Markov-Functional Models: Calibration, Correlation, and Practice
Summary
The document describes Markov-functional interest rate models as low-dimensional representations of the term structure built from functions of a Markov process. It notes their ability to calibrate to market prices and capture volatility smiles, while raising the question of why they may appear less prominent than short-rate models or path-dependent market models. Responses suggest this perception may be misleading: practitioners use them, particularly for Bermudan swaptions, while academic publications may focus less on implementation details that are difficult to turn into research papers.
A technical caveat concerns calibration to swaption prices: in a Markov-functional model, externally specified distributions may produce implausible relationships between forward rates and swap rates. For Bermudan options, distorted joint distributions or correlations can affect exercise boundaries even if marginal distributions are recovered. The discussion is qualitative and offers no empirical comparison across model classes; it presents competing views on practical popularity and model limitations.
Key ideas
- Markov-functional models use a low-dimensional Markov process to describe term-structure variables.
- They can calibrate to market prices and capture volatility smiles, but may not reproduce instantaneous correlations well.
- Recovering marginal distributions does not ensure realistic joint distributions among rates.
- Implausible rate correlations can distort Bermudan swaption exercise boundaries.
- The discussion distinguishes practical dealer use from visibility in academic literature.
Tags
Full text
# Why Markov Functional Models (Hunt 2000) are not yet so popular? # Why Markov Functional Models (Hunt 2000) are not yet so popular? I refer to MFM introduced by Hunt [2000]. These models can be seen a subset of interest rate market models. MFM allow us to describe the term structure elements using a set a functions of a low-dimensional Markov process (say 1 or 2). This gives to the model the ability to calibrate fairly well and to capture the smile. Of course, due to limited number of risk factors can fail to capture the instantaneous correlation structures between rates. However, being low-dimensional, Markovian and relatively good with the smile it did not make it so popular yet. If indeed this is true. What do you see as the reason? Why do people still prefer short rate modelling (maybe with stochastic vol) or even the path-dependent BGM? Thank you in advance. ## Answer by Mark Joshi (score 2, accepted) https://quant.stackexchange.com/a/16188 it's difficult to say that they are not popular. Some people definitely use them for live pricing. I'd say the real question is "why are they not popular in the academic literature"? One answer would simply be that most the questions that arise in their use are ones of fiddliness which do not make good papers. ## Answer by Arshdeep (score 1) https://quant.stackexchange.com/a/53756 In context of Bermudan Options, I believe that since the model determines everything exogenously, calibrating to swaptions may give you cases where the implied forward rate is negatively correlated to swap rates. Note this will never happen in an endogenous model where the short rate equation constrains this possibility. This will obviously distort exercise boundaries, as roughly, swap rate is a linear combination of forwards, and thus the implied correlation between swap rates may be completely unrealistic. Simply put, the model recovers marginals but lack of endogenous structure makes it difficult to control the joint distributions. ## Answer by JUW (score 1) https://quant.stackexchange.com/a/73915 The Markov-functional model is widely used by dealers around the Street in particular for Bermudan swaptions. So we cannot say it is not popular. Of course, there has been some criticisms since its birth. But somehow it seems that the Bermduan swaption market does not collapse (at least not yet) due to use of the MF model (maybe because the model does not matter that much for an one-way market and franchise business like the Berm market).
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.