Unproven — held loosely on purpose.
The mechanic is real: brokers charge three days of rollover on Wednesday to cover weekend settlement. Whether that makes Wednesday's rollover hour special is a different question. Testing it against the same hours on other weekdays, 5 of 18 markets differ with signs disagreeing, on samples a fifth the size of our other tests. A plausible mechanism is exactly the condition under which people over-read a weak result.
| Market | In the window | Rest | z | Movement | Samples |
|---|---|---|---|---|---|
| USD/JPY | 46.4% | 52.3% | -4.4 | 163% of normal | 1,819 |
| EUR/USD | 56.6% | 52.2% | 3.7 | 131% of normal | 2,361 |
| USD/CHF | 46.5% | 50.7% | -2.6 | 66% of normal | 1,289 |
| GBP/JPY | 51.8% | 55.1% | -2.3 | 84% of normal | 1,665 |
| BTC/USD | 56.9% | 53.0% | 2.0 | 93% of normal | 863 |
| Gold (XAU/USD) | 53.4% | 51.2% | 1.6 | 117% of normal | 1,750 |
Every claim here reduces to a window on the clock or the calendar, so it can simply be checked. We compare the claimed window against the rest of the week on 23 years of hourly data across 19 markets, measuring two separate things the folklore tends to conflate: direction (does the window close higher more often?) via a two-proportion z-test, and movement (is the average move bigger or smaller?). A window can be perfectly ordinary in direction and genuinely quieter in movement — saying which is the entire point.
The bar here is |z| ≥ 2, deliberately more lenient than the |z| ≥ 3 this site applies to its own numbers: when testing someone else's claim, failing to reject it even at a lenient bar is the stronger statement.