What Is Butterfly Arbitrage?

Last reviewed: May 17, 2026 by Options Analysis Suite Research.

What Is Butterfly Arbitrage?

For a fixed expiration, the call price function C(K) maps strike K to call value. Three constraints from no-arbitrage:

• Monotonicity. C(K) must be non-increasing in K (calls become less valuable as strike rises).
• Convexity (the butterfly constraint). C(K) must be convex in K, equivalently d^2 C / dK^2 >= 0 for all K. This is what "butterfly arbitrage" refers to.
• Boundary. C(K) -> max(F - K, 0) as K -> 0 (intrinsic at the bottom); C(K) -> 0 as K -> infinity.

The butterfly-arbitrage condition is the central no-arbitrage requirement because it has direct economic content: by Breeden-Litzenberger (1978), d^2 C / dK^2 = e^(-rT) * f^*(K), equivalently f^*(K) = e^(rT) * d^2 C / dK^2, where f^*(K) is the risk-neutral probability density at strike K. A negative second derivative means negative probability density - which is mathematically impossible and operationally exploitable.

The Butterfly Trade as the Arbitrage

If butterfly arbitrage exists, you can profit risk-free using a simple butterfly spread:

• Buy 1 call at strike K - delta_K
• Sell 2 calls at strike K
• Buy 1 call at strike K + delta_K

The payoff at expiration is non-negative everywhere: zero at extremes, peak at K. The cost of this position is C(K - delta_K) - 2*C(K) + C(K + delta_K), which approximates delta_K^2 * d^2 C / dK^2. If the second derivative is negative anywhere on the surface, the butterfly costs negative money to set up - you receive money for a position with non-negative payoff. That is the arbitrage.

Why Does the Butterfly Arbitrage Constraint Matter?

• Static-arbitrage absence is foundational. Any production option-pricing system must guarantee the surface has no internal arbitrage. Trading on an arbitrageable surface produces systematic losses against any market-maker who actually trades the relevant butterflies.
• Risk-neutral density extraction depends on it. RND extraction by Breeden-Litzenberger differentiates the call function. If the call function is not convex, the RND has negative regions - a useless distribution.
• Calibration validity depends on it. Calibrated parameters (Heston, SABR, eSSVI) must produce surfaces that satisfy butterfly absence. A "fitted" surface that violates butterfly arbitrage is not valid for downstream applications.

How Do Surface Models Enforce It?

• SVI (Gatheral 2004). The five-parameter SVI smile must satisfy the Roper (2010) butterfly-absence condition: an algebraic inequality on (a, b, rho, m, sigma) that excludes parameter regions where the smile is too curved. Calibration optimizers reject parameter sets that violate this.
• SSVI / eSSVI (Gatheral-Jacquier 2014). Surface-level extension. The phi(theta) function and rho parameter must satisfy butterfly-absence inequalities at every tenor. Production calibrators enforce these as hard constraints rather than soft penalties.
• Local volatility (Dupire). Constructs the local-vol function sigma(S, t) directly from the second derivative of the call function. Inverts butterfly arbitrage by construction: an arbitrage-free surface produces an interpretable local-vol; an arbitrageable surface produces complex (non-real) local vols.
• Stochastic-vol models (Heston, SABR). Butterfly absence is a property of the underlying stochastic process - any well-specified stochastic-vol model produces arbitrage-free prices automatically. The risk is calibration error producing parameters that imply something pathological.

The Roper Conditions

Roper (2010) gave the canonical algebraic specification of static no-arbitrage on a parametric implied-vol surface. For SVI, the relevant condition involves a function g(k) of the smile parameters; butterfly absence requires g(k) > 0 everywhere. The condition is:

g(k) = (1 - k * w'(k) / (2 * w(k)))^2 - (w'(k))^2 / 4 * (1/w(k) + 1/4) + w''(k) / 2 >= 0

where w(k) is the SVI total-variance function and primes denote derivatives in log-moneyness k. Practical calibration tests this on a dense grid of k values; violations trigger parameter rejection or regularization.

Worked Example

SPX 30-day SVI calibration produces parameters (a, b, rho, m, sigma) = (0.024, 0.18, -0.62, 0.005, 0.07). Computing g(k) on a grid k in [-0.30, +0.30]:

• min g(k) = 0.018 (positive everywhere)
• Surface satisfies butterfly arbitrage absence
• Implied RND is non-negative across all strikes

If a calibration produced negative g(k) at any k, the surface fit is flawed - either the input data has noise, the optimizer landed in a bad parameter region, or the model class is wrong for this regime. Production systems either retry with different starting points or fall back to a more flexible parameterization.

Calendar Arbitrage (Cross-Tenor)

A second arbitrage condition operates across tenors: total variance must grow monotonically in T at every log-moneyness k. Per-expiration SVI does not enforce this; SSVI/eSSVI do. Surface fitting that satisfies butterfly absence at each tenor but not calendar absence across tenors still has arbitrage - just one that requires options at multiple expirations to exploit.

Limitations and Caveats

• Static only. Butterfly arbitrage absence is a static property at one snapshot. Dynamic arbitrages (tradable when conditions change) are a richer topic not captured by the static condition.
• Quote noise. Real listed options have bid-ask spreads. A "butterfly arbitrage" within the spread is not actually exploitable - you need the surface to satisfy convexity within the bid-ask uncertainty bounds, not at every mid-quote.
• Strike grid sparsity. Single-name surfaces with sparse listings can have gaps where butterfly arbitrage cannot be tested directly because the strikes aren't listed.

Related Concepts

Risk-Neutral Density · SVI · eSSVI · Volatility Smile · Convexity · Calibration · Pricing Model Landscape

References & Further Reading

• Breeden, D. T. and Litzenberger, R. H. (1978). "Prices of State-Contingent Claims Implicit in Option Prices." Journal of Business, 51(4), 621-651. The foundational result connecting butterfly spread prices to the risk-neutral density.
• Roper, M. (2010). "Arbitrage Free Implied Volatility Surfaces." Working paper, University of Sydney. The canonical algebraic specification of static no-arbitrage on parametric implied-vol surfaces.
• Gatheral, J. and Jacquier, A. (2014). "Arbitrage-Free SVI Volatility Surfaces." Quantitative Finance, 14(1), 59-71. SSVI and eSSVI calibration with explicit arbitrage-absence enforcement.
• Carr, P. and Madan, D. (2005). "A Note on Sufficient Conditions for No Arbitrage." Finance Research Letters, 2(3), 125-130. Sufficient algebraic conditions for static-arbitrage absence in parametric surfaces.

View live arbitrage-free SPY surface and RND ->

This page is part of the Pricing Model Landscape and the canonical reference set on options market structure. Browse all documentation.