Why HJM Forward Curves Form Infinite-Dimensional State Variables

Summary

The document explains why Heath-Jarrow-Morton interest-rate models are often described as infinite-dimensional. In the fixed-maturity formulation, the model specifies dynamics for the instantaneous forward rate at each maturity. Since maturities form a continuum, the full forward curve must be tracked as the evolving state, rather than a single rate. The Musiela reparametrization expresses that curve by time to maturity, making it a function-valued state variable.

The explanation uses the intuition that the model can have separate Brownian drivers across maturities, yielding infinitely many degrees of freedom. It also notes that special model specifications can reduce the dynamics to a finite-dimensional representation. The discussion is conceptual: it does not derive the stochastic equations, identify conditions for finite-dimensional realizations, or give examples of particular models, so it serves as an introduction rather than a technical treatment.

Key ideas

• HJM describes forward-rate dynamics across a continuum of maturities.
• The entire forward curve can be treated as the model’s state variable.
• Musiela reparametrization indexes the curve by time to maturity.
• Distinct stochastic drivers across maturities can make the model infinite-dimensional.
• Some special HJM specifications admit finite-dimensional representations.

Tags

# HJM in infinite dimensions


# HJM in infinite dimensions












I recently started reading Filipovic's Consistency problems for HJM interest rate models and came across the Musiela reparametrization

$$r_t(x)=f(t,x+t)$$ so the forward curve can be thought of as a map $x\to r_t(x)$. The book goes on to tell that this maybe thought of as an infinite dimensional state variable.

Does anyone have a good explanation for this? Is it because we pick $r_t$ from a space of functions?

Does anyone have a good explanation for this? Is it because we pick $r_t$ from a space of functions?

## Answer by Magic is in the chain (score 4, accepted)

https://quant.stackexchange.com/a/47304

Keeping it simple, you know HJM SDE gives the dynamics of an instantaneous forward referencing a fixed maturity T, $f\left(t, T\right)$, but there is a continuum of such maturities - the whole forward curve as a function of T.

You can have each forward driven by a different brownian for example, so in general the HJM approach will be infinite dimensional. to visualise infinite dimensions, it may be helpful to recall one dimensional SDE, two dimensional SDE, and so on. But there are special cases for which it an be viewed as finite dimensional.

Please also see the discussion here: Musiela parameterization

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.