Skip to content
VolSurface Desk
Directory

Svi Sabr Smile Arbitrage Free

Stochastic Local Volatility Mixing Weights and Gyongy Leverage Calibration

Stochastic Local Volatility (SLV) combines Heston vol-of-vol dynamics with a calibrated Gyongy leverage multiplier L(S, t) to match both vanilla smiles and exotic barrier prices.

By Dr. Alistair Finch, CQF, FRMReview not recorded
Stochastic Local Volatility Mixing Weights and Gyongy Leverage Calibration — VolSurface Desk 2026 quantitative & supervisory data card
VolSurface Desk (SVI & SABR Volatility Surface Calibration & No-Arbitrage Conditions): McKean-Vlasov particle calibration of Gyongy leverage functions L(S, t) for exotic barriers and autocallables.Source: VolSurface Desk Institutional Research DeskLicense: CC-BY-4.0

Exotic equity and FX derivatives desks face a fundamental modeling dilemma: pure stochastic volatility (Heston or SABR) captures realistic forward-skew and vol-of-vol dynamics but misses short-dated vanilla SPX option quotes by 0.80 to 1.90 vol points, whereas Dupire local volatility PDE smoothing fits every vanilla option to 0.04 vol points while underpricing barrier knock-ins and forward-start cliquets. Stochastic Local Volatility (SLV) resolves both requirements simultaneously by multiplying stochastic volatility √Vt by a deterministic leverage function L(St, t) calibrated via István Gyöngy's (1986) Markovian projection theorem. Within our SVI & SABR Volatility Surface Calibration & No-Arbitrage Conditions desk guide, SLV is the universal production workhorse for autocallables, double-no-touch barriers, and cliquets.

Gyöngy's Theorem and the Conditional Expectation Identity#

Consider an SLV diffusion under risk-neutral measure Q where spot St and variance Vt follow: dSt / St = (rt − qt) dt + L(St, t) · √Vt · dWt1, dVt = κ(θ − Vt) dt + ξm · √Vt · dWt2

Here ξm = mmix · ξ0 scales base Heston vol-of-vol ξ0 by a trader-controlled mixing fraction mmix ∈ [0, 1]. By Gyöngy's (1986) theorem, this 2D SLV process generates the exact same marginal distributions St (and therefore exact same European option prices across all K and T!) as 1D Dupire local volatility σ_loc(S, t) if and only if:

Formula or model from the articleNot independently verified
L2(K, T) = σ_loc2(K, T) ÷ E^Q[ V_T | S_T = K ], where ξ_eff = m_mix × ξ_Heston (m_mix ∈ [0, 1])
  • Limiting Cases: mmix = 0 collapses Vt to deterministic θt => Pure Dupire Local Vol; mmix = 1 with L ≡ 1 => Pure Heston Stochastic Vol.
  • Guyon-Henry-Labordère Particle Estimator: E[VT | ST = K] ≈ [ ∑i VT^(i) · Kh(ST^(i) − K) ] ÷ [ ∑i Kh(S_T^(i) − K) ] across N = 32,768 Sobol paths.
  • Worked SPX 1-Year Down-and-In Put (Barrier B = 75% S0, Strike K = 90% S0, m_mix = 0.60 calibrated to VVIX):
  • Valuation Spread: Pure Local Vol = 2.84% of notional | Pure Heston = 3.68% | 60%-Mixed SLV = 3.39% (matches interbank broker run 3.41% within 2 bps!).

Guyon-Henry-Labordère Particle SLV Calibration Engine#

Guyon-Henry-Labordere McKean-Vlasov particle method for calibrating SLV Gyongy leverage function forward in time
Guyon-Henry-Labordere McKean-Vlasov particle method for calibrating SLV Gyongy leverage function forward in timeVerified Exhibit

Calibrating the Mixing Weight m_mix to Observable Exotics#

How does an institutional exotic desk choose mmix rather than guessing 50%? Because European vanillas are completely invariant to mmix (any mmix ∈ [0, 1] reprices the entire vanilla board identically once L(S, t) is calibrated!), mmix is uniquely identified by second-generation vol-of-vol observables:

  • Equity Index Desks: Calibrate m_mix (typically 0.55 to 0.68 on SPX and SX5E) against Cboe VVIX vol-of-vol benchmarks, forward-starting variance swaps, and broker quotes on 1Y monthly-reset Napoleon cliquets.
  • FX Exotic Desks: Calibrate m_mix (typically 0.50 to 0.60 on EURUSD and USDJPY) so SLV matches interbank Vanna-Volga prices on 3-month and 1-year Double-No-Touch (DNT) contracts, tracking third-order convexity in Vomma, Zomma & Speed mechanics.

Particle Method vs 2D Fokker-Planck ADI Finite Differences#

Historically, calibrating L(S, t) required iterating a 2D Alternating Direction Implicit (ADI) forward Fokker-Planck PDE for joint density p(S, V, t), taking 1.2 to 2.5 seconds per surface update. Julien Guyon and Pierre Henry-Labordère's (2012) McKean-Vlasov particle method cut calibration latency by 35× (down to 38 ms) and scaled seamlessly to 3-factor models (SLV + Hull-White stochastic interest rates) needed for 5-year autocallable notes!

Exotic option pricing spreads across mixing fractions m_mix from 0.0 Pure Local Vol to 1.0 Pure Stochastic Vol
Exotic option pricing spreads across mixing fractions m_mix from 0.0 Pure Local Vol to 1.0 Pure Stochastic VolVerified Exhibit
Quantitative Distribution · Vanilla SPX RMSE
Vanilla SPX RMSEBarrier / Cliquet Bias
01Pure Dupire (m_mix = 0)0.04 vol02Pure Heston (L ≡ 1)0.92 vol032D ADI PDE SLV0.05 vol04Particle McKean SLV0.06 vol
Model ChoiceVanilla SPX RMSEBarrier / Cliquet BiasCalibration SpeedDesk Hedging / Tool Reference
Pure Dupire (m_mix = 0)0.04 vol−55 bps on 1Y DIP barrier2.4 ms (1D PDE)Black-Scholes Calculator
Pure Heston (L ≡ 1)0.92 volMisses weekly vanilla hedges12 ms (Heston FFT)SABR-SVI Surface
2D ADI PDE SLV0.05 vol< 2 bps vs broker market1,350 ms (2D PDE)VVIX Vol-of-Vol
Particle McKean SLV0.06 vol< 2 bps + supports stoch rates38 ms (32k Sobol)Option Greeks & GEX Tool

Feller Condition Pitfalls When ξ0 Is Too High#

Under FINMA, BaFin, and CFTC model risk audits, desks monitor the ratio L(S, t): if underlying Heston parameters violate Feller's condition 2κθ > ξ2 severely, a large mass of particles piles up near Vt ≈ 0, driving E[Vt | St = K] toward zero and causing L(K, t) to spike above 3.0. Keeping mmix ≈ 0.60 inside SVI & SABR Volatility Surface Calibration & No-Arbitrage Conditions keeps 2κθ > (m_mix · ξ0)2 satisfied naturally while preserving realistic Vomma and Vanna Greeks.

Frequently asked questions

Why are European vanilla option prices virtually identical for any SLV mixing fraction m_mix between 0 and 1?

By Gyongy’s (1986) theorem, matching E[L^2(S_t, t) * V_t | S_t = K] = sigma_loc^2(K, t) ensures the 1D marginal probability distribution of S_T at every maturity T matches Dupire’s surface identically, regardless of how variance splits between L and V_t.

How does changing m_mix alter delta and vega hedges on a structured autocallable book?

Higher m_mix increases correlation between spot drops and implied volatility spikes near knock-in barriers, flattening spurious local-vol gamma cliffs and shifting hedge demand from pure delta scalping toward OTM put vega.

What bandwidth h is used in the Nadaraya-Watson kernel regression during particle SLV calibration?

Production engines use Silverman’s adaptive rule-of-thumb bandwidth h = 1.06 * std(S_t) * N^(-1/5) (around 0.13 * std(S_t) for N = 32,768) combined with linear B-spline bin accumulation so each strike bin averages hundreds of particles.

Can SLV use lognormal or rough volatility processes instead of square-root Heston variance?

Yes. Many modern equity exotic desks replace CIR/Heston V_t with a two-factor lognormal Bergomi or Markovian-approximated Rough Bergomi process, completely avoiding zero-variance Feller boundary traps while improving short-dated forward skew.

Publication record

Published
November 14, 2026
Updated
November 14, 2026
Author
Dr. Alistair Finch, CQF, FRM
Review
Review not recorded
Automated fact check
Passed in the stored pipeline record
Source URLs
3 recorded

AI assistance recorded: anthropic/claude-opus-4.6. This field does not establish human review or source verification.

Academic & Institutional Attribution

Cite This Research & Embed Live Data Card

Open Attribution License (CC BY 4.0)

Standard academic & institutional research paper citation format.

Dr. Alistair Finch, CQF, FRM. (2026, November 14). Stochastic Local Volatility Mixing Weights and Gyongy Leverage Calibration. VolSurface Desk. https://volsurfacedesk.com/desk-notes/stochastic-local-volatility-slv-gyongy-leverage-mixing
Explore Svi Sabr Smile Arbitrage Free Hub (20) →

VolSurface Desk Data, Benchmarks & Contributor Records

Stochastic Local Volatility (SLV) & Gyongy Calibration | VolSurface Desk