a certificate. It is a sample, and a sample is only useful if it is large enough to separate your edge from luck and varied enough to have seen a few different markets. Those are two different problems, and most traders solve neither. The honest answer is: it depends on your win rate, your reward-to-risk, and how much variety your sample has. But you can get to a real number without a statistics degree, and this post walks you through the reasoning so you can stop guessing. What sample size actually buys you The whole point of counting trades is to protect yourself from randomness. Flip a fair coin ten times and you can easily get seven heads. That does not mean the coin is biased. A strategy that wins 6 of its first 10 trades tells you almost nothing, because a 50% strategy would produce that result constantly. The more trades you log, the harder it is for a mediocre system to fake a good-looking result. Two things drive how many trades you need: Your win rate. A system that wins 30% of the time needs more trades to prove itself than one that wins 60%, because low-win-rate systems are lumpy. You can lose eight in a row and still have a profitable edge, but eight in a row inside a 40-trade sample looks like a disaster. How big your edge is. A strategy that clears a strong, consistent profit shows up fast. A wafer-thin edge (a small positive expectancy) hides inside noise for a long time, and you need a lot of trades before it separates from zero. This is why a fixed rule like '30 trades' is useless. Thirty trades might genuinely settle the case for a high-frequency scalp with an obvious edge, and be completely inadequate for a swing setup that trades twice a month with a coin-flip win rate. Rough numbers you can actually use Here is a practical ladder rather than a single magic figure. Under 30 trades: an anecdote You have a story, not evidence. This range is fine for a first look, to check that your rules are even followable and that the setup appears often enough to bother with. Do not size real risk off it. 30 to 100 trades: a hint Now you can see the shape of the thing. Is it roughly a trend follower with a low win rate and big winners, or a mean-reversion system with lots of small wins? You can start estimating expectancy and spotting whether the profit came from the whole sample or from two lucky months. Treat any edge here as provisional. 100 to 300 trades: a working sample This is the range where most retail strategies start to mean something. At 100-plus trades a low-win-rate system has had room to string together its losing runs, and a thin edge has had a chance to either show up or evaporate. If your strategy is still profitable across 200 trades that were spread over different market conditions, you have something worth forward testing. 300-plus trades: comfortable More is better for confidence, with one caveat covered below. Beyond a few hundred trades you are mostly firming up numbers you already believe, not discovering whether the edge exists. If you want a single sentence to remember: 100 trades is a floor for most strategies, 200 to 300 is where you can start trusting the numbers, and anything under 30 is a conversation starter, not a conclusion. Variety beats raw count Here is the part the trade-count debate almost always misses. Two hundred trades all taken from a single trending year in EURUSD is a worse sample than 120 trades spread across a trend, a range, and a high-volatility news period. Quantity without variety just measures how well your rules fit one market. A strategy that only ever saw 2021 has never met a choppy, mean-reverting environment, and it will find one in live trading. When you build your sample, deliberately drag it across: An uptrend, a downtrend, and a sideways range A calm period and a volatile one Different seasons, since some setups behave differently around seasonal patterns and month-end flows If your edge survives all of those, the raw trade count matters less. If it only worked in one regime, no number of trades from that regime will save you. The trap of a huge, over-optimised sample There is a failure mode on the other end too. You can keep tweaking a strategy against a big historical sample until it fits that exact history beautifully and predicts nothing about the future. That is overfitting , and a large trade count does not protect you from it. It can even make it worse, because a system with 40 tuned parameters can memorise 1,000 trades far more easily than a simple two-rule system can. The defence is not more data. It is fewer moving parts. Every extra rule, filter, or indicator you add to make the backtest look better is a place where you might be fitting noise. A blunt strategy that works okay across many conditions beats a finely tuned one that works brilliantly across the exact history you tuned it on. Count trades, not just time People conflate 'a year of data' with 'enough trades', and they are not the same. A setup that fires three times a month gives you 36 trades a year. To reach a working sample of 150 to 200 trades you would need four or five years of history, and that is before you account for regime variety. A scalping approach might hit 200 trades in a fortnight. So think in trades, then check the calendar afterwards to make sure those trades are not all bunched into one market mood. Both numbers have to pass. How to actually build the sample The slow, honest way is to click through historical candles bar by bar, mark your entries and exits by your written rules, and log every result including the ones you would rather forget. It is tedious, and that is exactly why it works. You cannot cherry-pick outcomes you have not seen yet if you are stepping forward one candle at a time. If you want a walk-through of the method, there is a full guide on backtesting without code . This is the part TradeSave+ is built for. The backtesting tool lets you step through historical price one candle at a time, place trades by your rules, and it records the outcomes into a journal automatically, so by the time you have your 150 or 200 trades you already have the win rate, expectancy, and equity curve sitting in front of you rather than in a spreadsheet you have to nurse by hand. Logging the sample and analysing it are the same action. Reading the sample once you have it A big enough sample answers three questions. Look at all three, not just the profit at the bottom. Did the edge come from the whole sample or a few outliers? Pull up the equity curve. A steady climb is believable. A flat line with two huge jumps means your 'edge' is really two trades, and 200 becomes 2 in a hurry. What is the worst losing run? Your longest string of losses in the backtest is the minimum you should expect live, and live will usually be worse. If a 12-trade losing streak in your history would have made you quit, the strategy is untradeable for you regardless of its average. Is the expectancy comfortably above zero, or barely? A thin positive edge across 300 trades can still be noise. A clear one across 150 is more convincing. This is where the debate between R-multiple and win rate earns its keep, because a low win rate with big winners is fine, as long as you have enough trades to prove the big winners are reliable and not flukes. The step nobody wants to hear Even a clean 200-trade backtest is a hypothesis, not a verdict. History is the one dataset where you already know the answer, so it flatters every strategy a little. The only test that counts double is putting the system in front of price it has never seen. Forward testing on a demo or in small size is slower and less satisfying than scrolling through years of candles in an afternoon, and it is the single most honest thing you can do. There is more on why in this piece on forward testing versus backtesting . So the number you are looking for is not really a number. Aim for at least 100 trades, prefer 200 to 300, spread them across different market conditions, keep the strategy simple enough that it cannot memorise the past, and then earn your confidence forward. A backtest of the right size does not tell you the strategy will work. It tells you the strategy is worth risking real attention on, which is all a backtest was ever able to do.