Skew Stickiness Ratio in Option Smile Dynamics

• SSR is defined as the ratio between the ATM implied volatility elasticity and the local skew, capturing the interplay between spot returns and volatility changes.
• It is computed using stochastic volatility frameworks and model-free formulations, providing a calibration-agnostic metric for risk and hedge management.
• Empirical and asymptotic analyses reveal SSR’s distinct limits across models, distinguishing between sticky-strike, sticky-delta, and rough volatility regimes.

The skew stickiness ratio (SSR) is a dimensionless statistic central to the analysis of option smile dynamics, quantifying the interplay between spot returns and implied volatility skews. Mathematically and empirically, SSR encodes how the at-the-money (ATM) volatility responds to spot or forward changes relative to the local slope ("skew") of the implied volatility surface. Its precise form and asymptotic limits, as well as its sensitivity to volatility path dependencies, have been systematically formalized across stochastic volatility frameworks, model-free representations, pathwise diffusions, and modern risk management platforms. SSR is fundamental for both theoretical model discrimination and robust risk and hedge management in derivative markets.

1. Formal Definitions and Theoretical Foundations

Given an underlying asset price process $S_t$ and Black–Scholes implied volatility $\sigma_{,t}(k,T)$ with log-strike $k=\log K/S_t$ and expiration $T>t$, SSR is defined as the ratio between the ATM implied-vol elasticity (with respect to spot moves) and the implied-vol skew: $\mathcal{R}_t(T) = \frac{\beta_t(T)}{S_t(T)}$ where

$S_t(T) = \partial_k \sigma_{,t}(k,T)\big|_{k=0}$

is the ATM skew and $\beta_t(T)$ characterizes the linear response: $$\delta\sigma_{,t}(T) = \beta_t(T)\,\delta X_t + \text{noise}, \quad \delta X_t = \frac{\delta S_t}{S_t}$$ Alternative but equivalent forms of SSR appear in terms of the covariation between ATM implied volatility and log-spot,

$$\mathcal{R}_{t,T} = \frac{1}{\mathcal{S}_{t,T}} \frac{\partial_t\langle\widehat{\sigma}^T,\log S\rangle_t}{\partial_t\langle\log S\rangle_t}$$

with $\mathcal{S}_{t,T}$ the ATM forward skew and angular brackets denoting quadratic covariation (Jaber et al., 18 Mar 2025, Friz et al., 2024). In the geometric Brownian motion setting (no stochastic volatility), SSR reduces to canonical values reflecting "sticky-strike" (SSR $\sigma_{,t}(k,T)$0) or "sticky-delta" (SSR $\sigma_{,t}(k,T)$1) regimes (Che et al., 11 Mar 2026). Model-free calculations and pathwise constructions for sticky/skew threshold diffusions further equate SSR to the ratio of the process's skewness parameter to its stickiness parameter, $\sigma_{,t}(k,T)$2 (Anagnostakis et al., 2024).

2. Model-Free Characteristic Function Formulation

A principal result is the model-free SSR formula in terms of the characteristic function $\sigma_{,t}(k,T)$3 of $\sigma_{,t}(k,T)$4 under the pricing measure: $\sigma_{,t}(k,T)$5 where $\sigma_{,t}(k,T)$6 and $\sigma_{,t}(k,T)$7 denotes a functional (Volterra-type) derivative with respect to the forward-variance curve $\sigma_{,t}(k,T)$8. This representation, first derived using Fourier integral techniques and functional differentiation, yields a practical, calibration-agnostic SSR value once the characteristic function is specified or observed—whether extracted from market data or derived from a volatility model (Friz et al., 2024). In stochastic volatility models with affine forward variance (AFV), this formula simplifies due to affine structure, leading to tractable Ricatti-type convolution solutions for the key objects.

3. Asymptotic Behavior: Short-Time and Long-Time Limits

In the limit of vanishing maturity $\sigma_{,t}(k,T)$9, SSR exhibits universal limiting behavior determined by the regularity of the kernel $k=\log K/S_t$0 in the volatility process specification: $k=\log K/S_t$1 given $k=\log K/S_t$2 as $k=\log K/S_t$3 (Friz et al., 2024). In classical diffusive cases ($k=\log K/S_t$4; e.g., Heston or lognormal volatility), this yields SSR $k=\log K/S_t$5, consistent with the "SSR=2" rule documented by Bergomi and empirically verified in index options for short tenors (Vargas et al., 2013). In rough volatility models with Hurst exponent $k=\log K/S_t$6 ($k=\log K/S_t$7), the short-maturity limit is $k=\log K/S_t$8 (Fukasawa, 5 Feb 2026, Friz et al., 2024). This generalizes and refines earlier lore, explicitly connecting observed SSR-in-sample to roughness in the volatility signal.

For large maturities ($k=\log K/S_t$9), SSR decays to unity: $T>t$0 which corresponds to the sticky-strike scenario (Vargas et al., 2013). In nonlinear and non-Gaussian volatility-of-volatility settings (e.g., asymmetric GARCH), SSR can exceed 2 for small maturities, with the discrepancy explained by the divergence between "smile skew" and cumulant skewness in the underlying return distribution.

4. SSR in Affine Forward Variance and Multi-Factor Models

In the affine forward variance framework, the spot and implied volatility processes evolve jointly: $T>t$1 with a cumulant exponent representation

$T>t$2

where $T>t$3 solves a convolution–Ricatti equation depending on $T>t$4 and correlation structures. The resulting SSR formula is

$T>t$5

enabling robust computation in classical stochastic volatility models (e.g., Heston) and their non-Markovian extensions (Friz et al., 2024). Recent developments avoid the need for the Markov property: generalized representations via Itô–Wentzell and Clark–Ocone formulae show that, for a stochastic volatility process,

$T>t$6

with explicit expressions for the quadratic covariation terms in terms of Malliavin derivatives and characteristic functionals (Fukasawa, 5 Feb 2026).

Multi-factor volatility models, such as the two-factor quintic Ornstein-Uhlenbeck process, calibrate SSR across the entire maturity range (Jaber et al., 18 Mar 2025). Here, SSR is expressed in terms of multivariate derivatives of the forward ATM volatility with respect to its driving factors, capturing the nuanced smile dynamics observed in both SPX and VIX markets.

5. Dependence on Forward-Variance Curve and Path-Dependence

SSR is highly sensitive to the entire path and shape of the forward-variance curve $T>t$7. In AFV models, a leading-order asymptotic for SSR is given by

$T>t$8

with $T>t$9 (Friz et al., 2024, Fukasawa, 5 Feb 2026). An upward-sloping forward-variance curve ("contango") amplifies $\mathcal{R}_t(T) = \frac{\beta_t(T)}{S_t(T)}$0; a downward-sloping curve ("backwardation") dampens it. In the limit of infinite market history, $\mathcal{R}_t(T) = \frac{\beta_t(T)}{S_t(T)}$1 itself becomes a complex functional of historical asset price paths, imbuing SSR with path-dependence that is not summarized by any finite-dimensional Markovian summary. Thus, two models yielding nearly identical implied-volatility surfaces may nonetheless differ markedly in SSR term structure, underscoring the necessity of SSR as an independent calibration target (Friz et al., 2024, Jaber et al., 18 Mar 2025).

6. Empirical Estimation, Pathwise Diffusions, and Applications

Empirical SSR is estimated either directly from option smiles (ATM vol and skew regression techniques) or, in sticky/reflecting diffusion models, via high-frequency path functionals. In the case of sticky-skew threshold SDEs,

$\mathcal{R}_t(T) = \frac{\beta_t(T)}{S_t(T)}$2

the SSR equals $\mathcal{R}_t(T) = \frac{\beta_t(T)}{S_t(T)}$3, quantifiable by forming occupation time/left-right local time path functionals and solving for $\mathcal{R}_t(T) = \frac{\beta_t(T)}{S_t(T)}$4 and $\mathcal{R}_t(T) = \frac{\beta_t(T)}{S_t(T)}$5 (Anagnostakis et al., 2024). High-frequency asymptotics ensure consistent estimation provided the local time at the threshold is reached.

SSR plays a critical role in modern risk and portfolio management. In model-independent risk engines based on perturbed optimal transport, SSR underpins the propagation of SPX shocks to VIX options by dictating the linear response of VIX implied volatility to changes in the SPX forward and vice versa. This yields substantial computational improvements (orders of magnitude speedup relative to full recalibration) with high first-order accuracy in cross-greeks and risk sensitivities, as shown in synthetic and backtest studies (Che et al., 11 Mar 2026).

7. Economic Significance and Calibration Implications

A high SSR ($\mathcal{R}_t(T) = \frac{\beta_t(T)}{S_t(T)}$6 in short maturities) signals that ATM implied volatility responds more strongly to spot moves than predicted by the sticky-strike rule (SSR = 1) or even by linear stochastic-volatility theory (SSR = 2), a feature backed by empirical analyses on SPX and DAX markets (Vargas et al., 2013, Jaber et al., 18 Mar 2025). This informs vega-delta hedging and risk management: failure to account for SSR in model calibration leads to systematic misestimation of cross-sensitivities, particularly in the pricing and risk management of short-dated options.

Table: Summary of SSR Limits and Empirical Ranges

Model Class	Short-Time SSR Limit	Long-Maturity SSR Limit	Empirical SPX Range
Linear SV (diffusive)	2	1	1–2
Rough SV (Hurst $\mathcal{R}_t(T) = \frac{\beta_t(T)}{S_t(T)}$7)	$\mathcal{R}_t(T) = \frac{\beta_t(T)}{S_t(T)}$8	—	$\mathcal{R}_t(T) = \frac{\beta_t(T)}{S_t(T)}$9
Nonlinear GARCH	$S_t(T) = \partial_k \sigma_{,t}(k,T)\big|_{k=0}$0 possible	$S_t(T) = \partial_k \sigma_{,t}(k,T)\big|_{k=0}$1	$S_t(T) = \partial_k \sigma_{,t}(k,T)\big|_{k=0}$2 short, $S_t(T) = \partial_k \sigma_{,t}(k,T)\big|_{k=0}$3 long
Sticky-skew diffusion	$S_t(T) = \partial_k \sigma_{,t}(k,T)\big|_{k=0}$4	—	—
Two-factor Quintic OU	0.9–2.0 (calibrated fit)	—	0.9–2.0

A plausible implication is that, even with near-indistinguishable implied volatility surfaces, differing SSR calibrations can reveal latent structural model deficiencies, particularly the inability to capture joint innovations in spot and implied volatility. Calibration routines incorporating SSR directly (via loss terms or auxiliary moments) align more closely with observed market dynamics, producing term structures matching market SSR across all tenors (Jaber et al., 18 Mar 2025, Friz et al., 2024). Robust estimation and use of SSR are thus essential for modern quantitative option modeling and for the development and validation of volatility surface models and risk engines.