What if the 2008 collapse, SVB, and GLD's April drop all had the same structural signal weeks before price moved?
I've been tracking a structural coherence metric on financial assets - it measures the internal coherence of an asset rather than price direction. Outputs a number from 0-1 and a regime depends on that: stable, transition or breakdown. When score drops into transition/breakdown while price hasn't moved yet that's where the risk for the assets sits.
Not a price prediction. Not a buy/sell signal. Think of it as a structural risk layer - the kind of metric risk manager use to ask "is the condition of this asset changing in a way I haven't priced in yet".
**The GLD case - April 2026 (live tracking, not backtest):**
* Apr 17: price 445.93$ - the peak. Score declining at 0.743
* Apr 22: price 435.26$. Score crosses into **Transition** at 0.644
* Apr 24: score reaches **Breakdown** at 0.507
* Apr 27: score hits low at 0.480
* Apr 29: price hits low 415.44$. The signal fired 7 days before this.
* May 4: score recovers to **Stable** at 0.775 - while price is still on its low 414.71$
The score recovered before price did, That's the pattern: structure degrades first, recovers first.
Total move from when transition warning fired (Apr 22, 435.26$) to price low (Apr 29, 415.44$): −4.6% in 7 days. This was a live signal.
**Backtesting:**
**Lehman Brothers, 2008**: Score entered transition 49 days before the September 15 bankruptcy. During those 49 days the stock still had partial support.
**SVB, March 2023**: Score entered transition 17 days before the FDIC seizure on March 10. In the days before the signal fired, SVB's stock was still trading above $200. By the time most risk systems reacted, the bank was gone.
Both are backtests so I want to be careful about claims as you can always find patterns in history, but what makes them worth noting is that the mechanism is the same one that produced the live GLD signal and the timing structure is consistent: score degrades while price still looks stable, then price catches up.
**What the metric is and isn't:**
It isn't a prediction system, the score fires on structural conditions when internal coherence breaks down. Whether that breakdown leads to a price event depends on the cause. The GLD April breakdown was short and resolved. SVB's didn't.
It also doesn't fire on external shocks. COVID March 2020 is a good example - the crash was exogenous, not a structural deterioration that built up inside the asset. The score stayed relatively stable through that period because the market structure itself wasn't degrading beforehand; it was hit from outside. That's a meaningful distinction: the score reflects internal structural health, not "bad things are coming from the world."
It won't fire every time something falls, and it isn't designed to. It fires when the internal coherence of an asset starts degrading before price reflects it. That's a narrower and more specific claim than "predicts crashes."