← Research
Markets
R Crosses 1/φ — when correlations lock, the crash is coming
JIM’S OVERSIMPLIFICATION
A crash is a conga line hitting a wall. Everyone locked step, nobody looking where they’re going, and when the front person stops, the whole line piles up. The math is identical to soldiers breaking a bridge by marching in sync. The bridge doesn’t care how heavy you are. It cares how synchronized you are.
A healthy stock market is a mosh pit. Everyone doing their own thing, bouncing off each other randomly. Some guy’s going left, someone else is going right, nobody’s in sync. That’s fine. That’s normal.
A crash is when the mosh pit turns into a conga line. Suddenly everybody’s moving the same direction at the same time. Fear replaces fundamentals. Nobody’s reading earnings reports anymore — they’re watching what everyone else is doing. The whole market becomes one animal.
Of course it crashes. That’s what happens when a million independent decision-makers all decide to make the same decision at the same time. It’s the same physics as a seizure — neurons that normally fire independently all lock together. The brain crashes for the exact same reason the market does. Too much synchronization.
CrashWatch doesn’t predict crashes. It watches correlations. When stocks that normally have nothing to do with each other start moving together across every time window — 10-day, 20-day, 50-day — that’s the conga line forming. The crash hasn’t happened yet. But the independence is gone.
Same thing happens with fireflies syncing their flashes. Soldiers accidentally breaking a bridge by marching in step. A crowd suddenly stampeding. Different systems, same failure mode: too much coupling.
Honest limit: this is simulated data, not a backtest on real crashes. We built it to demonstrate the principle, not to trade on. Do not bet money on this. The insight is real — phase transitions are physics. The application is unvalidated.
Correlations lock. The shape contracts. R crosses the threshold. Crash.
THE INSIGHT
A crash is a phase transition.
Normal market: stocks move independently. Low correlation.
Each stock responds to its own fundamentals, its own news.
The system is decoupled. R is low.
Pre-crash: stocks start moving together. Correlations rise.
Fear replaces fundamentals. Herd behavior takes over.
The system is locking. R approaches 1.
Crash: all stocks move as one. Total synchronization.
Individual identity gone. One giant coupled oscillator.
Same physics as a seizure — neurons that normally fire
independently all lock together. The brain crashes for the
same reason the market does: R crosses the threshold.
CRASHWATCH
CrashWatch monitors rolling correlation windows across multiple time scales. When correlations lock across all windows, the regime has changed.
From the pip package (begump):
Rolling windows: 10-day, 20-day, 50-day
Metric: mean pairwise correlation across stocks
Threshold: R approaching 1 = herd behavior = crash imminent
What it detects:
Not the crash itself. The REGIME CHANGE before the crash.
Correlations rising across all time windows simultaneously
= the market is losing its independence = phase transition incoming.
SIMULATED DEMONSTRATION
200 days normal trading (low correlation):
Stocks move independently. R stays low.
Each asset follows its own trajectory.
52 days pre-crash (correlations rising):
Stocks begin moving together. Fear contagion.
CrashWatch detects:
R(10d) = 2.29
R(50d) = 4.46
Regime: CORRELATED
Multi-scale correlation spike = all time windows agree
= the decoupling has broken = regime transition detected.
THE PHASE TRANSITION
Decoupled regime: stocks are independent oscillators. Each responds to its own signal. Low correlation. Healthy market.
Transition: correlations rise. Individual signals get drowned by collective fear. Coupling strengthens. R climbs toward 1.
Locked regime: all stocks move together. One signal dominates. The market is a single oscillator. This is the crash.
The transition from decoupled to locked is the same math as:
• Neurons locking in a seizure
• Atoms locking in a crystal (freezing)
• Fireflies synchronizing their flashes
• A crowd suddenly stampeding
Different systems. Same coupling threshold. Same R.
HONEST LIMITS
This is simulated data, not a backtest on real market crashes.
The demonstration above uses synthetic price series, not historical data.
No S&P 500 crash was backtested. No real market data was used.
Small-cap K=0.607 was KILLED.
An earlier claim about small-cap coupling came from simulation,
not real data. It was identified and removed.
CrashWatch detects correlation regime, not crash timing.
It tells you correlations are locked. It does not tell you
when the crash will happen or how deep it will go.
Regime detection ≠ prediction.
This is a demonstration of principle, not a trading system.
Do not use this to make investment decisions.
Do not bet money on correlation regime detection.
The insight is physical (phase transitions are real),
the application is unvalidated.
What would make this real:
Backtest on 2008, 2020, 2022 crashes with real price data
False positive rate measurement (how often does R spike without a crash?)
Sector-level analysis (do some sectors lock before others?)
Comparison to VIX and other established volatility measures
This is computational research, not financial advice. CrashWatch is a research tool demonstrating phase transition principles. It has not been validated on real market data. Do not use it for trading decisions.