March 16, 2020 — A Stability Autopsy

I. Introduction

On March 16, 2020, the S&P 500 fell 11.98%, recording its largest single-day decline since 1987. The VIX index closed at 82.69. Liquidity fractured, spreads widened dramatically, and the mechanics of risk transfer were tested at their structural limits.

Less visible, however, was a secondary shift occurring beneath the surface: the numerical topology of the options market inverted.

Modern derivatives infrastructure relies on iterative numerical solvers—most commonly Newton-type methods—to invert pricing models and extract implied volatility. These algorithms operate under assumptions of local stability. In ordinary conditions, those assumptions hold. On March 16, 2020, the surface entered regimes where classical stability bounds were no longer dominant.

This report analyzes the complete SPY options surface from that trading session (10,344 contracts). The data reveals that market stress is not merely a phenomenon of price volatility; it is an event of computational regime inversion. Rather than relying on observed convergence alone, we classify each contract using a computable local stability bound derived from fixed-point iteration theory.

Figure 1. Cross-sectional regime distribution within the 15–45 DTE lens. The Stable basin contracts materially during the March 16 regime inversion, while Oscillatory and Singular regions expand.

Data Source: Historical SPY options chain (March 16, 2020), end-of-day snapshot
Methodology: Stability classification based on local fixed-point bounds applied to Newton iteration

I. The Dataset

To isolate structural behavior, we replayed the end-of-day SPY surface through a standard, un-governed Newton-Raphson solver.

• Underlying: SPY (S&P 500 ETF)
• Date: March 16, 2020
• Contracts Analyzed: 10,344 (Full surface)
• Moneyness Range ($S/K$): Broad cross-section of listed strikes
• Solver Baseline: Standard Newton iteration (Tolerance: $1e^{-8}$)

By removing damping heuristics and fallback rules typically embedded in production systems, we expose the raw mathematical topography of the market under extreme volatility.

II. Regime Classification

Root-finding methods do not simply succeed or fail. They operate within topological regimes determined by local derivatives.

By evaluating the interaction between pricing error, Vega (first derivative), and curvature (Vomma), each contract can be classified into one of four regimes:

Figure 2. Local topology classification across moneyness under stressed conditions (15–45 DTE lens). Contracts migrate into Oscillatory and Divergent regimes as curvature interaction intensifies.

1. Stable — Local curvature supports monotonic convergence.
2. Oscillatory — Alternating overshoot driven by curvature concentration.
3. Divergent — Linear approximation breaks down; iteration leaves the basin of attraction.
4. Singular — Vega approaches machine precision thresholds; local linearization becomes unreliable.

The March 16 distribution was as follows:

• Stable: 35.7%
• Oscillatory: 42.0%
• Singular: 14.6%
• Divergent: 5.8%

In calm market conditions, the Stable regime typically dominates approximately 70% of the surface. On March 16, that basin was effectively halved.

More than 64% of contracts existed outside classical stability bounds.

The market did not merely become volatile; its numerical structure reorganized.

III. The Oscillatory Corridor

The largest expansion occurred in the Oscillatory regime.

In these contracts, Newton’s method often still converges. From a superficial perspective, the solver “works.” Yet the trajectory is turbulent.

The solver alternates between large overshoots and undershoots before eventually settling. Any premature termination — such as a hard iteration cap — produces materially distorted implied volatilities.

More subtly, even when price convergence is achieved, the intermediate instability propagates into the Greeks. Delta, Gamma, and Vega sampled mid-oscillation exhibit severe “Greek chatter,” producing unstable hedge ratios.

Convergence alone is not a sufficient indicator of structural reliability.

Iteration trajectory for a representative Oscillatory-regime contract. The standard Newton method converges but exhibits large intermediate Delta deviations before stabilization.

IV. The Singular Wall

A second expansion occurred in the Singular regime.

When Vega collapses toward machine precision, the Newton denominator becomes numerically fragile. The solver is no longer navigating a curve; it approaches a flat boundary.

The crash forced a large number of previously near-the-money calls deep into out-of-the-money territory. As contracts repriced abruptly, 14.6% of the surface entered a regime where linear approximation loses structural reliability.

In such conditions, fallback mechanisms are not optional—they are mathematically necessary.

V. Divergent Clusters

The most violent behavior appears in the Divergent regime.

Here, curvature overwhelms the linear assumption entirely. A single update step may imply volatility changes of several hundred or thousand percent, producing overflow or undefined outputs.

These failures are not random.

A computable local stability bound, evaluated prior to execution, flagged these contracts before divergence occurred.

The instability was structural and detectable.

VI. Terminal Stability vs. Path Stress

Returning a number is not sufficient. Numerical systems must quantify how that number was obtained.

Two complementary diagnostics were evaluated:

1. Terminal Confidence — A measure of local stability at convergence.
2. Path Confidence — A trajectory-based metric penalizing curvature stress and excessive step exposure encountered during iteration.

The distinction is critical.

Calm Baseline (60 Lowest VIX Days):

• Average Daily Terminal Confidence: ~83.41%
• Average Daily Path Confidence: ~77.91%
• Stable Regime: ~70.72%

March 16, 2020 (Crash):

• Average Terminal Confidence: 86.46%
• Average Path Confidence: 69.52%
• Stable Regime: 37.49%

At first glance, Terminal Confidence appears stable — slightly elevated during the crash.

This is not a contradiction. Elevated volatility increases option premiums and Vega, improving the informational density available to the solver. Terminal roots remain locally well-defined.

The crisis was not informational.

It was topological.

Path Confidence tells the deeper story.

While final convergence remained intact, trajectory stress intensified dramatically. The Oscillatory regime expanded to 42% of the surface, and Divergent clusters emerged. The solver often survived—but under severe curvature turbulence.

The March 2020 crisis did not degrade terminal stability; it amplified trajectory stress.

VII. Interpretation

The data yields a structural insight:

Market stress is not only price volatility. It is regime inversion within the numerical topology of the surface.

In calm markets:

• Stability basins dominate
• Path stress is moderate
• Singular regimes are negligible

In crash conditions:

• Stable regions contract sharply.
• Oscillatory corridors expand.
• Singular walls emerge.
• Path stress rises materially—even when final convergence succeeds.

Observed convergence alone is insufficient as a measure of computational integrity.

Financial infrastructure should not merely verify that a solver returns a value. It must measure the structural conditions under which that value was obtained.

Conclusion

The events of March 2020 are typically described in terms of price collapse, liquidity gaps, and macroeconomic shock.

Less examined is the stability of the numerical systems translating those prices into implied volatility surfaces.

If modern derivatives markets depend implicitly on iterative computation, then computational integrity is itself a dimension of market structure — and therefore a dimension of market risk.

During the COVID crash, the market’s numerical surface did not simply become volatile. Its stability basins fractured, its curvature intensified, and its trajectory stress amplified.

The surface converged.

But it converged under strain.

In the next essay, we examine the theoretical foundation of these diagnostics — why local stability bounds remain predictive under finite precision, and how regime-aware interpretation transforms solver behavior from blind iteration into governed computation.