Skip to content
VolSurface Desk
Directory

Svi Sabr Smile Arbitrage Free

Durrleman g(k) Density Condition: Eliminating Butterfly Arbitrage in Wing Strikes

Durrleman’s g(k) >= 0 inequality expresses Breeden-Litzenberger call convexity directly in terms of total implied variance w(k), w'(k), and w''(k) without option repricing.

By Dr. Alistair Finch, CQF, FRMReview not recorded
Durrleman g(k) Density Condition: Eliminating Butterfly Arbitrage in Wing Strikes — VolSurface Desk 2026 quantitative & supervisory data card
VolSurface Desk (SVI & SABR Volatility Surface Calibration & No-Arbitrage Conditions): Analytical g(k) >= 0 butterfly positivity diagnostics and Dupire denominator protection.Source: VolSurface Desk Institutional Research DeskLicense: CC-BY-4.0

Static butterfly arbitrage occurs on a listed option slice whenever a portfolio long one K − ΔK call, short two K calls, and long one K + ΔK call carries a negative net mid-price—equivalently, whenever the risk-neutral probability density q(K) = exp(rT) × ∂2C / ∂K2 drops below zero. Evaluating finite-difference call prices across thousands of strikes during millisecond optimizations is computationally wasteful and prone to floating-point cancellation in deep wings. Instead, institutional quant libraries inside our SVI & SABR Volatility Surface Calibration & No-Arbitrage Conditions architecture enforce Valdo Durrleman's (2005) exact analytical function g(k) ≥ 0 directly on total implied variance w(k) and its first two log-moneyness derivatives.

Exact Relationship Between Risk-Neutral Density and g(k)#

Let k = ln(K / FT) denote forward log-moneyness and w(k) = σBS2(k) × T denote total implied variance on a fixed maturity T, assuming asymptotic call price decay C(k) → 0 as k → +∞ (which holds whenever the Roger Lee moment formula wing slope caps keep right-wing variance slope strictly below 2). Differentiating the Black-Scholes call formula twice with respect to strike K yields the exact identity: q(K) = exp(rT) × (∂2C / ∂K2) = [ φ(d2(k)) / (K · √w(k)) ] × g(k)

where φ(z) > 0 is the standard normal density and d2(k) = −k / √w(k) − √w(k) / 2. Because both φ(d2) and K·√w(k) are strictly positive for every finite strike, the risk-neutral density q(K) is non-negative if and only if g(k) ≥ 0 across all k ∈ R!

Anatomy of the Durrleman Butterfly Function g(k)#

Analytical decomposition of Durrleman g(k) butterfly density function into three competing variance terms
Analytical decomposition of Durrleman g(k) butterfly density function into three competing variance termsVerified Exhibit

Anatomy of a Real-World Butterfly Failure ("The Axel Vogt Smile")#

Why do unconstrained Gatheral raw SVI five-parameter fits occasionally fail Durrleman's test? Consider a steep 5-DTE SPX pre-CPI chain: when total variance w(k) is very small (short T) while ATM skew |w'(k)| is large, the negative middle term −w'(k)2 / (4·w(k)) overwhelms both [1 − k·w'(k)/(2·w(k))]2 and w''(k)/2 just below ATM (where k < 0 and w'(k) < 0 make k·w'(k) > 0, eroding Term 1!).

Formula or model from the articleNot independently verified
g(k) = [ 1 − k·w'(k) / (2·w(k)) ]2 − (w'(k)2 / 4)·(1 / w(k) + 1 / 4) + w''(k) / 2 ≥ 0
  • Stress Case (Steep 5-DTE Slice): a = −0.0006, b = 0.095, ρ = −0.82, m = +0.025, σ = 0.014 evaluated at k = −0.012 (98.8% put).
  • Slice Derivatives at k = −0.012: w(−0.012) = 0.006041, w'(−0.012) = −0.16675, w''(−0.012) = +0.3007.
  • Three-Term Breakdown: Term 1 = +0.6962, Term 2 = −1.1524 (magnified by 1/w = 165.5), Term 3 = +0.1504 => g(−0.012) = −0.3058 < 0 (Butterfly Arb!).
  • Regularized Fix: Lifting vertex smoothing σ from 0.014 to 0.026 raises min_k g(k) to +0.084 while changing mid-IV by < 0.04 vol.

Traders can verify individual call and put butterfly spreads directly in our Black-Scholes Option Calculator.

Production Grid Enforcement and Dupire PDE Stability#

Why do exotic and risk engines care even if min g(k) = −0.02 sits inside wide bid-ask spreads where transaction costs block retail arbitrage? Because Dupire local volatility PDE finite-difference smoothing computes σ_loc2(K, T) = (∂w / ∂T) / g(k). If g(k) approaches zero or crosses negative anywhere on the grid, local variance explodes to ±∞ and crashes Crank-Nicolson PDE solvers and OCC STANS 20-minute intraday Monte Carlo margin snapshots!

Minimum Durrleman g(k) values across four SVI calibration regimes before and after butterfly penalty regularization
Minimum Durrleman g(k) values across four SVI calibration regimes before and after butterfly penalty regularizationVerified Exhibit
Quantitative Distribution · Guaranteed min g(k)
Guaranteed min g(k)
01Unregularized Raw SVIUnbounded (< 0 …02SVI-JW ṽ_t Floor> 0 analytically
Enforcement TechniqueMathematical MechanismGuaranteed min g(k)Fit FlexibilityLinked Desk Resource
Unregularized Raw SVIPure ∑(σfit − σmkt)2Unbounded (< 0 common)Highest (overfits noise)SABR-SVI Surface
Chebyshev g(k) Penalty+λ ∑ max(0, ε − g(k_j))2≥ +ε on dense gridHigh (slice-level SVI)Vega-Weighted SVI Optimizers
SVI-JW ṽ_t Floorṽt ≥ vt · 4·pt·ct / (pt+ct)2> 0 analyticallyExact ATM vol & skewSVI-JW Trader Mapping
Global Power-Law SSVIη(1 +ρ) ≤ 2> 0 for all (k, T)Moderate (global 3-param)SSVI Power-Law Kernel

Regulatory Margin and Clearing Implications#

Both Cboe market-maker risk gateways and clearing member engines overseen by the CFTC run automated pre-quote Durrleman scans across SVI & SABR Volatility Surface Calibration & No-Arbitrage Conditions and Breeden-Litzenberger risk-neutral PDF tail extrapolation. Maintaining g(k) ≥ 0.01 ensures every listed strike admits a valid martingale pricing measure and well-conditioned Greeks in our Option Greeks & GEX Calculator.

Frequently asked questions

Does g(k) >= 0 alone suffice to rule out all vertical strike arbitrage?

Almost: strict absence of butterfly and call-spread arbitrage requires both g(k) >= 0 for all k in R and the asymptotic boundary condition lim_(k -> +infinity) d_1(k) = -infinity, which holds automatically whenever the right-wing total variance slope stays strictly below 2.

Why not check butterfly arbitrage using discrete call prices C(K-dK) - 2C(K) + C(K+dK) >= 0?

In 5-delta and 1-delta wings, call prices decay exponentially to 1e-9 while floating-point rounding introduces 1e-16 noise, causing catastrophic cancellation in discrete second differences; g(k) evaluates cleanly in O(1) variance units across all moneyness.

Where on an equity index smile does Durrleman g(k) typically hit its lowest value?

For negatively skewed index smiles like SPX, g(k) reaches its minimum slightly below or near the ATM vertex where slope |w'(k)| transitions steeply while total variance w(k) is still small, rather than in the linear asymptotic wings.

What minimum positive threshold epsilon do production desks enforce for g(k)?

Most institutional surface engines enforce g(k) >= 0.005 to 0.010 across [-5*sqrt(w_ATM), +4*sqrt(w_ATM)] so Dupire local variance sigma_loc^2 = (dw/dT) / g(k) never spikes above 10x ATM variance.

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). Durrleman g(k) Density Condition: Eliminating Butterfly Arbitrage in Wing Strikes. VolSurface Desk. https://volsurfacedesk.com/desk-notes/durrleman-condition-butterfly-arbitrage-density-positivity
Explore Svi Sabr Smile Arbitrage Free Hub (20) →

VolSurface Desk Data, Benchmarks & Contributor Records

Durrleman g(k) Condition & Butterfly Arbitrage Checks | VolSurface Desk