This is the one lesson here with a negative result, and it is the most useful one in the series — not because of what it says about round numbers, but because of how it was tested.
Round numbers are tidy prices where traders are assumed to cluster their orders — and when the reaction at them is compared against the reaction at deliberately ordinary prices, the two are indistinguishable.
Round numbers are supposed to matter. Gold at 4,400. GBPUSD at 1.3500. The Dow at 45,000. The reasoning is that humans place orders at tidy figures — nobody sets a target at 4,397.30 when 4,400 is right there — so orders bunch up at round prices and price reacts when it gets there.
The reasoning is sound. Whether it shows up in price is a separate question, and it is the kind of question that almost never gets asked properly.
Suppose you claim a particular lamp post is unlucky, because you have tripped over near it three times. To test that, counting your trips near the lamp post is useless on its own. You have to count your trips near an ordinary stretch of pavement too. If you trip equally often there, the lamp post is innocent.
Price reacts somewhere constantly. Measuring reactions at round numbers alone will always produce an impressive-looking number. The only honest test is to run the identical measurement at prices with nothing special about them, and compare.
| Instrument | Round touches | Reaction at round | Control touches | Reaction at control | Difference |
|---|---|---|---|---|---|
| XAUUSD | 5,213 | 99.7 | 5,358 | 101.2 | -1.5% |
| GBPUSD | 4,119 | 26.5 | 4,338 | 26.6 | -0.4% |
| EURUSD | 3,329 | 20 | 3,705 | 20.1 | -0.5% |
| US30 | 4,343 | 138.3 | 4,234 | 138.3 | 0% |
| NAS100 | 3,652 | 83 | 3,711 | 85.6 | -3% |
What was counted: A level is round when it is a whole multiple of the instrument's round step. A touch is the first hourly candle to trade through a level the previous candle did not reach. The reaction is how far price pulled back away from the level over the following twelve hours. The control repeats the identical test at half-step prices, which are equally frequent and entirely arbitrary.
Reactions are in the instrument's own pips or points. Median rather than mean, so a handful of violent reversals cannot carry the result.
There is no effect. Across five instruments and more than twenty thousand touches each side, the reaction at round numbers is within a couple of per cent of the reaction at prices chosen for being unremarkable. On the Dow the two figures are identical.
This does not mean nobody places orders at round numbers. They plainly do. It means those orders are not concentrated enough to produce a reaction that stands out from what price does everywhere else — which is the only version of the claim that would be tradeable.
Everything else in this series measures a concept and reports a rate. This page reports a rate and a control, and that is the difference between a finding and a statistic.
If you take one habit from these lessons, take this one: whenever someone shows you that a pattern works X% of the time, ask what X% is for everything else. A setup that wins 60% of the time is remarkable if random entries win 50%, and worthless if they win 62%. Most trading education never mentions the second number, which is how ideas like this one survive for decades without being checked.
Not necessarily — but demote them, and be honest about the job they are doing.
Elena is short gold from 4,430 and picks 4,400 as her target because it is a round number and "price always reacts there".
What the data says about the premise. Gold's median reaction at a round number was 99.7 pips, against 101.2 pips at ordinary prices — a difference of -1.5%. There is no bounce waiting for her at 4,400 that would not equally be waiting at 4,395.
Is the target still fine? Yes, but for a different reason. A round target is easy to set in advance and easy to stick to, and both of those are worth something. Just do not expect the number itself to defend her.
Where the belief would cost her. If she moved her stop to 4,450 "because 4,450 is round and will hold", she would be putting her invalidation at an obvious price with no measured protection — the worst of both worlds. That is the version of this belief that actually loses money.
What she does instead. Keeps 4,400 as a target, places her stop by volatility rather than by tidiness, and stops describing the round number as a reason.
This is the most transferable idea in the whole series. Any pattern you are about to trade can be tested against a control in the backtester: run your rules, then run the same rules on entries picked at random, and compare. If the gap is small, the pattern was never the reason.
Test a pattern against a control →Run the drill in the free backtester → Free, no sign-in to begin.
Do round numbers act as support and resistance?
Not measurably. On gold, the median pullback after price first reached a round level was 99.7 pips; at half-step prices chosen for being unremarkable it was 101.2 pips. That is a difference of -1.5%, across thousands of touches on each side, and the other four instruments agree.
Why does everyone say round numbers matter then?
Because the reaction at round numbers, looked at on its own, is genuinely substantial — price does pull back after touching them. The problem is that price pulls back after touching everything. Without a control you cannot tell the two apart, and almost nobody runs the control.
Should I remove round numbers from my chart?
Not necessarily, but demote them. They are fine as take-profit targets, where the value is that a tidy number is easy to commit to and stick to. They are not a reason to enter, and they are a poor place to hide a stop — obvious prices are where everyone else's stop already is.
What about really big round numbers, like gold at 4,000?
This test used the ordinary trading step — every ten dollars on gold — because that is what traders actually watch. Much larger figures are far rarer, so any sample would be small and any conclusion weak. This page does not claim to have tested those.
What is the practical takeaway?
Two things. Do not put stops just beyond round numbers, because they are obvious even without an edge. And more importantly, start asking "compared to what?" of every trading claim you meet — a pattern that works 60% of the time means nothing until you know what random entries did.
How do I run a control test myself?
In the backtester: run your rules over a stretch of history and record the results, then run the same rules on entries taken at fixed intervals with no setup at all. The gap between those two is your actual edge, and it is usually smaller than the headline number.
Any trading claim can be tested against a control. Run your rules, then run the same rules on random entries, and compare. If the gap is small, the pattern was never the reason — and you have saved yourself a strategy.
Open the free backtester →Beginner exploration
Open each answer for a plain-language way to read Round Numbers, Tested Against a Control, test it carefully and decide what to explore next.
This page focuses on “Round Numbers, Tested Against a Control”.Round numbers tested against a control at half-step prices. On gold the median reaction was 99.7 pips at round levels and 101.2 at ordinary ones — a difference of -1.5%. No measurable edge, on any of five instruments.For “Round Numbers, Tested Against a Control”, a beginner should identify what the learning guide measures, assumes or teaches before acting on its conclusion.Treat this page's account of “Round Numbers, Tested Against a Control” as a learning reference rather than a prediction, signal or promise of future performance.
For “Round Numbers, Tested Against a Control”, translate the idea into a definition you could apply the same way on two different charts.While exploring “Round Numbers, Tested Against a Control”, work through one example slowly and record which inputs or observations determined the result.Keep your “Round Numbers, Tested Against a Control” record honest: list the limitation or counterexample before using the concept in a trading plan.Before leaving “Round Numbers, Tested Against a Control”, practise the definition on unseen history and review consistency before judging performance.
Turn one idea from “Round Numbers, Tested Against a Control” into a rule with explicit inputs, dates, costs and pass-or-fail conditions.Ask AI to expose missing assumptions in that “Round Numbers, Tested Against a Control” test, not to guess the next market move.Use the FXAbsolute AI Backtesting Lab to inspect calculations connected to “Round Numbers, Tested Against a Control” and the assumptions behind them.Reproduce any important “Round Numbers, Tested Against a Control” result and reserve unseen data before deciding that an apparent pattern is useful.
Continue your exploration of Round Numbers, Tested Against a Control with the beginner AI prompt guide, or inspect public calculations in the AI Backtesting Lab.