# Does a quiet Asian session set up a big London move?

**Verdict: refuted** — the correlation is positive: volatility clusters.

Spearman rho = **+0.244** between the Asian range and the London/NY range of the same session day, over **169** session days (191 seen, the rest short of the 75 % minute floor). Permutation p = 0.0019 over 10,000 shuffles; the null's own spread is 0.079.

In raw USD rather than basis points: rho = +0.270, p = 0.0001. The normalisation does not decide it.

## The quintile table (basis points)

| Asian range | days | median Asian | median London | mean London |
|---|---:|---:|---:|---:|
| 1 (61–92 bp) | 34 | 76 | 129 | 152 |
| 2 (93–116 bp) | 34 | 104 | 122 | 140 |
| 3 (116–141 bp) | 33 | 126 | 140 | 166 |
| 4 (141–197 bp) | 34 | 168 | 155 | 169 |
| 5 (201–1005 bp) | 34 | 254 | 172 | 256 |

Median range: Asian 126 bp (57.59 USD), London/NY 143 bp (65.18 USD). The London/NY window is the wider of the two on 57.4 % of days.

## The positive control

A rank correlation of known size injected into this sample's own Asian ranks, 500 draws a level, judged by the verdict's own rule. That rule is two gates — |rho| >= 0.15 and p < 0.01 — and at 169 days the significance gate is the stricter of the two: the permutation null's 99th percentile of |rho| is 0.201. So the effective gate is |rho| >= 0.201, and that is what the control below is judged against.

| injected rho | certified | median measured |
|---:|---:|---:|
| 0.00 | 0.8 % | -0.009 |
| 0.05 | 2.6 % | +0.051 |
| 0.10 | 8.2 % | +0.107 |
| 0.15 | 26.8 % | +0.149 |
| 0.20 | 53.2 % | +0.206 |
| 0.25 | 73.4 % | +0.245 |
| 0.30 | 93.2 % | +0.297 |
| 0.40 | 100.0 % | +0.391 |

Smallest injected effect this test certifies at 95 % or better: **rho = 0.40**.
 At an injected rho of zero it fires 0.8 % of the time.


## Two things I looked at afterwards

Neither was pre-registered. Both are questions the table above raised, and they are reported as questions asked after seeing the answer.

**The quietest bucket breaks the trend.** From quintile 2 upwards the median London range rises with the Asian range, but quintile 1 — the quietest Asian sessions, the folklore's own case — has a median London range of 129 bp against quintile 2's 122 bp. That is the coiling story, in the only place it could hide. It is 34 days against 34, a difference of +7 bp, and a permutation test on the two buckets' medians gives p = 0.578. It does not survive being looked at.

**The two windows are not the same length**, so "which session is wider" is not a fair question as asked: London/NY is 540 minutes against the Asian window's 480. Raw, London/NY is the wider of the two on 57.4 % of days. Dividing each by the square root of its own length, which is roughly how a range grows with time, that share is 52.7 %.

Method, data and the kill rule: `README.md`, committed before the run. This is a test record, not a recommendation.