← Stage Two Research

Bitcoin Crash Signal

September 11, 2026 · BTC at $76,676 (2026-09-10) · -38.6% from ATH · daily closes (Coin Metrics)

Warning flags: GREEN · 0 of 6 active. 6 of the gauges below are threshold “flags” (marked ⚑); the rest are context. The stage is GREEN below 2 active flags, YELLOW at 2, RED at 3+ — so a single flag can be lit (and show yellow on its own row) while the overall stage is still GREEN. One flag alone means little; two is a warning, three is rare and serious.
≥20% drop within 30 days12% · calmer than typical
Since 2018: 12% (4/34, CI 5-27%) · full history: 14% (11/76, CI 8-24%) · estimate level: trend × vol × drawdown-from-ATH. Typical (since 2018): 20% of windows. 30-day crashes are mostly shocks and leverage cascades; the 90-day number is the operational one.
≥20% drop within 90 days25% · calmer than typical
Since 2018: 25% (4/16, CI 10-49%) · full history: 28% (9/32, CI 16-45%) · estimate level: trend × vol × drawdown-from-ATH. Typical (since 2018): 37% of windows.

What owning Bitcoin means

A ≥20% drawdown begins within any given 90-day window ~37% of the time (since 2018). That base rate is the headline. On the current sample the regime cells mostly can't beat it — with confidence intervals this wide and only ~4 distinct bear markets on record, this page is a risk-literacy tool with a modest regime tilt, not a crash-timing engine. The one thing it can demonstrate at 95% confidence: a calm uptrend (above trend, low vol) crashes materially less over the next 30 days.

The gauges

Trend ⚑+9.6% vs 200-day averageGREEN
Don't fight the tape: sustained downtrends are where cascades live. (This ratio is the Mayer Multiple, as a percentage.)
Volatility ⚑32nd percentile (30-day realized, annualized)GREEN
Turbulence clusters. One BTC honesty note: LOW vol here often precedes big moves in EITHER direction — calm is compression, not safety.
Drawdown from all-time high-38.6%RED
BTC's strongest single state variable. Whether being deep below the ATH raises odds (mid-cascade) or lowers them (bottom zone) is settled by the table, not assumed — it depends on trend and vol.
Trend extension ⚑45th percentile stretch above the 200-day averageGREEN
How stretched price is vs its own history — blow-off tops are stretched tops. This is the wire that fires near peaks while everything else still looks healthy (it would have been on at the April and November 2021 highs). Percentile rather than a fixed line, because fixed thresholds like Mayer 2.4 decay — the Oct 2025 top never reached it.
Leverage (perp funding) ⚑+0.004%/8h avg, 45th percentileGREEN
What longs pay shorts. Crowded positive funding is the fuel for liquidation cascades — good 1-2 week warning, weak at 3 months. Percentile vs the past ~3 months (OKX BTC-USDT perp).
Stablecoin supply (USDT+USDC) ⚑$256B, -2.0% over 13 weeksYELLOW
The system's internal dry powder. Expanding supply is fuel; a shrinking supply is the native margin call — it contracted hard through the 2022 deleveraging. One of the few gauges here that measures liquidity rather than transformed price. Warning when contracting, red below -5% in 13 weeks.
Equity stress (VIX) ⚑16.5 (43rd percentile, +1.9 in 20 days)GREEN
BTC stopped being an island around 2020 — it now sells off with equities in risk-off regimes. Elevated and rising VIX is the macro wire.

Context — shapes how bad a drop gets, not when it starts (not in the odds)

Dollar (broad index)118.1, -0.8% in a monthGREEN
Dollar strength leads BTC weakness by ~a month, but the correlation is weak (~-0.2). Context, never a trigger.
Fed net liquidity5,857bn, +61bn over 4 weeks (adding)GREEN
Balance sheet minus RRP minus Treasury account. Crypto is the most liquidity-sensitive risk asset, but the sample is too short to put this in the math. Context.
Valuation (MVRV)1.44 (36th percentile, as of 2026-09-10)GREEN
Market value vs realized value — BTC's CAPE. A severity amplifier, not a timer, and decaying: it peaked ~3.5 at the Oct 2025 top vs ~7 in 2021 and ~9-10 in 2017 as ETFs moved coins off-chain. Percentile, never a fixed line.
Sentiment (Fear & Greed)56 (Greed)GREEN
Far better as a BOTTOM signal (extreme fear precedes recoveries) than a crash predictor — extreme greed is weak and noisy as a top signal. Context only.
How the odds are computed (and the full history table)

Same engine as the S&P version: two questions sort every day since 2010 into a box — above or below the 200-day average, and how choppy lately — refined by distance from the all-time high when there's enough history. The number is how often Bitcoin closed at least 20% below that day's price within the window. (Why 20%? A 10% dip begins within 90 days 51% of the time since 2018 — ambient noise, not a signal.) Three honesty upgrades for Bitcoin's short history: counts are de-overlapped (one crash can't pose as ninety correct calls), every number carries a 95% confidence range, and two eras are shown because Bitcoin's volatility has compressed — old frequencies overstate today's risk. Cells with too few independent starts say "insufficient data" instead of pretending.

One caveat the intervals don't fully capture: de-overlapping makes windows non-overlapping, but it can't make them independent — the 2018 and 2022 bears each spun off several sequential 20% legs from one macro episode, so a cell showing “7 hits” may really rest on two or three distinct events. With only ~4 bear markets on record, the true uncertainty is a touch wider than even these ranges show. Which is why the honest read of most cells is simply the base rate.

90 days · base rate 37% [23–54%]

No cell separates from the base rate at 95% confidence — on the current sample this horizon carries no usable regime signal, so read every cell as the base rate.
below trend, high vol= base rate
2018+: insufficient data (1 starts) · full: insufficient data (1 starts)
below trend, mid vol= base rate
2018+: 47% (7/15, CI 25-70%) · full: 50% (10/20, CI 30-70%)
below trend, low vol= base rate
2018+: 42% (8/19, CI 23-64%) · full: 45% (13/29, CI 28-62%)
above trend, high vol= base rate
2018+: insufficient data (1 starts) · full: insufficient data (3 starts)
above trend, mid vol= base rate
2018+: 33% (4/12, CI 14-61%) · full: 32% (8/25, CI 17-52%)
above trend, low vol ← today= base rate
2018+: 29% (6/21, CI 14-50%) · full: 31% (12/39, CI 19-46%)

30 days · base rate 20% [13–29%]

Statistically distinguishable — calmer than the base rate: above trend, low vol. Every other cell overlaps the base rate; treat it as the base rate.
below trend, high vol= base rate
2018+: insufficient data (1 starts) · full: insufficient data (1 starts)
below trend, mid vol= base rate
2018+: 15% (4/27, CI 6-32%) · full: 25% (9/36, CI 14-41%)
below trend, low vol= base rate
2018+: 19% (9/48, CI 10-32%) · full: 22% (16/72, CI 14-33%)
above trend, high vol= base rate
2018+: insufficient data (2 starts) · full: insufficient data (6 starts)
above trend, mid vol= base rate
2018+: 23% (5/22, CI 10-43%) · full: 22% (10/46, CI 12-36%)
above trend, low vol ← todaydistinct: calmer
2018+: 7% (4/54, CI 3-18%) · full: 12% (12/104, CI 7-19%)

Read this before acting

This catches regime fragility — downtrends, turbulence, crowded leverage. It cannot see shocks: FTX, the COVID cascade, and exchange or stablecoin breaks arrived as news, and the August 2024 unwind started from a calm tape. Against those, only position sizing and standing hedges protect you — never signals. A low reading means no internal fragility, not no risk.

And the deeper honesty: Bitcoin has had ~4 cycle tops — every "top indicator" fit to them has failed out of sample (Pi Cycle missed Nov 2021 and 2025 entirely; MVRV-Z peaked at ~3.5 in Oct 2025 vs ~7 in 2021). This tool never calls tops. It reports the measured frequency of the current regime with its sample size, and Bitcoin's sample is small — the confidence ranges ARE the product.