Quantitative Derivatives Smile Calibration Arbitrage-Free SVI

Gatheral SVI Volatility Surface & Arbitrage Lab

Explore Jim Gatheral’s (2004) Stochastic Volatility Inspired (SVI) parametric smile model. Calibrate raw total variance $w(k)$, verify Roger Lee asymptotic extreme-strike moment bounds, extract Breeden-Litzenberger risk-neutral densities, and prevent butterfly arbitrage.

Model: Raw SVI $w(k) = a + b[\rho(k-m) + \sqrt{(k-m)^2+\sigma^2}]$ Asymptotics: Roger Lee Bound $b(1+|\rho|) \le 2$ No-Arbitrage: Durrleman Condition $g(k) \ge 0$
Institutional Highlights
  • • Standard smile model across Wall Street equity & FX desks
  • • Prevents negative local volatilities in Dupire PDE pricing
  • • Closed-form Jump-Wings (SVI-JW) trader mapping
  • • Strict mathematical guarantee against butterfly spread arbitrage
ATM Volatility $\sigma_{\text{ATM}}$
--%
Annualized at $k = 0$
ATM Skew $\partial \sigma / \partial k$
--
Log-moneyness slope
Min Variance $\sigma_{\min}$
--%
$w_{\min} = a + b\sigma\sqrt{1-\rho^2}$
Roger Lee Max Slope
-- / 2.0
Must be $\le 2.0$
Arbitrage Verdict
Evaluating...
Durrleman Butterfly $g(k) \ge 0$
Raw SVI Parameters
Adjust Gatheral (2004) total variance parameters
Select a preset to inspect distinct volatility market regimes.
0.040
-0.05 (Depressed) 0.20 (Elevated)
0.35
0.02 (Flat) 1.30 (Steep Wings)
-0.65
-0.95 (Equity Skew) +0.95 (Call Skew)
0.04
-0.40 +0.40
0.15
0.01 (Sharp Vertex) 0.45 (Smooth Bowl)
0.25
Diagnostics & Jump-Wings (SVI-JW)
Left Wing Slope $b(1-\rho)$ --
Right Wing Slope $b(1+\rho)$ --
Minimum Butterfly $g(k)$ --
Minimum Variance $w_{\min}$ --
JW ATM Variance $v_t$ --
JW ATM Skew $\psi_t$ --
JW Left Slope $p_t$ --
JW Right Slope $c_t$ --
JW Min Variance $\tilde{v}_t$ --
Chart Notes: Tab 1 plots the annualized Black-Scholes implied volatility smile $\sigma_{\text{BS}}(K)$ alongside linear wing asymptotes. Tab 2 evaluates the Breeden-Litzenberger risk-neutral density and Durrleman's butterfly condition $g(k)$ (arbitrage occurs whenever $g(k) < 0$). Tab 3 compares total variance $w(k)$ against the theoretical Roger Lee boundary $2|k|$.
5×5 Arbitrage Sensitivity Heatmaps
Analyze boundary conditions across structural parameter pairs
Roger Lee Wing Slope Ratio (%)
Wing Slope $b$ (rows) vs Skew $\rho$ (cols) $\to \frac{\max(b(1\pm\rho))}{2.0} \times 100\%$
$b \backslash \rho$ -0.85 -0.50 0.00 +0.50 +0.85
Green: Valid ($\le 100\%$). Red: Extreme-strike arbitrage ($> 100\%$, Roger Lee breach).
Butterfly Condition $\min_k g(k)$
Smoothing $\sigma$ (rows) vs Level $a$ (cols) $\to \min_{k} g(k)$
$\sigma \backslash a$ -0.02 0.00 0.03 0.06 0.10
Green: Valid non-negative density ($g \ge 0$). Red: Butterfly arbitrage ($g < 0$).

Mathematical Foundations of Gatheral’s SVI Model

1. Raw SVI Formulation

Introduced by Jim Gatheral in 2004, the Raw SVI model parameterizes total implied variance $w(k) = \sigma_{\text{BS}}^2(k) T$ as a function of log-moneyness $k = \ln(K / F)$:

$$w(k) = a + b \left[ \rho (k - m) + \sqrt{(k - m)^2 + \sigma^2} \right]$$
  • $a \in \mathbb{R}$: Baseline overall variance level.
  • $b \ge 0$: Wing angle determining the asymptotic spread between put and call wings.
  • $\rho \in (-1, 1)$: Smile orientation/skew. Negative values reflect equity crash risk.
  • $m \in \mathbb{R}$: Horizontal translation shifting the minimum variance location.
  • $\sigma > 0$: Smoothing scale governing the curvature radius at the vertex.

2. Minimum Variance & Vertex Location

Differentiating with respect to $k$ yields the unique vertex location $k^*$ where $\partial w / \partial k = 0$:

$$k^* = m - \frac{\rho \sigma}{\sqrt{1 - \rho^2}}, \quad w(k^*) = a + b \sigma \sqrt{1 - \rho^2}$$

To prevent negative implied variance ($w(k) \le 0$), we require $w(k^*) = a + b \sigma \sqrt{1 - \rho^2} > 0$.

3. Roger Lee's Moment Formula & Wing Asymptotics

As $|k| \to \infty$, the square root term $\sqrt{(k-m)^2+\sigma^2} \to |k - m|$. The asymptotic slopes are linear:

$$\lim_{k \to +\infty} \frac{w(k)}{k} = b(1 + \rho), \quad \lim_{k \to -\infty} \frac{w(k)}{|k|} = b(1 - \rho)$$

Roger Lee (2004) demonstrated that total variance cannot exceed $2|k|$ asymptotically without permitting infinite moments or negative pricing. This imposes the universal bound:

$$b(1 + |\rho|) \le 2$$

4. Durrleman-Gatheral Butterfly Arbitrage Condition

By the Breeden-Litzenberger (1978) theorem, the risk-neutral probability density must remain strictly non-negative:

$$g(k) = \left( 1 - \frac{k w'(k)}{2 w(k)} \right)^2 - \frac{w'(k)^2}{4} \left( \frac{1}{w(k)} + \frac{1}{4} \right) + \frac{w''(k)}{2} \ge 0$$

Whenever $g(k) < 0$, the corresponding butterfly spread option package has a negative price, enabling immediate riskless arbitrage.

Quantitative Mastery Quiz: SVI & Volatility Arbitrage

Test your structural understanding of Gatheral's SVI parameterization, no-arbitrage bounds, and risk-neutral density dynamics.

Frequently Asked Questions: SVI & Arbitrage-Free Calibration

The Stochastic Volatility Inspired (SVI) model was introduced by Jim Gatheral in 2004 as an analytical parameterization of total implied variance $w(k) = \sigma_{\text{BS}}^2(k) T$ as a hyperbolic function of log-moneyness $k = \ln(K/F)$: $$w(k) = a + b \left[ \rho (k - m) + \sqrt{(k - m)^2 + \sigma^2} \right]$$ It enforces asymptotic linearity in extreme strikes consistent with Heston stochastic volatility models while offering an exact five-parameter representation of the volatility smile.

Roger Lee (2004) proved that total implied variance cannot grow faster than $2|k|$ at extreme strikes without implying negative or infinite price moments. For Raw SVI, the asymptotic put and call wing slopes are $b(1 - \rho)$ and $b(1 + \rho)$. Roger Lee's no-arbitrage boundary requires that: $$b(1 + |\rho|) \le 2$$ Any SVI parameter set violating this bound creates extreme strike arbitrage where deep out-of-the-money options violate fundamental risk-neutral bounds.

Absence of butterfly arbitrage requires the Breeden-Litzenberger risk-neutral probability density $p(K) = e^{rT} \frac{\partial^2 C}{\partial K^2}$ to remain non-negative for all strikes. In SVI total variance notation, this translates to the Durrleman condition: $$g(k) = \left( 1 - \frac{k w'(k)}{2 w(k)} \right)^2 - \frac{w'(k)^2}{4}\left(\frac{1}{w(k)} + \frac{1}{4}\right) + \frac{w''(k)}{2} \ge 0$$ If $g(k)$ drops below zero at any strike, a standard call or put butterfly spread centered at that strike yields a negative cost, creating an immediate static arbitrage opportunity.

Raw SVI operates on five parameters ($a, b, \rho, m, \sigma$) for a single expiration slice. Natural SVI re-parameterizes the smile into terms closer to trader intuition: minimum variance, skew, and wing levels. Surface SVI (SSVI), formulated by Gatheral and Jacquier (2014), parameterizes entire 2D volatility surfaces across strikes and maturities simultaneously with analytical guarantees against both butterfly arbitrage and calendar spread arbitrage ($\partial w / \partial T \ge 0$).

Cubic splines frequently introduce spurious oscillations in implied volatility slopes and second derivatives, generating severe butterfly arbitrage (negative risk-neutral densities) between observed market strikes. In contrast, SVI provides a smooth, convex, economically grounded parametric smile that guarantees asymptotic stability, fits vanilla quotes within bid-ask spreads, and allows reliable local volatility extraction via Dupire's equation.

Related Quantitative Derivatives & Microstructure Labs

Rough Volatility & Bergomi Lab

Explore Gatheral, Jaisson, and Rosenbaum (2018) fractional Brownian motion, sub-diffusive Hurst parameter $H < 0.5$, and power-law short-dated skew explosion.

Dupire Local Volatility Lab

Extract non-parametric local volatility surfaces $\sigma_{\text{loc}}(S, t)$ from arbitrage-free vanilla call options prices via Dupire's equation.

Cont-Stoikov Order Book Lab

Model continuous-time Markov chain order book dynamics, order cancellation scaling, and first-passage price increase absorption probabilities.

Cont-de Larrard Order Book Lab

Heavy-traffic diffusion limit, conformal harmonic exit probabilities $\frac{2}{\pi}\arctan(q_b / q_a)$, and micro-volatility $\sigma^2 = \frac{\pi\mu\delta^2}{4D^2}$.