Pricing Swaptions - Which Models Do Trading Desks Use?

AlgoQuantHub Weekly Deep Dive

Author

Nicholas Burgess
July 17, 2026

Welcome to the Deep Dive!

Each week on The Deep Dive we explore cutting-edge ideas in algorithmic trading, quantitative research, and modern financial engineering, bridging theory and practice in how markets behave.

This week we explore the pricing of European, American and Bermudan swaptions. For European swaptions a change of numeraire turns pricing into the familiar Black-76 model. From there, we tackle the much harder world of Bermudan and American swaptions, where early exercise forces traders to abandon closed-form solutions in favour of numerical methods.

Table of Contents

Feature Article: Pricing European Swaptions

A swaption is an option to enter into an interest rate swap at a future date and predetermined fixed rate. They are essential tools for managing uncertainty in future interest rates and are heavily used in mortgage markets, where homeowners' ability to prepay or refinance fixed-rate mortgages creates embedded optionality that must be hedged.

At first sight, a swaption looks as though it should require a full interest rate model to price. Its payoff depends on an entire schedule of future cashflows, each discounted along a stochastic yield curve, so the natural instinct is to reach for Hull-White, LMM or another term-structure model before attempting a valuation. Yet the market prices vanilla European swaptions using what is, at heart, the same mathematics as Black-Scholes. The reason lies in one of the most elegant ideas in quantitative finance: change of numeraire.

The general pricing identity,

holds for any traded asset chosen as the numeraire. For a payer swaption, the payoff is

where A(T) is the swap annuity (the PV01 of the fixed leg) and S(T) is the forward par swap rate. The key observation is that the annuity itself is a traded portfolio of zero-coupon bonds. Choosing the annuity as the numeraire causes the awkward A(T) term to disappear from the expectation, leaving

At this point the problem suddenly looks much simpler. Assuming the swap rate follows a lognormal diffusion, the remaining expectation is exactly the familiar Black-Scholes integral, giving the Black-76 swaption formula.

Viewed this way, Black-76 is simply another member of the generalised Black-Scholes family. The only thing that changes between equities, FX, futures and interest-rate derivatives is the pricing measure—and therefore the cost of carry. Dividend-paying equities use b=r−q; FX uses b=r−rfb (Garman-Kohlhagen); futures use b=b0=0. Swaptions follow the same logic: under the annuity measure, the forward swap rate behaves exactly like a futures price, which is why Black-76 emerges naturally with the discount factor replaced by the annuity.

This perspective also explains how European swaption markets actually operate. Black-76 is not the model used to describe the evolution of interest rates—it is the market's quotation framework. Every point on the volatility cube (expiry, tenor and strike) is quoted under its own annuity measure, so the implied volatilities are not globally consistent. Since negative interest rates became commonplace in EUR, CHF and JPY markets, traders have also moved increasingly from lognormal Black volatility to normal (Bachelier) volatility, which naturally accommodates negative rates and expresses volatility directly in basis points. In practice, the first question on any trading desk is not "What is the volatility?" but "Which volatility convention are we using?"

For European swaptions, this framework is entirely sufficient. The market supplies the volatility, Black-76 supplies the price, and the annuity naturally defines the trade's DV01 and vega. The moment a product gains early exercise rights, however, the elegant shortcut disappears. Pricing now depends on how interest rates evolve through time, forcing us to choose an actual interest-rate model calibrated back to these same European prices. That raises a far more interesting question: which models should you learn, and which ones do trading desks actually use?

Keywords: Swaptions, Black-76, Change of Numeraire, Annuity Measure, Interest Rate Derivatives, Quantitative Finance, Generalized Black-Scholes

Bonus Article: Pricing American / Bermudan Swaptions

Once early exercise enters the picture, Black-76 reaches the end of its usefulness. Bermudan swaptions—the workhorses behind callable bonds, callable swaps and many structured products—derive much of their value from the optimal exercise decision itself. That decision depends on the future evolution of interest rates, so the pricing problem becomes one of dynamic optimisation rather than a simple expectation under a single measure.

To explore this, this week's implementation builds a complete C++ pricer for European, Bermudan and American swaptions using the Black-Derman-Toy (BDT) short-rate tree. BDT is no longer the model of choice on modern trading desks, but it remains one of the best learning tools because it extends the familiar Cox-Ross-Rubinstein equity tree directly into interest-rate modelling while exposing every stage of calibration.

BDT parameterises each node as

where the level aia_iai​ fits today's discount curve while the spacing σi\sigma_iσi​ controls the tree's volatility. Forward induction allows each time step to match two observable market quantities simultaneously: the discount curve (bootstrapped from OIS, deposits, FRAs/futures and swaps) and a volatility term structure derived from caps, floors or—preferably—co-terminal European swaptions, which provide the natural hedge instruments for Bermudan books.

Pricing then becomes a three-stage backward induction problem. First, calibrate the tree using Arrow-Debreu state prices. Second, roll back the underlying swap value from maturity to every earlier node. Finally, roll back the option itself, comparing continuation value with immediate exercise wherever exercise is permitted. European swaptions evaluate this comparison only at expiry, Bermudans at a finite set of exercise dates, and Americans at every node.

The weakness of BDT is also the reason it disappeared from production. Its volatility function does double duty: the slope of the volatility term structure implicitly determines the model's mean reversion. Fit a downward-sloping volatility strip and you unintentionally force stronger mean reversion, altering the very exercise premium you are trying to value. Calibration and dynamics become inseparable.

Modern production models deliberately separate these effects. Hull-White (LGM) remains the industry workhorse for callable products because mean reversion becomes an independent calibration parameter while Gaussian dynamics permit extremely fast calibration to European swaption prices. The cost is a flat volatility smile, although the allowance for negative interest rates—which once appeared unrealistic—became highly desirable during the negative-rate era.

When volatility skew becomes important, desks often move to Cheyette (quasi-Gaussian) models, which introduce local volatility while remaining Markovian enough to retain efficient lattice and PDE solvers. At the highest level sits the Libor Market Model (BGM/LMM), where the entire forward curve evolves consistently with market smile and correlation structures. The price for this realism is computational complexity: lattices become impractical, Monte Carlo takes over, continuation values are estimated using Longstaff-Schwartz regression, and sensitivities become slower and noisier to compute.

Trading desks use different models to price swaptions depending on their exercise features. The choice of model depends on the balance between calibration accuracy, computational efficiency, and risk management requirements. The practical hierarchy is therefore quite different from the order in which many people encounter these models academically. Learn CRR to understand lattice mechanics. Learn BDT to understand yield-curve calibration and early exercise. Price most callable products using Hull-White/LGM. Move to Cheyette when volatility skew materially affects exercise decisions. Reserve LMM for problems where full forward-rate dynamics and correlation genuinely drive the trade.

Our implementation illustrates the expected ordering of option values for a 5×5 ATM payer swaption with unit notional:

  • European: 0.0260

  • Bermudan: 0.0286

  • American: 0.0293

Full compilable C++ source code, together with a test verification suite, is available here. The European tree price also sits within 1% of the Black-76 benchmark, providing a useful validation of the implementation before introducing early exercise.

For readers looking to explore these models in greater depth, more complete implementation frameworks, production-quality pricing libraries and structured trading examples are available in my AlgoQuant Playbooks and professional trading toolkits.

Keywords: Bermudan Swaptions, American Swaptions, Black-Derman-Toy, Hull-White, LGM, Cheyette, Libor Market Model, Longstaff-Schwartz, Interest Rate Modelling

AlgoQuant Playbooks & Trading Toolkits

Explore my quantitative trading and financial markets toolkit store, featuring practical implementation frameworks for yield curve construction, derivatives pricing, risk modelling, and live market trading.

Feedback & Requests

I’d love your feedback to help shape future content to best serve your needs. You can reach me at [email protected]