Backtesting SoftwareTest Your Edge on Real Market History

Automated backtesters take rules written as code and run them over years of history in seconds. You get thousands of trades, a parameter sweep and an equity curve before you've finished your coffee. MetaTrader's Strategy Tester, Python libraries like backtrader, and the strategy engines built into charting platforms all live here.

Manual replay tools do the opposite. They hide everything to the right of the current candle and step price forward one bar at a time while you read the chart and click your own entries, stops and targets. No code, no optimiser, no thousands of trades — you get however many setups you can work through by hand.

The mistake is buying across the divide, and it's extremely common: discretionary traders buy an optimiser, discover their setup can't be expressed as code, and never open it again; mechanical traders buy a replay tool and spend three weeks clicking through what a script would have answered in an afternoon.

There's a clean test for which side you're on. Could a stranger read your rules, never see the chart, and take exactly the trades you'd take? If yes, your strategy is mechanical — automate it. If your honest answer involves the words "it depends" or "you'd see it", it's discretionary, and no optimiser can test it. Replay is the only honest option. If you're not sure which you have, that ambiguity is itself the finding — writing rules precisely enough to be testable settles it faster than any software will.

Every serious tool in this space is good at something and bad at something else, and the marketing never tells you which. Here's the honest version.

Third-party plan limits and prices above were checked against each vendor's own pricing page on the date shown at the top of this page. These vendors run frequent sales — check their current pricing yourself before deciding.

Four things separate real backtesting software from a chart you can drag. It should replay real historical data bar by bar so hindsight can't creep in. It should log every trade and turn them into stats — win rate, average R, profit factor, drawdown. It should cover the markets you actually trade. And it should be fast enough that you can run hundreds of setups without it becoming a chore.

That last one gets dismissed as a nice-to-have and it is the one that decides whether you finish. A sample of 100 setups is not a big ask; at two minutes of clicking per setup it's a few evenings, and at ten it's a project you abandon. The arithmetic of backtesting throughput is worth doing before you pick a tool, not after.

If the software lets you drag the chart backwards and forwards freely, you will do it. Not deliberately — you'll nudge it two candles to "check", and you'll have seen the outcome before you marked the entry. That is the entire mechanism behind traders reporting 90% win rates in testing and 40% live. It isn't dishonesty, it's that the tool made the honest version optional.

Real replay makes it impossible rather than discouraged: the right-hand side does not exist, you commit before the next candle prints, and a trade you'd have skipped live is a trade you skip here. Everything else on a feature list is secondary to that one property. The related failures — cherry-picking the sample, quietly editing rules mid-test, calling twenty trades a result — are covered in the seven mistakes that make backtest results worthless.

Watch for the subtler version too. Software that fills you at a level price merely touched, or that lets you enter on a candle only after seeing it close, is producing a result you cannot reproduce live. That gap has a name and a size — why backtest results differ from live trading puts numbers on it.

Every backtester ends the same way: a results screen. This is the part people skim, and it is where most of the wrong conclusions get made — because the number that is easiest to read is the one that matters least.

Win rate is the number to distrust first. On its own it says nothing. A 40% win rate is excellent at 1:3 and fatal at 1:1, so any win rate quoted without the reward:risk beside it is decoration. The pairing that actually means something is expectancy — your average result per trade, in R — because that is the figure that compounds and the one you multiply by how often you trade.

Profit factor is the fastest honest read. Gross profit divided by gross loss: 2.0 means you won two for every one you lost. The useful band is bounded on both sides, and the top edge surprises people. Under 1.3 the edge is inside the margin that spread, slippage and one bad fortnight erase. Between 1.3 and 2.5 is where almost everything durable lands. Above about 4.0 you should audit rather than celebrate — over a genuine out-of-sample run that is rare enough that the base rate favours lookahead, an unrealistic fill, or three outliers carrying a thin sample.

Maximum drawdown is the only one that depends on the ORDER of your trades. Profit factor, expectancy and win rate are all order-independent — shuffle the trades and they do not move a thousandth. Drawdown does. The same twenty trades with the losers clustered at the front produce a far deeper hole than the same twenty spread out, and only one of those two accounts is still being traded at the end. It is the number that decides whether you could actually have held on, which is a different question from whether the strategy made money.

And none of it counts until the sample does. Thirty trades constrains almost nothing; a strategy with no edge at all clears a profit factor of 1.5 over thirty trades often enough that you should assume yours might have. The check worth running is a confidence interval on expectancy — if it straddles zero, the sample cannot yet separate your edge from a coin, and the answer is more trades rather than more optimisation.

You can run all of that on your own results in a browser: the profit factor calculator takes a pasted column of R-multiples and returns profit factor, expectancy, realised reward:risk, peak-to-trough drawdown, longest losing streak and the confidence interval, and tells you plainly when the sample is too small to support a verdict. The sample size calculator answers how many more trades you need, and the risk of ruin calculator answers whether the account survives the drawdown the results imply.

1. Is your strategy mechanical or discretionary? Answered by the stranger test above. This decides the category, and nothing else matters until it's settled.

2. Do you need your broker's exact data? If your rules hinge on a level being taken by a fraction of a point, or on spread at a specific hour, you need tick data from the feed you trade — that's MetaTrader or a desktop app with imported data. If you're testing structure, sweeps and session behaviour, any honest feed answers the question, and arguing about the last 0.2 of a pip is a way of avoiding the test.

3. How many setups can you realistically get through in a week? Multiply it by four. If that number isn't in the low hundreds, your bottleneck is throughput, not features — and you should pick whichever tool gets you from opening it to clicking a trade fastest, then optimise later.

Two questions that feel important and aren't: how many symbols the tool offers (you will test one or two, seriously, and testing thirty is how correlated markets inflate an apparent sample), and how many years of history it holds (until you've logged a few hundred trades, depth is not your constraint).

Browser-based replay tools run roughly $15–35 a month. Desktop simulators are sold as a licence plus a data package and run into the hundreds one-off. Automated testers built into trading terminals are typically free, because the broker wants you trading. Coding it yourself is free and costs you weeks.

You should not pay anything to start. Every credible option in this space has a free tier or a demo, and the honest thing to work out first is not which has the longest feature list — it is whether you will actually sit down and do the reps. Most people who buy a backtesting licence use it for a fortnight. So spend the first evening finding that out for nothing, on whichever tool you like the feel of, and only then look at what continuing costs you.

When you do look, look at what the free tier withholds rather than at the headline price. There are two ways to limit a free plan. One takes features away — no intraday replay, two sessions, a week of data retention — which means the free version cannot answer the question you came with. The other gives you the whole product and limits how much of it you use in a week. Only the second lets you find out, for nothing, whether your setup works on the timeframe you actually trade.

Free is the question most people searching this actually have, so here it is in one place. MetaTrader's Strategy Tester is fully free, for automated Expert Advisors only. TradingView's free plan is a great chart without Bar Replay, which is listed from the paid Essential tier upward. FX Replay's free plan is the real product capped at 2 sessions, 50 trades and a week of data retention. Forex Tester offers a demo of a paid licence. Python is free and costs you weeks.

CRTLAB's free sample is the whole product — 38 markets, full history, every timeframe and tool, stats scored as you go — with 10,000 candles of replay. That is several months of 15-minute sessions on one market, which is enough to reach a real sample before you have spent anything. Unlimited replay is $14.99 a month, cancel any time.

One practical note that decides it for a lot of people: if you are on a Mac, the desktop simulators are out or awkward. The backtesting software for Mac page covers exactly what runs natively and what needs Wine or a Windows virtual machine.

CRTLAB is browser-based backtesting software built around candle-by-candle replay — the manual side of the divide, deliberately. 38 markets: 17 US stocks and ETFs, seven crypto pairs, NAS100, US30, SP500, EURUSD, GBPUSD, USDJPY, gold, silver, oil and Bitcoin — replayed on real data, with your win rate, average R and profit factor tracked automatically as you go.

CRTLAB opens all 38 markets — indices, forex, metals, energy, crypto and 17 US stocks and ETFs — on the full history we hold, on every timeframe including intraday, with every drawing tool, indicator and overlay switched on. No card, no account, about 30 seconds. What is metered is candles advanced: 10,000 free sample first, or unlimited at $14.99 a month, cancel any time.

Size each test properly while you're at it — a backtest that ignores position sizing isn't reflecting real money. The position size calculator takes about ten seconds per market, and the backtest sample size calculator will tell you when you've logged enough trades for the result to mean anything.