How Long Does It Take to Backtest a Strategy?

Nobody abandons a backtest because the method was wrong. They abandon it because it was taking four evenings a week and they were on trade nineteen.

That's the failure mode this article is about. Every guide tells you to log 100–300 trades and none of them tell you what that costs in hours, so people start with an open-ended commitment, discover on night three that they're 6% of the way there, and quietly stop. Then they conclude that backtesting doesn't work, when what actually happened is that they never estimated the job before starting it.

So let's estimate it properly. The number is knowable, it's mostly arithmetic, and once you can see it you can attack the part that's actually expensive — which is almost certainly not the part you think.

Where the backtesting hours actually go (it isn't the trades)

Here's where the hours actually go when you replay by hand. Marking a trade — entry, stop, target, logging the outcome — takes somewhere between 40 seconds and two minutes once you've done it a few dozen times. Call it 90 seconds.

If your sample is 150 trades, that's under four hours of trade-marking. Four hours. That is not a project.

The rest of the time — and untouched, it's the overwhelming majority — goes on advancing through candles where nothing happens. Your setup doesn't form every bar. It forms a few times a week, and between one occurrence and the next you have to move through every candle in between, because the moment you start skipping ahead to find the next one you're using information you wouldn't have had live, and the whole exercise is worthless.

That's the real shape of the job:

Backtesting time ≈ (time to cross the dead space) + (90 seconds × number of setups)

On an intraday chart, before you do anything about it, that first term starts out five to twenty-five times larger than the second. Which means every hour you save comes from how fast you can cross empty market — and that the goal isn't to eliminate that term, it's to beat it down until it's roughly the same size as the trade-marking term. At that point you're done optimising, because the remaining time is the work itself.

Most people attack the wrong term entirely. They build a slicker logging spreadsheet and save eleven minutes across a project that's costing them thirty hours.

The arithmetic, with real numbers

Two inputs give you an estimate: how often your setup occurs, and how much chart you have to cross to find each one.

Say your setup appears three times a week on one instrument. For a 150-trade sample:

150 setups ÷ 3 per week = 50 instrument-weeks
50 weeks ≈ 11.5 months of chart time to cross

Eleven and a half months of price. Now, how long does crossing eleven and a half months of price take? Entirely depends on the timeframe you're replaying and how fast you can move through it.

Replay timeframe	Candles in ~11.5 months	At 1 candle/sec	At 20 candles/sec
1-minute	~360,000	100 hours	5 hours
5-minute	~72,000	20 hours	1 hour
15-minute	~24,000	6.7 hours	20 minutes
1-hour	~6,000	1.7 hours	5 minutes

Approximate: assumes 24-hour markets on a five-day week. Index and stock sessions have fewer candles, forex more.

Add the four hours of trade-marking to any row and you have your real estimate.

Look at what that table is telling you. The same 150-trade sample is a two-evening job on the 15-minute chart and a two-and-a-half-week job on the 1-minute chart at the same click rate. It's the same strategy, the same sample, the same rigour. The only thing that changed is throughput.

And notice the right-hand column against the left. Speed control is the difference between 100 hours and 5. Nothing else in the entire process has a 20× lever in it.

The three levers, in order of how much they're worth

1. Speed through the dead space. By a wide margin the biggest one, as the table shows. If you can only advance one candle per click, a 1-minute sample is arithmetically out of reach — not hard, out of reach, because 360,000 clicks is not a thing a person does. Being able to hold fast-forward through a quiet Tuesday afternoon and then crawl candle-by-candle through the twenty minutes around your setup is what makes an intraday sample possible at all.

The key is that it has to be variable. A fixed fast speed is nearly as bad as a fixed slow one — you'll blow straight through the entry you were waiting for, and then you're either re-running the segment or, worse, marking an entry you've already seen resolve.

2. Narrow the window you replay. If your setup only ever forms in the first two hours of the New York session, you are crossing 22 hours of chart for every 2 you need. Restricting your test to the session your rules actually operate in is a legitimate 5–10× saving and it costs you nothing, because those other candles were never eligible for a trade under your own rules.

Be careful with the honest version of this: you can skip hours in which your rules make you ineligible to trade, but you cannot skip hours because nothing interesting happened in them. The first is a rule; the second is hindsight. If you're not sure which side you're on, the test is whether you could have known the window was skippable before seeing it — a session filter passes, "that day looked boring" doesn't.

3. Cut logging friction. Worth the least, but it's free. Log as you go, never afterwards. Fixed fields, decided before you start, so you're never inventing a column at trade 60 and having to backfill 59. If you're alt-tabbing to a spreadsheet for every trade you're adding 20–30 seconds each and, more damagingly, breaking concentration in a task where concentration is the thing being trained. What to log in a trading journal covers the field set worth committing to.

What you cannot speed up, and shouldn't try

There's a hard floor, and pushing through it is how a fast backtest becomes a worthless one.

The decision itself. When your setup is forming, you have to make the same call you'd make live, with the same information, at the same moment — before the next candle prints. Rushing that is not efficiency, it's changing the experiment. If you catch yourself deciding faster in replay than you would with money on it, you're no longer testing the thing you'll trade.

The skips. A setup that formed and that you chose not to take is data, and it takes the same 90 seconds to record. The instinct to not bother logging skips is the single fastest way to make a sample dishonest, because skips are where your discretion lives. Backtesting mistakes covers what happens to a result when they go missing.

The scary bits. The temptation when you're moving fast is to speed up through drawdown — three losses in a row and the fast-forward gets a bit more generous. That's the exact stretch of the sample you're doing this for. Slow down there, not up.

There's also a subtler failure worth naming: speed loosens your rules. When you're moving quickly you start accepting setups that are approximately right, because stopping to adjudicate a marginal one breaks your rhythm. Twenty trades later your effective entry criteria have drifted from what you wrote down and you're measuring a strategy nobody has ever traded. If you notice this happening, the fix isn't to slow down — it's that your rules weren't precise enough to be applied at speed. Write them so they can be.

A realistic schedule for 150 trades

Here's what it looks like as a plan rather than an open-ended commitment. Assume a 5-minute chart, one instrument, one session, variable replay speed:

Chart to cross	~50 instrument-weeks
Crossing time, session-filtered, at speed	4–6 hours
Trade marking, 150 × 90 sec	~4 hours
Review and segmentation at the end	~2 hours
Total	10–12 hours

That's five or six focused evenings. Not a season of your life — an estimate you can commit to, hold yourself to, and finish.

And it's worth saying what that buys, because the number sounds small for the claim: after ten hours you have a real win rate, a real average R, a real worst losing streak, and a sample large enough that a bad week doesn't make you doubt the whole thing. Compare that against learning the same lessons live at, say, 1% risk per trade over 150 trades. The ten hours is the cheapest tuition available in this business by a very wide margin.

If the estimate you get for your own setup comes out above 25 hours, don't just grind it. That's a signal — either your setup is too rare to test on one instrument (test the same setup on a second correlated market, but count the correlation problem against your effective sample size), or you're replaying a finer timeframe than your rules actually need.

Do this before you start, not after

Spend ten minutes on the estimate. Count your setup's frequency over the last month of chart, pick your timeframe, look up the row in the table, add four hours.

If the number is 12 hours, commit to it and schedule it. If it's 80, change something — timeframe, session filter, replay speed, instrument — until it isn't, before you begin. The traders who finish samples aren't more disciplined than the ones who don't. They just knew what they were signing up for.

Then check the other end of the arithmetic: how many trades your particular edge actually needs before the result means anything. A big edge is provable in far fewer trades than a thin one, and there's no point budgeting for 300 if 80 would have settled it — or budgeting for 150 when your edge is thin enough that nothing short of 600 will.

FAQ

How long does it take to backtest a trading strategy?

For a manual, discretionary backtest of 150 trades on a 5-minute chart with variable replay speed and a session filter, budget 10–12 hours — about five or six evenings. The dominant variable is not the number of trades but how fast you can advance through the candles between them, which is why the same sample can take 5 hours or 100 depending on your timeframe and speed control.

How many trades should a backtest have?

Enough that the result clears noise, which depends on the size of the edge rather than on a universal number. A strong edge of around +0.5R per trade can be distinguishable in under 50 trades; a thin +0.1R edge needs a sample in the high hundreds. The backtest sample size calculator gives you the number for your own observed win rate and reward:risk.

Is it faster to backtest on a higher timeframe?

Dramatically — there are 60 times fewer 1-hour candles than 1-minute candles in the same span of history. But test the timeframe you actually trade. A fast backtest of a strategy you don't run is worth nothing, and higher-timeframe setups occur less often, which partly cancels the saving in candles.

Can I speed up backtesting by skipping to the interesting parts?

No, and this is the one shortcut that destroys the result. Deciding a stretch of chart is skippable requires knowing what happened in it, which is information you wouldn't have had live. Skipping hours your rules make you ineligible to trade — a session filter, say — is fine, because that decision is made in advance and applies whether or not anything interesting occurs.

How long does 100 trades take on the 1-minute chart?

Longer than most people expect: roughly 240,000 candles to cross for a setup that occurs three times a week, which is 65+ hours at one candle per second and about 3 hours at 20 per second. On the 1-minute chart, replay speed control isn't a convenience — it's the difference between a feasible sample and an impossible one.

Is manual backtesting worth the hours compared to automated?

If your strategy is mechanical, no — code it, and get the same answer in seconds. If it's discretionary, the hours are the point: you're not just measuring a win rate, you're building the pattern recognition that lets you act on it under pressure. An automated backtester cannot do that, and it cannot test rules that require judgement in the first place.

---

Cut the arithmetic down to size. CRTLAB replays real historical price candle by candle in your browser across 21 markets free — no card, no trade limit — and logs every trade as you go. Pro adds full replay speed control, so you can hold fast-forward through a quiet week and crawl through the minute around an entry, plus up to ten years of history on all 38 markets. Start free and see how many setups you can get through in one evening.