SABR Model

Interest-rate options behave differently from equity options. At-the-money swaptions and caplets trade at lower implied volatilities than out-of-the-money equivalents—an inverted smile peculiar to fixed-income markets. The SABR model (Stochastic Alpha Beta Rho), introduced in 2002, prices these instruments by allowing both forward rates and volatility to drift stochastically, with parameters tailored to the leverage and mean-reversion properties of bond markets. It has become the market standard for interest-rate option pricing and Greeks, especially among dealers and hedge funds.

Why equity models fail for rates

The Black-Scholes model and even the Heston model were built with equities in mind. Equities have a natural lower bound (zero) but no obvious ceiling. Forward interest rates, by contrast, are level-constrained: they cannot go far negative (in most economies, central banks set a floor), and they rarely spike without economic dislocation.

Moreover, the implied volatility behaviour in rates markets is inverted compared to equities. In equities, out-of-the-money puts (far downside) trade at higher implied volatilities than at-the-money options. In rates, out-of-the-money swaptions—especially those far from the at-the-money strike—trade at lower implied volatilities. This reflects the belief that extreme moves in rates are less likely than extreme moves in equity prices.

The SABR framework: CEV and lognormal

SABR models the forward rate (F_t) using a constant-elasticity-of-variance (CEV) process:

(dF = \alpha F^\beta dW_F)

and volatility (\sigma_t) using a lognormal martingale:

(d\alpha = \nu \alpha dW_\sigma)

where (W_F) and (W_\sigma) are correlated Wiener processes with correlation (\rho), and (\nu) is the volatility of volatility.

The four parameters are:

• Alpha: the initial value of volatility; scales the model to current market levels
• Beta: elasticity of volatility to the forward rate (0 = normal, 1 = lognormal, between = CEV)
• Rho: correlation between forward-rate and volatility shocks; typically negative (rising rates = falling vol)
• Nu: volatility of volatility; how much the volatility process jitters

The CEV exponent and strike behaviour

The (\beta) parameter is crucial. When (\beta = 1), the forward follows lognormal dynamics (like Black-Scholes). When (\beta = 0), it is normal (arithmetic Brownian motion). Most practitioners calibrate (\beta) between 0.5 and 1, depending on the market and tenor.

A lower (\beta) means the forward’s volatility is higher at low rates and lower at high rates—this naturally produces the inverted smile. Out-of-the-money calls (high-strike swaptions) have lower implied volatility because the forward’s conditional volatility is low at elevated levels. This matches what traders observe and is one reason SABR became dominant.

Closed-form approximation

Unlike Heston, SABR does not have an exact closed-form solution. However, Hagan et al. (who published the model) provided an accurate approximation formula for the implied volatility as a function of strike, forward, and SABR parameters:

$$\sigma_{impl}(K) = \frac{\alpha}{F_0^\beta} \cdot z \cdot \frac{x(z)}{1 + \frac{\rho\nu\alpha}{4F_0^\beta} + …}$$

where (z) and (x(z)) encode the forward-strike relationship and the model’s dynamics. This approximation is accurate to within a few basis points and evaluates in microseconds, making it ideal for real-time hedging.

The option price itself is then computed using this implied volatility plugged into the Black-Scholes formula or a normal-volatility formula, depending on convention.

Calibration to swaptions and caps

In practice, a dealer calibrates SABR parameters to market prices of liquid swaptions and caps. Given observed implied volatilities across strikes and tenors, the quant team solves:

Minimize: Σ(SABR implied vol - market observed implied vol)²

Over the parameters (\alpha, \beta, \rho, \nu).

Once calibrated, SABR generates prices and Greeks for exotic or bespoke swaptions, bermudan swaptions (which allow exercise on multiple dates), and other exotics that don’t trade liquid prices.

Stability of calibration is important: parameters should not jump wildly day-to-day unless the market has genuinely shifted. Quants employ regularization (penalizing parameter changes) to achieve smooth, stable calibrations.

Greeks: delta, vega, and gamma

SABR Greeks are derived either from the closed-form approximation (via finite differences) or from Monte Carlo. The delta hedging ratio tells a dealer how much of the underlying forward to buy or sell to neutralize the option’s directional exposure.

Vega measures sensitivity to shifts in the initial volatility (\alpha). Dealers often quote swaption prices in implied-volatility terms (e.g., “25 basis points volatility”), so vega is how much the price changes if the market’s volatility quote shifts by 1 bp.

More subtly, SABR produces Greeks that vary with the strike—a phenomenon absent in Black-Scholes but present in real markets. An out-of-the-money swaption has different gamma (curvature) and theta (time decay) than an at-the-money one, and SABR captures this.

Strengths and limitations

Strengths:

• Explains inverted volatility smile and skew in rates markets
• Closed-form approximation is fast and accurate
• Parameters are economically intuitive
• De facto market standard; all major dealers and pricing systems support it

Limitations:

• The approximation formula breaks down far out-of-the-money (extreme strikes)
• Calibration is non-unique; different parameter sets can fit the same strikes equally well
• Beta is sometimes treated as fixed (not calibrated), introducing model risk
• Normal dynamics of rates near zero are not well-captured; some practitioners use a displacement parameter (shifted SABR) to fix this

Extensions: displaced SABR and multi-curve

Displaced SABR shifts the forward rate by a constant, allowing negative forwards to be modeled while maintaining lognormal dynamics above zero. This became necessary post-2008 when negative rates emerged in some economies.

Multi-curve SABR calibrates separate volatility surfaces to different basis curves (funding curves, projection curves), essential for pricing floating-rate notes and basis swaps.

See also