How One-Factor Short-Rate Models Generate Yield Curves
Summary
The document explains how a one-factor short-rate model can determine prices across many maturities even though it is named for the short rate. In a model such as Hull–White, discount bond prices at different maturities are functions of the modeled state variable and model parameters. The short-rate process also relates to forward rates through the term structure, so the model can produce a full curve rather than only one quoted point.
The key limitation is that a single stochastic factor constrains how the curve moves. The discussion notes that one-factor Gaussian models tend toward highly correlated rate changes and mainly parallel shifts, with some maturity-dependent effects. Thus, deriving a curve is distinct from capturing realistic curve dynamics. Multi-factor models can represent level, slope, and curvature changes more flexibly, but require more calibration and implementation effort; the document provides conceptual explanation rather than empirical calibration results.
Key ideas
- A short-rate model uses a state variable to determine bond prices across maturities.
- The name short-rate model does not mean that only one point on the yield curve is modeled.
- A one-factor model limits the range of possible yield-curve movements and tends toward correlated changes.
- Multi-factor models can represent level, slope, and curvature dynamics with greater complexity.
Tags
Full text
# Modeling the Yield Curve with Short-Rate Models
# Modeling the Yield Curve with Short-Rate Models
I have been reading Gregory Counterparty Credit Risk. In the book he criticizes one-factor short-rate models for their limited ability to capture realistic yield curve movements. While they allow for some steepening and flattening due to mean reversion, they primarily capture parallel shifts, making them restrictive for risk management. In contrast, multi-factor models (e.g., three-factor models) better reflect observed yield curve movements, accounting for level (parallel shifts), slope (twists), and curvature (butterfly movements). These models lead to significantly different risk measures, such as potential future exposure (PFE). However, they are more complex to implement and calibrate.
Question: While we can model the short rate using both one-factor and multi-factor models, when it comes to modeling the whole yield curve, how can a short-rate model simulate the entire curve if it only models the short-term rate (one point on the curve)?
## Answer by Andrea (score 2)
https://quant.stackexchange.com/a/82054
I see why the name "short rate" model can be confusing.
They really are "one factor model". This factor is for convenience identified with the short rate, but you could have used any other rate.
In HullWhite, every discount factor is function of the short rate
$P(t, T, r) = e^{A(t, T) - B(t, T) r}$
and so you could try to re-write the HW dynamics as a 10y-rate-one-factor-model.
Moreover, if you look at the HW dynamics in the HJM framework, you can see that all rates are on equal footing
$df(t,T)=\alpha(t,T)dt + \sigma e^{-\alpha (T-t)} dW_t$
for one single $dW_t$, which is what really matters.
Whether this is realistic or not, is a different question.
## Answer by Pedro (score 1)
https://quant.stackexchange.com/a/82051
From a purely mathematical point of view the answer is: if your short rate process is $r(t)$ then the instantaneous (HJM) forward rate is computable in terms of bond prices, which are computable in terms of the integral of the short rate.
But this does not guarantee at all any realistic evolution of the forward curve. For example, in the Hull-White one factor Gaussian model you can compute that up to some distortions due to maturity, you essentially get parallel shifts, and forwarded rates are (inevitably) perfectly correlated.
Note that the short rate is somewhat "special" (it is a derivative) so it contains more information than just a point, hence integrating it gives you a lot of information. So it is not really a point in the curve, but rather a point with some extra information (direction, if you will).
In the case of some HJM type models the single value $r(t)$ determines the forward curve, and this is an artifice of Markovian models, which might be what you observe? But I'm not sure if this is what you mean.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.