Option pricing under fast-varying and rough stochastic volatility

Josselin Garnier, Knut SølnaAnnals 2018 · DOI · arXiv

Abstract

Recent empirical evidence suggests that asset price volatility is "rough," characterized by a Hurst exponent $H < 1/2$, which contradicts the standard Markovian assumptions ($H = 1/2$) used in many financial models. This paper provides a mathematical reconciliation between these two regimes. We demonstrate that in the limit of fast mean reversion, the first-order correction to option prices is independent of the Hurst exponent. This result explains the continued efficacy of Markovian models in practical calibration despite the underlying roughness of market volatility.

1. Introduction: The Rough Volatility Paradox

In quantitative finance, the dynamics of asset volatility are central to option pricing. For decades, standard models have assumed that volatility follows a Markov process, such as an Ornstein-Uhlenbeck process, where the future state depends only on the current value. These models are favored for their mathematical tractability and their ability to fit the observed "volatility smile."

However, empirical studies (e.g., Gatheral et al., 2018) have revealed that volatility paths are significantly more jagged than Brownian motion would suggest. This "roughness" is quantified by the Hurst exponent $H$; while Markovian models assume $H = 1/2$, market data often indicates $H \approx 0.1$. This discrepancy poses a theoretical challenge: if volatility is non-Markovian and rough, why do smooth Markovian models remain so effective for pricing and hedging? The authors tackle this by analyzing the asymptotic behavior of option prices under a fast-mean-reverting rough volatility regime.

2. Modeling Framework

The authors employ a fractional Ornstein-Uhlenbeck (fOU) process to represent the stochastic volatility factor. The fOU process generalizes the standard OU process by incorporating a fractional Brownian motion, allowing for the adjustment of the Hurst exponent $H$:

  • $H = 1/2$: The process is standard Markovian Brownian motion.
  • $H < 1/2$: The process is "rough" or jagged, exhibiting short-range correlations.

The primary objective is to analyze the asymptotic behavior of option prices when the volatility's mean-reversion time is very small relative to the option's time to maturity.

3. Methodology: The Martingale Approach

To derive option prices, the authors utilize a martingale method to perform an asymptotic expansion. They focus on the regime where volatility varies rapidly. By analyzing the covariation between the asset price and volatility fluctuations—accounting for the "leverage effect" (the negative correlation between returns and volatility)—they identify the leading-order corrections to the Black-Scholes price.

The core of the derivation involves showing how the integrated volatility behaves over the option's lifespan. In the fast-mean-reversion limit, the specific path-dependency of the rough process becomes secondary to its long-term distributional properties.

4. Key Insights: The Universality of Fast Mean Reversion

The paper’s most significant contribution is the discovery of a "universality" result in the fast-varying regime:

4.1 Hurst Independence

The authors prove that for fast mean reversion, the Hurst exponent $H$ disappears from the leading-order price correction. This means that as long as volatility reverts to its mean quickly enough, the specific "roughness" of the path does not alter the first-order impact on option prices. From the perspective of an option with a standard maturity, the rough process is effectively "averaged out," behaving much like a standard Markovian process.

4.2 Reconciling Theory and Practice

This result provides a rigorous justification for the industry's reliance on Markovian-based calibration tools. Even if the underlying market dynamics are rough, the fast-varying nature of volatility ensures that Markovian approximations remain robust for pricing purposes.

4.3 The Role of the Characteristic Exponent

The authors introduce a characteristic term structure exponent, $\zeta(H)$. They demonstrate that for fast mean reversion and $H \le 1/2$, $\zeta(H) = 0$. This implies that the term structure of the volatility skew remains flat at leading order, consistent with Markovian results and contrary to what is observed in slow-varying rough regimes.

5. Key Results

  • Proposition 1: The authors derive an analytical expression for the first-order corrected option price. They show that this correction is identical in form to the one derived for fast-mean-reverting Markovian models (e.g., those by Fouque et al.).
  • Implied Volatility Skew: The resulting formula for implied volatility (the "skew") matches the classic results used in the industry, confirming that the "roughness" does not distort the standard linear-in-log-strike behavior of the volatility smile at leading order.

6. Limitations and Future Directions

While this paper bridges a significant gap, it acknowledges several boundaries:

  • First-Order Approximation: The "Hurst independence" applies to the leading-order correction. Higher-order terms in the asymptotic expansion may still reveal the influence of $H$.
  • Regime Sensitivity: The reconciliation is specific to the fast mean-reversion regime. In markets where volatility varies slowly, roughness would dominate, requiring more complex fractional models.
  • Stationarity Assumptions: The derivation assumes a stationary volatility process. While the authors discuss non-stationary variants, the primary results rely on the process having reached its steady state.

7. Conclusion

Garnier and Sølna demonstrate that the "roughness" of volatility is not necessarily at odds with the "smoothness" of pricing models. By focusing on the timescale of mean reversion, they show that the jaggedness of volatility paths becomes a second-order effect in the fast-varying limit, allowing practitioners to maintain the use of tractable Markovian frameworks while acknowledging the empirical reality of rough markets.

Summary updated

BibTeX
@article{garnier2018option,
  author = {Josselin Garnier and Knut Sølna},
  title = {{Option pricing under fast-varying and rough stochastic volatility}},
  journal = {Annals},
  year = {2018},
  doi = {10.1007/s10436-018-0325-4},
}

Loading related content...

Figure