Bounds for VIX futures given S&P 500 smiles

Julien Guyon, Romain Menegaux, Marcel NutzFinance Stochastics 2016 · DOI · OpenReview

Abstract

We derive sharp, model-free bounds for the prices of VIX futures by utilizing the full information contained in S&P 500 option smiles. By formulating the problem as a sub/superreplication task involving the S&P 500 index, its vanilla options, and forward-starting log-contracts, we establish a dual relationship with a martingale optimal transport problem. While classical bounds for VIX futures often rely solely on calendar spreads of log-contracts—frequently resulting in uninformative lower bounds—our approach incorporates the entire volatility smile. We introduce a family of tractable Functionally Generated Portfolios (FGPs) that significantly tighten these bounds. Numerical experiments using both SABR-generated smiles and empirical market data demonstrate that while classical upper bounds are often near-optimal, the lower bounds can be improved dramatically, providing a more accurate "floor" for VIX future pricing.

Introduction

The VIX Index, widely recognized as the market's "fear gauge," reflects the expected 30-day volatility of the S&P 500. Despite the high liquidity of VIX futures, their valuation presents a unique challenge: the VIX is not a directly tradable asset. Unlike standard equity options, where the underlying can be held to hedge a position, VIX futures must be hedged using the underlying S&P 500 index and its derivatives.

Traditional pricing models often rely on specific stochastic volatility assumptions. However, these models carry model risk. This paper addresses the problem by seeking "model-free" bounds—price ranges that must hold under any arbitrage-free model consistent with observed S&P 500 option prices (the "smile").

The Pricing Problem and Classical Limitations

The VIX is defined based on the prices of S&P 500 options, specifically through a portfolio of out-of-the-money options that replicate a log-contract. The classical approach to bounding VIX futures involves a calendar spread of these log-contracts. However, these classical bounds are often remarkably loose. In many market scenarios, the classical lower bound for a VIX future is mathematically zero, offering no practical utility for risk management or pricing.

The primary objective of this research is to tighten these bounds by leveraging the full distributional information available in the S&P 500 option smile across all strikes, rather than just the aggregate log-contract value.

Methodology: Duality and Optimal Transport

The authors frame the bounding problem through the lens of sub/superreplication. To find the upper bound, one seeks the cheapest portfolio of tradable assets—the S&P 500, its vanilla options, and forward-starting log-contracts—that is guaranteed to pay out at least as much as the VIX future at maturity.

Linear Programming Duality

A key theoretical contribution is the proof that this hedging problem is dual to a specialized Martingale Optimal Transport (MOT) problem. Specifically, the cost of the optimal superreplicating portfolio is equal to the maximum expected value of the VIX across all risk-neutral measures consistent with the observed S&P 500 smiles. The authors prove that the duality gap is zero, meaning the bounds derived are "sharp" and cannot be improved without further assumptions.

Functionally Generated Portfolios (FGPs)

Solving the full MOT problem numerically can be computationally intensive. To provide a more practical tool, the authors introduce Functionally Generated Portfolios. These portfolios are determined by a single concave or convex function. FGPs are significantly easier to compute than the full linear programming (LP) solution but are far more sophisticated than classical bounds. They allow for a tractable way to capture the "tail risk" of the S&P 500 distribution, which is essential for establishing a non-zero lower bound.

Numerical Results

The framework was tested using both synthetic data from the SABR model and historical market data from May 2016.

SABR Model Analysis

In a 2-month maturity scenario under the SABR model, the improvement in the lower bound was particularly stark:

  • Classical Lower Bound: 0%
  • FGP Method: ~5.2%
  • Full LP Solver: 7.2%
  • Upper Bound: 22.8% (Both classical and LP methods yielded similar results).

The jump from 0% to over 7% demonstrates that the full smile contains critical information that the classical log-contract approach ignores.

Market Data Application

Using S&P 500 option data from May 5, 2016, the authors found:

  • The lower bound improved from a trivial 0% to a meaningful 8.4%.
  • The upper bound remained stable at 16.7%.

These results indicate that while the market's existing methods for capping VIX prices (the upper bound) are relatively efficient, the "floor" or lower bound was previously significantly underestimated by model-free theory.

Conclusion

This research provides a robust, model-free framework for pricing VIX futures. By incorporating the full S&P 500 smile and utilizing the duality between hedging and martingale optimal transport, the authors have turned a previously uninformative lower bound into a practical tool for market participants. The introduction of Functionally Generated Portfolios offers a balance between computational tractability and pricing accuracy, ensuring that these sharp bounds can be applied in real-time financial environments.

Summary updated

Abstract

We derive sharp bounds for the prices of VIX futures using the full information of S&P 500 smiles. To that end, we formulate the model-free sub/superreplication of the VIX by trading in the S&P 500 and its vanilla options as well as the forward-starting log-contracts. A dual problem of minimizing/maximizing certain risk-neutral expectations is introduced and shown to yield the same value.The classical bounds for VIX futures given the smiles only use a calendar spread of log-contracts on the S&P 500. We analyze for which smiles the classical bounds are sharp and how they can be improved when they are not. In particular, we introduce a family of functionally generated portfolios which often improves the classical bounds while still being tractable; more precisely, they are determined by a single concave/convex function on the line. Numerical experiments on market data and SABR smiles show that the classical lower bound can be improved dramatically, whereas the upper bound is often close to optimal.
BibTeX
@inproceedings{guyon2016bounds,
  author = {Julien Guyon and Romain Menegaux and Marcel Nutz},
  title = {{Bounds for VIX futures given S\&P 500 smiles}},
  booktitle = {Finance Stochastics},
  year = {2016},
  doi = {10.2139/ssrn.2840890},
}
Figure