Gatheral Raw SVI Five-Parameter Total Variance Fitting on SPX Listed Options

Fitting a 30-day S&P 500 option chain across 195 active strikes with Gatheral's raw Stochastic Volatility Inspired (SVI) equation reduces weighted implied volatility root-mean-square error to 0.11 vol points while preserving linear wing variance out to 2-delta crash puts. Originally introduced at Merrill Lynch in 1999 and formalized by Jim Gatheral in 2004, raw SVI writes total implied variance w(k) = σBS2(k) × T as an algebraic hyperbola over forward log-moneyness k = ln(K / FT). Within our SVI & SABR Volatility Surface Calibration & No-Arbitrage Conditions blueprint, raw SVI serves as the foundational single-slice engine benchmarked against the SABR vs Gatheral SVI surface standard.

Geometric Meaning of the Five Raw SVI Coordinates#

Every coordinate in w(k; a, b, ρ, m, σ) governs a distinct geometric transformation of the total variance smile:

• a ∈ R (Vertical Variance Shift): Moves the entire total variance hyperbola up or down; must satisfy a + bσ√(1 − ρ2) ≥ 0 so minimum variance never crosses zero.
• b ≥ 0 (Wing Opening Angle): Controls the sum of left and right asymptotic slopes, scaling overall smile kurtosis across out-of-the-money puts and calls.
• ρ ∈ (−1, 1) (Counter-Clockwise Rotation): Tilts asymmetry between downside put wings sL = −b(1 − ρ) and upside call wings sR = b(1 + ρ); SPX 30-day slices typically calibrate around ρ ∈ [−0.78, −0.62].
• m ∈ R (Horizontal Vertex Translation): Shifts the log-moneyness location where the hyperbola attains its minimum variance curvature, typically sitting +1.5% to +3.5% above the SPX forward price.
• σ > 0 (ATM Vertex Smoothing): Rounds the hyperbola apex at k = m; as σ approaches zero, raw SVI degenerates into a sharp piecewise-linear V-smile.

Quasi-Explicit Raw SVI Calibration Architecture (ADC Method)#

Solving the Non-Convex Five-Dimensional Trap via Zeliade ADC#

Naively running five-dimensional Levenberg-Marquardt directly on (a, b, ρ, m, σ) produces severe parameter collinearity between a, m, and σ: flat smiles allow σ and a to trade off inversely, causing intraday parameter jumping even when option mid-quotes move by a single tick. De Marco and Martini's quasi-explicit Zeliade reparameterization eliminates this instability by splitting the 5D problem into a convex 3D inner quadratic program nested inside a 2D outer search over (m, σ).

Setting y = (k − m) / σ, total variance becomes strictly linear in three transformed coordinates (a, d = bρσ, c = bσ):

Empirical SPX Tenor Behavior and Desk Diagnostics#

When calibrating across Cboe SPX weekly and monthly chains, raw SVI parameters follow tight structural scaling laws. Traders convert raw outputs into human-readable skew metrics using Natural SVI and Jump-Wings parameter mapping and verify every calibrated wing against the Durrleman butterfly density positivity condition before quoting client blocks or running Greeks through our Implied Volatility Calculator and Black-Scholes Option Calculator.

Guarding Against Short-Dated Butterfly Dents#

In expirations shorter than 5 trading days, heavy retail and hedge fund bidding for 10-delta SPX puts often tempts unconstrained optimizers to push b higher while shrinking σ below 0.015. Under CFTC and U.S. SEC automated market-access controls, desk risk engines impose a hard floor σ ≥ 0.018 × √(T / 0.0822) and cross-check adjacent slices inside SVI & SABR Volatility Surface Calibration & No-Arbitrage Conditions so adjacent expiries never intersect.