Swaption Volatility Cube

An equity option's implied volatility depends on two things: how long until expiry, and how far the strike sits from the current price. Plot those two against volatility and you get a surface, like a topographic map. A swaption adds a third dimension: the underlying itself isn't a single stock price, it's an entire swap, and swaps come in different lengths — a swaption expiring in 1 year can be an option on a 2-year swap or a 10-year swap, priced completely differently. Stack expiry, underlying swap tenor, and strike on three axes and the flat map becomes a solid block: the volatility cube.

The three axes

A point in the cube is written $\sigma(T_\text{exp}, T_\text{tenor}, K)$, and each dimension answers a separate question:

$$\sigma(T_\text{exp}, T_\text{tenor}, K) = \text{ATM vol}(T_\text{exp}, T_\text{tenor}) + \text{skew adjustment}(K).$$

In plain English: start from the at-the-money volatility for a given expiry and swap tenor — a two-dimensional slice of the cube, called the ATM matrix — then add a skew term that shifts the volatility up or down depending on how far the strike $K$ sits from the at-the-money rate. Traders usually parametrize that skew slice with a model like SABR, fit separately at each expiry/tenor point, rather than storing every possible strike explicitly.

Worked example 1 — reading one cell

Suppose the ATM matrix shows a 5-year-into-10-year swaption (5Y expiry, 10Y underlying swap) has an at-the-money implied volatility of 22%. A SABR fit at that same expiry/tenor gives a skew such that a strike 100 basis points below the money (a receiver's floor-like strike) trades at 24.5% implied vol, while a strike 100bp above trades at 21%. A trader pricing a receiver swaption struck 100bp out-of-the-money must use 24.5%, not the 22% ATM number — using the wrong slice of the cube here understates the option's value, since receivers deep out-of-the-money on a downward-sloping skew are typically priced with higher vol, reflecting the market's pricing of tail risk in rates falling sharply.

Worked example 2 — comparing two cube points

Compare a 1-year-into-2-year ATM swaption at 28% vol against a 10-year-into-2-year ATM swaption at 17% vol — same underlying swap length, very different expiry. Converting both to an approximate annualized standard deviation of the 2-year swap rate over each option's life: $28\% \times \sqrt{1} = 28\%$ of the rate's level over one year, versus $17\% \times \sqrt{10} = 53.8\%$ cumulative over ten years. The shorter-expiry option has the higher annualized vol, but because it runs for less time, its cumulative uncertainty is smaller — a reminder that comparing raw vol numbers across expiries without adjusting for time horizon compares the wrong thing.

The surface above shows two of the cube's three axes (strike and tenor, at one fixed expiry) — imagine sliding a third slider for expiry and watching the whole surface reshape underneath it; that motion is the cube.

What this means in practice

Desks that trade swaptions, callable bonds, and Bermudan swaptions all consume the same cube, because every one of those products reduces to some combination of options on swap rates of specific expiries, tenors, and strikes. Building and maintaining the cube — interpolating between sparsely quoted points, fitting SABR smiles at each node — is a full-time modeling function on a rates desk.