Backtesting means testing a method on past data. Market replay is its manual form: the market is played back candle by candle, without showing what comes next, and you take your decisions as you would live. The difference between the two fits in one word, and that word decides how much the result is worth.
Contents
Traders define backtesting as applying a method to past data to find out what it would have returned. The backtesting meaning most people have in mind is the automated one: a program sweeps the history and gives a great many figures very fast. Done by hand, it becomes replay: the market moves forward candle by candle and stops where you are.
The distinction that matters is not technical, it is cognitive. An automated test has no opinion, so it does not cheat, but it only tests what you managed to write down as rules. A replay tests what you actually do, hesitations included, on the strict condition that what comes next stays hidden. Re-reading a chart whose outcome you already know tests nothing at all: it confirms.
And what replay does not reconstruct has to be said. It replays candles, meaning a price through time. It rebuilds neither the order book, nor the premiums and implied volatility of an option chain, nor the real slippage of a large order. On those subjects, an honest replay says what it does not know rather than pretending.
A useful test = a setup defined in advance × enough occurrences × the outcome hidden
The setup defined in advance is the condition most often missed. A setup that sharpens as the replay goes describes the trades you took, not a rule.
The number of occurrences is what replay buys you: it is compressed time, and it is its only real promise.
The hidden outcome is not negotiable. Scrolling a chart you have already seen produces flattering statistics describing no decision at all.
The setup holds in replay and does not hold live. The method is not what is at fault, the gap between the two populations is: 52 decisions taken calmly against 19 taken in session. That figure only appears if you compute the same statistics on both sides, and it points at execution work rather than strategy work.
It is the one promise replay keeps for certain, and it is enough to justify it. A setup appearing twice a week needs six months to give fifty live occurrences, and one evening in replay.
Below about thirty cases, no figure drawn from a test deserves changing anything. That floor is exactly what replay lets you reach before putting money at risk.
Compute the same statistics on both sides, on the same setup. The gap between replay expectancy and live expectancy is a direct measure of what session conditions cost you, and it is a figure nothing else gives.
It also settles a question people otherwise ask themselves for years: should I change method, or learn to hold the one I have.
The first flaw is the chart you have already seen. Replaying a period you remember tests no decision, it confirms a memory. It is the most common flaw and the hardest to admit, because the result is excellent and it really does feel like work.
The second is the setup that sharpens along the way. Adding a condition after an occurrence that would have lost produces a rule cut from its own data, and it will never repeat. The rule is written first, in one sentence, and it does not move until the test ends.
The third is taking replay at its word. It knows neither the order book, nor slippage on a large order, nor an option chain. A setup whose profitability rests on a few ticks is precisely the one replay measures worst.
The classic mistake. Replaying a period you lived through. The result is excellent, the work looks serious, and nothing has been tested: you do not take a decision when you already know the outcome, you reconstruct it. An older period or an instrument you do not follow costs exactly the same time and returns a figure that means something.
| Indicator | Formula | What it measures | When to use it |
|---|---|---|---|
| Replay | decisions taken by hand, outcome hidden | What you actually do | Testing a discretionary method |
| Automated backtest | a program sweeps the history | What your written rules return | Testing a fully coded method |
| Demo account | real time, fake money | Execution, at real speed | Getting used to a platform |
| Live trade journal | what actually happened | What you do with real money | Everything else in this glossary |
One sentence, not a page, and it does not move until the test is over. A setup that sharpens along the way produces a rule cut from its own data, and it will never repeat in session.
It is the only protection against confirmation. A chart you remember gives an excellent, empty test. An older period, or an instrument you do not follow, costs the same time and is worth far more.
Win rate, expectancy and average R on the replay, then on the same setup live. It is not the replay figure that teaches you something, it is the gap between the two.
It is applying a method to past data to find out what it would have returned. Backtesting trading strategies this way says what a rule would have given, and nothing about what you would have done with it in session. Done by a program, it sweeps the history and gives many figures fast. Done by hand, it becomes replay: the market moves forward candle by candle and stops where you are.
An automated backtest tests written rules, perfectly and without opinion, but it only tests what you managed to write down. Market replay tests what you actually do, hesitations included, on the strict condition that the outcome stays hidden. A discretionary method cannot be tested any other way.
Thirty at the very least, and fifty is more comfortable. Below that, the gap you observe comes from the luck of the sample and no figure deserves changing a rule. That floor is exactly what replay lets you reach in one evening instead of six months.
Three rules are enough: write the setup in one sentence before starting and never touch it again, replay a period you did not live through, and compute the same statistics on the replay and on your live trading. The third is the one people skip, and it is the only one producing new information.
No, and it is a limit worth knowing before relying on it. A replay plays back candles, meaning a price through time. It rebuilds neither an option chain with its premiums and implied volatility, nor the order book, nor the real slippage of a large order.
The Backtesting screen replays history candle by candle without showing what comes next, with a simulated execution engine. Costs are set when the session opens, spread, slippage, commission and swap, and each order type has its rule: a market order fills at the candle close with spread and slippage, a limit order fills at the level or better with no slippage, a stop order takes the slippage because it consumes liquidity. Stepping back is possible, and it really erases what was done after the point you return to, on screen and in the database, so a session never keeps two versions of the same decision.
Trades taken in replay are analysed with the same statistics as your real trades, which is the whole point: the win rate, R and expectancy of a backtest session compare directly with those in your journal. A session's resume point, however, lives in memory: closing the tab loses it. The result itself is saved at every step.
To go deeper on this: read the full guide. Every term is gathered in the glossary.