Wing Arbitrage & Negative Density · SVI 402

A volatility surface can lie to you, and the wings are where it lies most.

By now you can fit a smile, read a risk-reversal as the market's directional lean, and read a butterfly as its appetite for tail convexity. But none of that machinery is worth anything if the surface you fit implies something that cannot exist. And here is the uncomfortable truth of this course: an option surface is not a free-form curve you are allowed to draw however you like. It is a disguised probability distribution. Every quoted price is a constraint on that distribution, and if your quotes are mutually inconsistent, the distribution they imply will do something a real distribution can never do — it will go negative. When that happens, you are not looking at a clever skew view. You are looking at free money, or, far more often, at a fitting artifact pretending to be a signal. This chapter teaches you to tell those two apart.

Let us build the intuition before the formula. Start with a vanilla call. Its payoff is the amount by which the underlying finishes above the strike, and nothing below. Now imagine sliding the strike upward by a tiny amount. The call gets cheaper, because you have made it slightly harder to finish in the money. How much cheaper? Exactly in proportion to the probability — the risk-neutral probability — of finishing above that strike. So the slope of the call price as you move the strike is telling you about the cumulative probability in the tail. The call-price curve, read across strikes, is secretly carrying the entire distribution inside its shape.

That is the heart of the Breeden–Litzenberger result, and it is one of the most beautiful facts in derivatives. The first derivative of the call price with respect to strike gives you the risk-neutral probability of finishing above that strike, shifted by one. Take one more derivative — the second derivative of the call price with respect to strike — and you recover the risk-neutral density itself, discounted. In words: the curvature of the call curve at a strike is, up to a discount factor, the probability mass the market is assigning to the underlying landing right at that strike. The smile is a density in costume.

Now the no-arbitrage condition writes itself. A probability density can never be negative. There is no such thing as a minus-three-percent chance of an outcome. Therefore the second derivative of the call price with respect to strike must be non-negative at every strike. Equivalently, the call price must be a convex function of the strike. The moment your fitted surface produces a region where the call curve bends the wrong way — where curvature turns negative — Breeden–Litzenberger hands you a density that dips below zero, and that is a hard, model-free arbitrage. This is called a butterfly arbitrage, because the trade that harvests it is a butterfly spread: buy the strikes on either side, sell two of the middle, and if the density is negative there, you collect a credit for a position that can never pay out negatively. That is the definition of money for nothing.

There is a second condition, gentler but just as important, that lives along the strike axis rather than in curvature. As you raise the strike, a call can only get cheaper or stay flat — never more expensive — and it cannot fall faster than the discount rate allows. The call price must be monotone decreasing in strike, with a bounded slope. Violate that and you have a vertical-spread arbitrage: a cheaper call that dominates a more expensive one. Convexity guards the curvature; monotonicity guards the slope. Together they are necessary and, for a single expiry, very nearly sufficient for the smile to be free of static arbitrage.

So where do the wings come in, and why does SVQ watch them specifically? Because the wings are exactly where you have the least data and the most extrapolation. Deep out-of-the-money strikes trade thinly, their quotes are wide and stale, and any smooth parameterization you fit — SVI, SABR, a spline, whatever your chapter four toolkit gave you — is essentially guessing out there. A small error in the fitted slope of the wing gets differentiated twice to produce the density, and second derivatives are brutal amplifiers of noise. A barely-visible kink in the implied-volatility wing becomes a violent negative spike in the density. That is why the wings lie most: they are the region where the surface is most extrapolated and the density most fragile.

In Saqrid

On a single expiry, the wing-arbitrage panel in SVQ does three things at once. First, it reconstructs the risk-neutral density by twice-differentiating the fitted call curve, and it shades any strike where that density goes negative — those red wing patches are live butterfly arbitrages, or, much more likely, a warning that the fit is unstable out there. Second, it checks RR and BF monotonicity: moving from the at-the-money toward the wings, the risk-reversal (the call-wing minus put-wing skew) should move smoothly and the butterfly (the wing convexity premium) should not fold back on itself. A non-monotone RR or BF as a function of delta is the surface-quote symptom of the same disease the density spike shows in price space. Third, it labels which it found. When SVQ flags a wing, the discipline is identical to the institute's whole throughline: the question is whether that flag is evidence of a real, tradeable dislocation, or merely the fit's own extrapolation folding over on itself. Almost always it is the latter — and knowing that prevents trading a number the fit itself invented.

You will notice the two diagnostics are the same statement told twice. Negative density is the condition seen in price space, through the curvature of calls. Non-monotone risk-reversal and butterfly are the condition seen in quote space, through the shape of the skew in delta. A surface that is free of butterfly arbitrage has a density that is everywhere non-negative and a butterfly term that behaves monotonically in the wings. When SVQ shows you both flags firing on the same strikes, you can be confident the problem is real and located. When only the density spikes but the RR and BF curves are smooth, suspect a numerical artifact in your differentiation. Cross-reading the two is how you avoid both false alarms and missed dislocations — and it is the difference between a quant who trusts the screen and one who interrogates it.