How to Backtest a Trading Strategy Free, Online

What "backtesting free online" actually means

Three different things get sold under that phrase, and only one of them is a backtest.

The first is a free chart. You scroll back through history, look at a setup, and decide you would have taken it. This feels like research and produces nothing measurable, because you can already see what happened next. Hindsight is not a control you can switch off by promising to be honest.

The second is a free trial. The tool is complete, but the clock is running, and a strategy test that needs a weekend of replay does not sit comfortably inside a countdown that started the moment you registered.

The third - the one worth your time - is a free simulator: historical price replayed bar by bar with the future genuinely hidden, orders that fill at a price and are charged spread and commission, position sizes derived from a risk rule, and a report at the end that computes your statistics for you. "Online" simply means this runs in a browser: no licence key, no installer, no Windows-only build, and no separate expedition to find and clean historical data, because the datasets ship with the tool.

That last point is underrated. Assembling a free desktop setup means downloading historical data, checking it for gaps, importing it, and configuring the tester before you place a single trade. Most people who set out to backtest for free stall in that phase and never reach the part that produces an answer.

How to backtest a trading strategy for free: the seven steps

This is the whole method. It is the same on any instrument and any timeframe, and no step in it costs money.

1. Write the strategy down before you open a chart

One entry condition, one stop rule, one exit rule, in plain sentences you could hand to a stranger. This is the step people skip, and it is the real reason most backtests stall - not tooling. A rule like "enter on a strong pullback" cannot be tested, because "strong" gets decided after you see the outcome. "Enter when price retraces to the 20 EMA in an uptrend and closes back above it" can be tested, because it is either true on a bar or it is not.

2. Pick one instrument and one timeframe

Not five. A strategy tested across five markets at once produces a blended result that hides which market it actually works on. Start with the pair you would trade live - EUR/USD is the usual first choice because its spread is tight and its behaviour is well documented - and add instruments only after you have a verdict on the first one.

3. Jump to a random historical date

Left to yourself, you will start somewhere you vaguely remember, and vague memory selects for periods that suited your strategy. A random start date removes that. It costs nothing and it is the cheapest defence there is against curve fitting.

4. Step forward bar by bar, with the future hidden

This is the mechanical difference between backtesting and reminiscing. You see the chart exactly as you would have seen it live, decide, and only then reveal the next bar. If the tool lets you peek ahead - or if you are scrolling a static chart and deciding retrospectively - the results will flatter you, and that is the most common backtesting mistake there is.

5. Size every position from a fixed risk rule

Set risk as a percentage of the account - 1% is the standard - and let the stop distance determine the lot size. Fixed-lot testing quietly turns your results into a story about which trades happened to be big, not about whether the strategy has an edge. Any simulator worth using does this arithmetic for you.

6. Take every setup that meets the rules

Every one, including the ugly ones you would "obviously" have skipped. Skipping is selection bias, and it inflates every statistic downstream of it. If a filter is genuinely part of your strategy, write it into the rules in step 1 and apply it consistently; if you cannot write it down, it is not a rule, it is a mood.

7. Stop at 100+ trades and read the report

Below about 30 trades, win rate is noise. A hundred or more, spanning both trending and ranging conditions, is where the numbers start meaning something. Then read the four that matter: expectancy per trade, profit factor, maximum drawdown, and the shape of the equity curve. Positive expectancy with a drawdown you could actually sit through means move to forward testing. Negative expectancy means you just saved yourself months of live losses - which is also a win, and it also cost nothing.

How to backtest gold (XAU/USD) for free

Gold is the market people most often try to backtest for free and most often get wrong, because the method above is right but the sizing intuition carried over from forex is not.

Gold is not a currency pair. It is priced in dollars and cents per ounce with a different contract value per point, so a move that sounds the same size as a forex move is a completely different amount of money. Any simulator that carries gold properly handles this arithmetic for you - but you still have to respect it when you choose a stop.

The bigger issue is that gold's volatility regime shifted dramatically inside the period most people test. From our own measurement of 3,419 XAUUSD daily bars:

Two consequences for a free gold backtest. First, a stop distance from 2018 is not a stop distance for 2025. Gold's average day ran about five times wider at the end of that window than in the middle of it, so a fixed stop that was generous in 2018 sits inside the noise today. Size off ATR or a percentage of price instead, and the stop travels with the regime rather than being fitted to one.

Second, gold is steadier than the raw numbers suggest. As a share of price the daily range stayed between 0.83% and 1.53% across the whole decade - roughly 1.8 times EUR/USD, not the three-times-plus the point counts imply. The dollar range exploded mostly because gold itself went from around $1,150 to well over $3,000.

Practically: when you pick a test window for gold, deliberately include at least one quiet year (2017-2018) and one violent one (2020 or 2025), so the result is a strategy verdict rather than a single volatility regime in disguise. The full year-by-year table and the rest of gold's mechanics are in how to backtest gold (XAU/USD).

Computed 2026-08-27 from the FxBacktest tick-derived OHLC archive: 3,419 XAUUSD and 3,440 EURUSD daily bars, 2015-2025. Ranges are the daily high-low, averaged by year; percentages divide that range by the day's close.

What free actually limits - and what it doesn't

Free tiers are not all shaped the same, and the shape decides whether the tier is usable. The fair kind limits storage - how many finished sessions you keep - or the depth of secondary analytics. Your current test is never interrupted; you simply curate what you archive, and the verdict numbers stay visible.

The kind to walk away from breaks the loop itself: trials that expire mid-test, tools that block trade placement until you pay, or reports that show a teaser and paywall the actual statistics. A single question sorts them: can this free tier take one complete 100-trade test from start to finished report? If yes, the tier is real. If answering requires a card number, "free" is doing marketing work rather than describing the product.

Being specific about our own: FxBacktest's free tier runs the full bar-by-bar engine across all 93 instruments and all seven timeframes, with no card and no countdown. The report you get for free includes win rate, average risk-reward, net P&L, profit factor, expectancy, maximum drawdown, recovery, the balance curve and streak analysis. What free caps is saved sessions - three of them - and a handful of secondary breakdowns such as wins and losses by day of week. It does need a free account with an email address; the account is what makes saved sessions and a continuous testing history possible at all. New accounts also get ten days of Pro, so you can see the deeper analytics before deciding whether they matter to you.

Measured: the depth a free archive has to have

The usual failure of free backtesting is not the interface - it is history that stops two years back, which quietly caps how good your answer can be. For transparency, here is what sits behind our own free tier, counted from the archive:

That is 2.2 million hourly bars across 28 instruments, with the majors reaching back to 2010 and gold to July 2009. Depth matters more than instrument count: sixteen years spans the 2015 franc shock, the 2016 referendum, the 2020 crash and the 2024-25 gold run, so a strategy meets genuinely different regimes instead of one recent stretch. When you compare free options, ask for the start year per instrument. It decides whether the tool can answer your question at all, and it is the number most often left off marketing pages.

Counted 2026-08-27 directly from the FxBacktest OHLC archive (28 instruments, 2,199,262 hourly bars, earliest bar 21 July 2009 on XAUUSD). Derived from Dukascopy tick data and refreshed nightly.

Bottom line

Backtesting a strategy for free online stopped being a compromise the moment browser simulators started executing orders and computing statistics. The burden of proof now sits on paid software: it has to name a capability you specifically need - coded parameter sweeps, tick-level fills inside the candle, offline work - and for most discretionary traders it cannot. Write one strategy down, replay it to a hundred trades on the market you actually trade, and let the report decide what happens next. If that market is gold, do it across two different volatility regimes. The method costs nothing but the weekend.