Book 3 · The Mini-Beast · Chapter 15 of 44
Market · Critical Volatility Threshold
σ* is κ* in different units.
σ ≥ σ* ⇒ regime shift
Orientationσ* Is κ* in Different Units
The critical volatility threshold is not a new object. It is the critical focal curvature, computed on the Fisher manifold and expressed in volatility units. This chapter states it, gives the confirmed crossings, and sets out what a clean backtest would have to look like.
The Two Confirmed CrossingsWhat Was Observed
6 May 2010 — the Flash Crash
14:32 EST — κ reaches 0.13, the threshold approach. 14:37 EST — κ crosses κ* = 0.15, fold activation. 14:47 EST — Dow −9.2%, 998 points, orbit completion. 15:10 EST — recovery, the U phase. Total duration 36 minutes.
The dm³ fold-time prediction is τ_fold = π/ω, which at the daily ω = 0.28 rad/day gives ≈ 11.2 days and at an intraday ω_intra gives ≈ 22 minutes. Observed: 36 minutes — within a factor of two of the intraday extrapolation. That is agreement worth reporting and also worth being uneasy about: a factor of two is not a sharp test, and the intraday ω is not independently pinned.
16 March 2020 — the COVID crash
σ crossed 0.17, near the top of the band, ahead of a −12% intraday move. The second confirmed crossing matters more than the first, because the mechanism was different — an exogenous macro shock rather than an endogenous liquidity cascade — and the threshold behaved the same way.
Three testable predictions
- κ* threshold. Volatility surface curvature must reach 0.12–0.18 within 10 minutes before each flash crash. Backtestable on TAQ data.
- Fractal dimension. Intraday price increments must exhibit d_f in the range 1.7–1.9 during regime shifts.
- Mean-reversion rate. Post-crash volatility decay must fit the μ_max = −0.67 ± 0.08 exponential envelope.
Backtest DesignHow Not to Fool Yourself
Two confirmed crossings are two data points. A threshold claim needs the complementary evidence, and the design has to be fixed before the data is touched.
- Define the estimator first. κ(t) = |d²σ/dt²| / |dσ/dt| is the curvature proxy used in the guided exploration. Fix the smoothing window and the volatility estimator in advance and do not revisit them.
- Count the false positives. Every crossing of 0.12–0.18 that was not followed by a regime shift is evidence against. The confirmed-crossings framing hides this; a real test reports the full contingency table.
- Report the base rate. Over a 20-year horizon, how often does κ enter the band at all? If the answer is ‘most weeks’, the threshold has no predictive content regardless of how well the two famous events fit.
- Pre-register the horizon. The open prediction from the pedagogy chapter is whether the dm³ regime-shift signal can beat buy-and-hold on 20-year horizons. That is the only version of the question that cannot be won by selection.
A student who runs this honestly and finds the threshold has no edge has produced a publishable negative result and has done the framework a service. The point of stating σ* numerically is to make that outcome possible.
BridgesWhere This Connects
- WP-28 · Auditing the Circadian Trader ClaimA market-adjacent threshold in this corpus with no derivation, dataset or citation. The failure mode this chapter is designed to avoid.
- WP-30 · How to Audit a Mathematical ClaimPhantom code, phantom citation, phantom row — the three shapes of nothing to check for in any threshold claim.
- WP-31 · The Calibration PipelineThe four-stage method the backtest design in this chapter is an instance of.
- WP-85 · The Missing AnchorWhat happens when a number has no observable underneath it.