SPX-VIX Risk Computations Via Perturbed Optimal Transport

Imagine you are a master chef running a very complex restaurant. Your restaurant has two main menus: the SPX Menu (which lists prices for the overall stock market) and the VIX Menu (which lists prices for "fear" or market volatility).

In the real world, these two menus are deeply connected. If the stock market gets shaky, the "fear" menu changes price. If the fear menu changes, it tells you something about where the stock market might go next.

For a long time, chefs (traders) tried to predict how these menus would change together using rigid, pre-written recipes called Stochastic Models. These recipes assumed specific rules about how the world works (e.g., "volatility always moves like a bouncing ball"). The problem? The real world doesn't always follow the recipe. Sometimes the market behaves weirdly, and these rigid models break or give bad advice.

A few years ago, a new chef named Guyon invented a smarter way. Instead of guessing a recipe, he looked at the actual prices on both menus today and found the perfect "bridge" (a mathematical map) that connects them. This bridge is called Optimal Transport. It's like finding the most efficient way to move ingredients from the stock market shelf to the fear shelf without wasting a single drop.

The Problem:
This new bridge was great for pricing things today. But what happens tomorrow? If the stock market jumps up by 1%, how much does the fear menu change?
To find out, the old way was to stop the kitchen, throw away the current bridge, build a brand new bridge from scratch based on the new prices, and then measure the difference.

The Issue: This is incredibly slow. It's like rebuilding your entire restaurant every time a customer orders a different sandwich. It takes too long to calculate the "risk" (how much money you might lose).

The Solution: Perturbed Optimal Transport (POT)
This paper introduces a brilliant shortcut. Instead of rebuilding the bridge, the authors realized you can just nudge the existing bridge and see how it bends.

Here is how they do it, using simple analogies:

1. The "Rubber Sheet" Analogy (Linear Response)

Imagine the bridge connecting the stock market and fear is made of a giant, stretchy rubber sheet.

The Old Way: If you push the sheet, you measure the whole new shape by rebuilding it.
The New Way (Linear Response): The authors realized that for small pushes, the rubber sheet stretches in a predictable, straight line. They calculated the "stiffness" of the sheet (called the Fisher Information Matrix).
The Magic: Once you know how stiff the sheet is, you don't need to rebuild it. You just do a quick math calculation: "If I push here, it stretches exactly this much."
Result: You get the answer in milliseconds instead of hours. It's like knowing exactly how much a spring will stretch just by poking it, without needing to weigh the whole spring again.

2. The "Shadow Puppet" Analogy (Dimensional Reduction)

The bridge connects three things: The Stock Market Today ( $S_1$ ), The Fear Level ( $V$ ), and The Stock Market Tomorrow ( $S_2$ ). This is a 3D puzzle, which is hard to solve quickly.

The Insight: The authors noticed that the relationship between "Stock Today" and "Stock Tomorrow" (given the current Fear level) is actually very stable. It's like a shadow puppet show where the hand (the relationship) doesn't change, only the light (the prices) moves.
The Trick: They realized they could ignore the 3rd dimension (Stock Tomorrow) for the calculation. They just adjusted the 2D shadow (Stock Today + Fear).
Result: This turns a heavy 3D calculation into a light 2D one. It's like solving a puzzle by looking at its shadow on the wall instead of the whole 3D object. It's incredibly fast and almost as accurate as the full version.

3. The "Skew Stickiness" Rule (SSR)

In the real world, when the stock market moves, the "fear" smile doesn't just move up or down; it twists.

The Problem: The math bridge didn't know how to twist.
The Fix: The authors added a rule called Skew Stickiness Ratio (SSR). Think of this as a "twist rule" based on how the market actually behaves historically. It tells the bridge: "When the stock market moves left, the fear smile doesn't just slide; it rotates slightly."
Result: The bridge now reacts realistically to market shocks, capturing the messy, real-world behavior without needing a complex, rigid recipe.

Why Does This Matter? (The Hedging Backtest)

The authors didn't just do the math; they tested it in a simulation (a "backtest").

The Test: They created 50 fake portfolios of "fear" options and tried to protect (hedge) them against market crashes.
The Competitor: They compared their new "nudge" method against a standard, old-school "recipe" model.
The Winner: The new method was a clear winner. When the market got volatile (scary), the new method protected the money much better. The "old recipe" left holes in the umbrella, while the "nudge" method kept the portfolio dry.

Summary

This paper is about speed and accuracy.

Old Way: Rebuild the whole model every time the market moves. (Slow, rigid, often wrong).
New Way (POT): Nudge the existing model using math shortcuts (Linear Response and Dimensional Reduction). (Fast, flexible, and surprisingly accurate).

It's the difference between rebuilding a house every time the wind blows versus just knowing exactly how much the windows will rattle and reinforcing them accordingly. This allows traders to manage risk in real-time, keeping their portfolios safe even when the market gets wild.

Here is a detailed technical summary of the paper "SPX–VIX Risk Computations Via Perturbed Optimal Transport."

1. Problem Statement

The joint modeling of S&P 500 (SPX) and VIX derivatives is a critical challenge in equity volatility markets.

The Challenge: SPX options encode the distribution of future equity prices, while VIX options are derivatives on the square root of forward variance. Traditional parametric models (e.g., Heston, Bergomi, Rough Volatility) impose structural assumptions on volatility dynamics that may not align with observed market surfaces. Conversely, while model-free approaches like Martingale Optimal Transport (MOT) (specifically Guyon's entropic formulation) provide perfect static calibration to both SPX and VIX smiles, they lack a practical framework for risk management.
The Gap: In existing implementations, calculating risk sensitivities (Greeks) requires a "bump-and-recalibrate" approach: perturbing market data and re-solving the full, computationally expensive MOT calibration. This obscures the structural relationship between market shocks and risk and is too slow for real-time trading.
The Goal: Develop a framework to generate risk sensitivities for SPX-VIX derivatives directly from a calibrated MOT coupling without full recalibration, while incorporating empirical volatility dynamics (like Skew Stickiness).

2. Methodology: Perturbed Optimal Transport (POT)

The authors propose Perturbed Optimal Transport (POT), a framework that treats the calibrated MOT coupling as a point on a statistical manifold. By analyzing the local geometry of this manifold, they derive risk sensitivities analytically.

A. Mathematical Foundation

Entropic Projection: The base model solves a discrete entropic optimal transport problem to find a coupling $\mu^*$ between SPX spot ( $S_1$ ), VIX ( $V$ ), and future SPX ( $S_2$ ) that matches market marginals and satisfies financial constraints (Martingality and Variance Consistency).
Exponential Family & Fisher Information: The optimal coupling $\mu^*$ belongs to an exponential family (Gibbs distribution). The authors show that the Fisher Information Matrix (Hessian of the log-partition function) of this distribution governs the local geometry.
Linear Response (LR) System: Using the Implicit Function Theorem on the dual formulation, they derive that the first-order change in the coupling (and thus any payoff) due to marginal perturbations is linear and governed by the inverse Fisher Information Matrix.
$\Pi'(0) = \langle h, (H_0)^{-1} g_G \rangle$
Where $h$ is the market perturbation, $H_0$ is the Fisher matrix, and $g_G$ is the covariance of the payoff with sufficient statistics.

B. Incorporating Empirical Dynamics (SSR)

Standard MOT is static. To model how the VIX smile reacts to SPX moves, the authors introduce the Skew Stickiness Ratio (SSR) as a linear constraint.

They linearize the SSR dynamics (Bergomi's definition) to relate changes in the VIX future level to shifts in the VIX implied volatility skew.
This SSR relation is embedded as an additional linear constraint in the perturbation system, allowing the framework to propagate SPX shocks to the VIX smile dynamically without parametric stochastic volatility assumptions.

C. Dimensional Reduction (DR)

To further improve efficiency, the authors identify a Conditional Coupling Invariance structure.

Assumption: Under small perturbations, the conditional distribution of $S_2$ given $(S_1, V)$ remains fixed.
Result: The 3D transport problem ( $S_1, V, S_2$ ) reduces to a 2D entropic projection on $(S_1, V)$ . The conditional kernel is fixed from the base calibration, and only the marginal coupling on $(S_1, V)$ is updated.
Benefit: This allows for "mini-recalibration" that converges in 5-6 steps, capturing non-linear effects for finite shocks while avoiding the full 3D optimization.

3. Key Contributions

Linear Response (LR) System: Derivation of explicit analytic formulas for risk sensitivities using the Fisher Information Matrix of the calibrated Gibbs coupling, eliminating the need for full recalibration.
Linearized SSR Dynamics: Integration of Skew Stickiness Ratio as a linear constraint within the OT framework, enabling model-independent propagation of SPX shocks to the VIX volatility smile.
Dimensional Reduction (DR): Identification of a structural invariance that reduces the perturbed transport problem from 3D to 2D, preserving martingale and variance consistency while drastically lowering computational cost.
Numerical Validation: Demonstration that LR and DR methods produce sensitivities nearly identical to full recalibration but with orders of magnitude less computation time.
Hedging Backtests: Empirical evidence showing that hedges constructed using POT-derived sensitivities outperform those from benchmark stochastic local volatility models in terms of hedged P&L variance.

4. Experimental Results

The paper validates the framework through extensive numerical experiments:

Accuracy vs. Recalibration:
- VIX Futures: SPX Delta and Vega sensitivities computed via LR-POT matched the full recalibration benchmark with negligible error (e.g., Delta difference < 1%).
- VIX Options: Across a strip of strikes, the perturbation-based Greeks (Delta and Vega) closely tracked the recalibration results. Minor discrepancies occurred only in the wings where non-linearities are strongest, but the overall shape and magnitude were consistent.
Computational Efficiency:
- Base Calibration: ~348 seconds.
- Full Recalibration (per shock): ~370 seconds.
- DR-POT Method: ~18.5 seconds.
- LR-POT Method: ~5.7 seconds.
- Conclusion: The POT methods offer a 20x to 60x speedup over full recalibration.
Hedging Backtest (Jan 2024 – Feb 2026):
- 50 randomized VIX option portfolios were hedged using VIX futures and SPX instruments.
- Result: Hedges using POT-derived sensitivities consistently achieved lower hedged P&L variance compared to a Stochastic Local Volatility (SLV) benchmark.
- Volatility Regimes: The improvement was most pronounced during periods of high market volatility, demonstrating the framework's robustness in stress scenarios.

5. Significance

This paper establishes Perturbed Optimal Transport (POT) as a unified, model-independent framework for SPX-VIX markets.

Theoretical Advance: It bridges the gap between static calibration (MOT) and dynamic risk management by leveraging information geometry (Fisher Information) to define a local perturbation theory.
Practical Impact: It solves the "bump-and-recalibrate" bottleneck, enabling real-time risk calculation for complex SPX-VIX derivatives.
Market Consistency: By incorporating SSR dynamics directly into the transport constraints, it captures empirically observed volatility behaviors without resorting to potentially misspecified parametric stochastic volatility models.
Hedging Superiority: The framework provides more accurate risk sensitivities, leading to demonstrably better hedging performance in live market conditions compared to traditional industry standards.

In summary, the authors demonstrate that the geometry of the optimal transport solution itself contains the necessary information to propagate risk, offering a computationally efficient and financially consistent alternative to traditional parametric modeling.