The Divorce
Detector
Everyone quotes the Bitcoin–Nasdaq correlation. Almost nobody watches for the moment it breaks. This page runs a regime machine over the rolling correlation — married, drifting, divorced — and keeps a dated ledger of every separation, with the market conditions measured at each one.
Updated 08 September 2026 · 19:16 UTC · Prices as of 08 September 2026 · Source: Yahoo Finance (BTC-USD, QQQ, ^VIX)Current status: DRIFTING
The gauge reads the 60-day rolling correlation of daily returns. Above 0.45 the two are married; below 0.10 they are divorced; the band between is the drift. Hysteresis means the label needs real evidence to change — one odd week does not file the papers.
DRIFTING
DRIFTING for 13 trading days
State changes require crossing both an entry and an exit threshold (married ≥ 0.45, exits < 0.35; divorced ≤ 0.10, exits > 0.25). Correlation measured on paired daily log returns over Nasdaq trading sessions only.
30-day+0.12Fast read — reacts first, lies often.
60-day+0.26The regime input. This is the gauge.
90-day+0.33The slow trend underneath.
Breaks on record1Divorced episodes ≥ 3 days in 3 years.
The relationship, day by day
Three windows of the same rolling correlation. Shaded bands mark the divorced episodes — watch how the fast 30-day line leads the regime machine into and out of each break. The lower panel is the VIX over the same stretch.
30-day60-day90-dayDivorced episode
Every break on record
Each row is a divorced episode: contiguous trading days where the 60-day correlation fell through 0.10 and stayed low. Conditions are measured, not narrated — VIX at the break, VIX the week before, and what each asset did during the separation.
| Break start | Reconciled | Days apart | Lowest corr | VIX at start | VIX week prior | BTC during | QQQ during |
|---|---|---|---|---|---|---|---|
| 2023-12-04 | 2024-05-22 | 171 | -0.12 | 13.1 | 12.8 | +58.0% | +16.0% |
Do breaks telegraph themselves?
We tested the simplest suspect: equity volatility. For every recorded break we took the average VIX in the week before it started, and compared it with every other week in the sample. The numbers, not a story:
VIX, week before breaks12.8Mean weekly VIX in the 7 days before each divorced episode began.
VIX, all other weeks17.3Mean weekly VIX across the rest of the 3-year sample.
Breaks preceded by VIX > 200%Share of break episodes whose prior week averaged above 20.
Methodology & limitations
Method. Daily closes for BTC-USD, QQQ and ^VIX from Yahoo Finance, 3-year window, export timestamped 08 September 2026 · 19:16 UTC. Bitcoin's close is carried onto Nasdaq trading days (never the reverse), daily log returns are paired, and Pearson correlation rolls over 30/60/90-day windows. The regime machine reads the 60-day window with hysteresis (enter married ≥ 0.45, exit < 0.35; enter divorced ≤ 0.10, exit > 0.25). An episode is a contiguous divorced run of at least 3 trading days.
Limits. Three years is one macro cycle, not a law of nature. QQQ is the tradable Nasdaq-100 proxy, not the Composite. Correlation of daily returns misses intraday coupling and weekend crypto moves. The VIX lead test is univariate and descriptive — a difference in means is not a forecasting model, and 1 episodes is a small sample. Missing data is never treated as zero; if a feed fails, the page says so instead of printing a number.
Limits. Three years is one macro cycle, not a law of nature. QQQ is the tradable Nasdaq-100 proxy, not the Composite. Correlation of daily returns misses intraday coupling and weekend crypto moves. The VIX lead test is univariate and descriptive — a difference in means is not a forecasting model, and 1 episodes is a small sample. Missing data is never treated as zero; if a feed fails, the page says so instead of printing a number.
