← All papers
First page of The Right Wing of the SABR Smile with an Absorbing Boundary: Resolution of Conjectures of Henry-Labordère

The Right Wing of the SABR Smile with an Absorbing Boundary: Resolution of Conjectures of Henry-Labordère

Zeyu Cao, Shaosai Huang

math.PR Oct 3, 2026 · v2 q-fin.MF
Selected deterministic steps are formalized in Lean: the Black–Scholes call-price rate, the implied-volatility limit transfer, and the intrinsic-distance calculation, with a claim manifest linking statements to declarations.
We determine the fixed-maturity right tail of the SABR model with $β\in(0,1)$ absorbed at zero, for every correlation $ρ$: the probability $P(k)$ that the forward ends above $f_0e^k$ satisfies $-k^{-2}\ln P(k)\to(1-β)^2/(2ν^2T(1-(ρ\wedge0)^2))$. The implied volatility therefore tends to $ν\sqrt{1-(ρ\wedge0)^2}/(1-β)$. Henry-Labordère's conjectured limit $ν/(1-β)$, which Hagan's leading-order formula also gives, holds for $ρ\ge0$ and fails for $ρ<0$, where survival keeps the noise orthogonal to volatility above a moving square-root barrier. The Gaussian estimate behind his second conjecture does not hold for $ρ<0$; the rate is instead given by the intrinsic distance within the survival domain.

Determine the fixed-maturity right tail and implied-volatility wing of the SABR model with β in (0,1) absorbed at zero, for every correlation ρ. This resolves Henry-Labordère's conjectures that the limit is ν/(1-β) and that a Gaussian estimate based on unrestricted geodesic distance gives the rate.

A Lamperti–CEV transform splits the forward into a correlation term, an orthogonal Gaussian martingale in quadratic-variation time, and a nonpositive singular drift. For ρ<0 the upper bound discretizes the volatility excursion and bounds the combined volatility and moving-barrier energy. The lower bound uses a Girsanov volatility tube and Ornstein–Uhlenbeck survival blocks above a square-root barrier. The rate is identified with the intrinsic distance in the survival domain using Fermi coordinates on the hyperbolic plane, and a model-free argument converts tail rates into implied-volatility limits. Selected deterministic steps are machine-checked in Lean.

-k^{-2} ln P(k) tends to (1-β)^2/(2ν^2T(1-(ρ∧0)^2)), so implied volatility tends to ν√(1-(ρ∧0)^2)/(1-β). The conjectured limit holds for ρ≥0 and fails for ρ<0. Monte Carlo results show that Hagan's formula overstates the right wing, with the excess growing much faster for negative correlation.

kmodel (ρ=-0.6)Hagan (ρ=-0.6)model (ρ=0)Hagan (ρ=0)
10.2550.2580.3600.368
30.3980.4320.5380.557
50.4840.5610.6600.688
∞0.9601.2171.2001.236
Right-wing implied volatilities (Monte Carlo model vs Hagan–Obłój formula), f0=1, α=0.3, ν=0.6, T=1, β=0.5