On this page
In the Gauss+ model, the short rate moves towards a medium-term factor, that factor moves towards a long-term factor, and the long-term factor moves towards a fixed anchor. The targets therefore form a chain, rather than three separate returns to the same number. For an FRM® Part II candidate, the most useful test is to identify the target and sign of each factor’s drift before doing any arithmetic.
Why build a chain of rates?
A single short-rate factor can produce a compact story: today’s rate is pulled towards a long-run level while random news pushes it around. That story is useful, but one factor offers limited freedom when an option book responds to different parts of the yield curve. A short-dated cap and a long-dated swaption need not be explained by the same volatility setting. If the model is tuned to one maturity and misses another, the apparent hedge can be misleading even when the short-rate path looks plausible.
Gauss+ adds separate short, medium and long state variables, each with its own adjustment speed and shock size. The added structure permits richer curve and volatility behaviour. It does not guarantee a perfect fit, and a calibration still depends on the prices, conventions and data chosen. The point for an exam answer is the trade-off: more factors can represent more shapes, but they also require more parameters and careful interpretation. GARP places market risk measurement and management in the FRM Part II curriculum; the FRM programme overview helps locate that wider context.
Read the three targets before the shocks
Call the short factor r, the medium factor m and the long factor l. The short factor is pulled towards m. The medium factor is pulled towards l. The long factor is pulled towards a fixed level μ. A rate above its own target has a negative mean-reversion drift; a rate below its own target has a positive one. That rule works factor by factor. It is unsafe to look at the short rate and the final anchor alone and infer the short rate’s immediate drift.
The ordering of adjustment speeds is normally fastest for the short factor, slower for the medium factor and slowest for the long factor in the assigned Gauss+ reading. Think of the speed as the strength of the pull for a given gap, not as a promise that the short rate must move most in every realised interval. A large adverse shock can overwhelm the deterministic pull. A small gap can also produce a small drift even when its speed is high.
For each factor, separate the deterministic component from the random innovation. The deterministic component uses its own speed, the distance to its own target and the time step. The random component uses its own volatility and a shock scaled to the time step. The medium and long shocks may be correlated, so a scenario that moves them together is possible. Correlation says something about co-movement of shocks; it does not change which level is the mean-reversion target.
A small illustrative drift calculation
Suppose, solely as invented round numbers, that the short factor is 2%, the medium factor is 3%, the long factor is 2.5%, and the fixed anchor is 3%. Let the annual adjustment speeds for the short, medium and long factors be 2, 0.4 and 0.1 respectively. Consider a quarter of a year and ignore random shocks so that the direction is easy to see. These are teaching inputs, not observed rates or fitted parameters.
The short factor is one percentage point below its target of 3%. Multiplying that gap by its speed of 2 and a quarter-year step gives a positive deterministic change of 0.5 percentage points. Its illustrative next value, before a shock, is therefore 2.5%. The medium factor is half a percentage point above its target of 2.5%; multiplying the negative gap by 0.4 and one quarter gives a negative change of 0.05 percentage points. Its illustrative next value is 2.95%. The long factor is half a percentage point below the anchor; multiplying by 0.1 and one quarter gives a positive change of 0.0125 percentage points, leaving 2.5125%.
Notice what the calculation does not say. The three next values are not a forecast of market rates. We deliberately omitted shocks, used an elementary discrete time step and invented every input. In a full model, shocks can be positive or negative and the factors are recalculated together. The calculation isolates the exam skill: name each target, take target less current factor, multiply by the right speed and time step, then interpret the sign. If you reverse the medium-factor gap, the model’s economic story breaks even if your multiplication is flawless.
What the extra factors buy, and what they cost
The short factor reacts to information that affects the near end of the curve. The medium factor gives the short rate a moving destination, so an apparently high short rate can still be pulled upward if the medium factor is higher. The slow long factor lets that destination itself move over time. This is why a chain can produce richer dynamics than a one-factor return to a fixed constant. It is a structural description, not a claim that any latent factor is directly quoted on a screen.
A market yield for a particular maturity is an observable price-derived quantity; m and l here are model state variables. They may be estimated from an observed curve, but naming them ‘medium rate’ and ‘long rate’ does not make them identical to a particular bond yield. That distinction matters if a question asks which quantities are observed and which are inferred. The Omni methodology page is a useful place to pair this model explanation with deliberate practice in interpreting model assumptions.
Calibration is also different from a prediction. Fitting a volatility pattern across maturities can make present option prices more coherent under chosen inputs. It does not ensure that future movements will follow the model or that a hedge remains safe in every market state. A complex model can fit noise, and some instruments may provide weak information about a particular factor. For risk management, a fitted answer is a starting point for sensitivity and stress checks, not a substitute for them. This qualification also prevents the common overstatement that ‘three factors solve interest-rate risk’.
A direct comparison with a simpler Vasicek idea
A one-factor Vasicek-style model has one mean-reverting short rate and a fixed target. A BIS-hosted discussion of the Vasicek specification illustrates the idea of a short rate returning towards a long-run mean. Gauss+ keeps mean reversion but inserts moving intermediate targets. The useful comparison is therefore about target structure. In the simple model, one factor’s target is fixed. In the Gauss+ chain, the immediate short-rate target can change because the medium factor itself changes, and the medium factor’s target can change because the long factor changes.
Do not read that comparison as a ranking of models for every task. A simpler specification may be easier to estimate and explain. A richer specification may be useful when term-structure behaviour matters, but demands more calibration work and more careful validation. In an exam setting, first identify the model the stem actually gives. Importing a fixed Vasicek target into a Gauss+ question is a category error, just as importing a three-factor chain into a one-factor calculation is unnecessary.
How the moving targets change a scenario
Consider a second invented situation in which the short factor is 4%, the medium factor is 3% and the long factor is 5%. Even if the fixed anchor is only 2%, the short factor’s immediate deterministic pull is downward because its own target, the medium factor, is below it. At the same time the medium factor is pulled upward because its target, the long factor, is above it. The long factor is pulled downward towards the fixed anchor. This is a useful counterexample to the intuition that all rates should move in the same direction whenever the anchor is low. It shows why the chain must be read locally before making a claim about the whole curve.
If the medium factor subsequently rises far enough, the short factor’s target can move above the short factor and the sign of its drift can reverse. Nothing about that reversal requires a change in the fixed anchor. It follows from the interaction of the state variables. This makes the model capable of representing a rate environment in which the near end and the longer end adjust at different times. It also explains why memorising a single slogan such as ‘rates revert down’ is brittle. The direction is conditional on the current state of the entire cascade, and the realised rate path remains subject to shocks.
What calibration can and cannot establish
Suppose a risk manager adjusts the three volatility parameters to reproduce option prices at several maturities. A good match to those particular prices is evidence that the chosen parameters reproduce those observations under the model. It is not evidence that every unobserved maturity is priced accurately. The set of instruments used to calibrate the model matters, as do the conventions for measuring their prices. A different set of options may imply a different compromise. A candidate should be able to describe this limitation without pretending to compute a full calibration from the short reading.
Likewise, a richer volatility shape is possible because separate factors have different persistence and shock sizes, but ‘possible’ is not ‘automatic’. If the parameters are poorly chosen, the model can still miss the market pattern. Correlated innovations can change the joint behaviour of the medium and long factors, yet they do not replace the need to specify each factor’s own dynamics. In an exam answer, distinguish structural capability from an achieved empirical fit. This is a general model-risk discipline: identify what has been assumed, what has been estimated and what has actually been observed.
The mistakes worth checking at the end
The first trap is to send all three variables towards μ. Only the long factor has μ as its immediate target; the other targets are themselves stochastic. The second is to assume the fastest reversion speed means the biggest observed movement. Drift depends on both speed and gap, while realised movement also includes a shock. The third is to call the medium and long factors quoted market rates. They are state variables that help explain a curve. The fourth is to treat the model’s fit as proof of its future accuracy.
A compact check is to write three arrows on scrap paper: short towards medium, medium towards long, long towards anchor. Next to each arrow, mark whether the current factor is above or below its target. Then calculate only the drift that the question requests. If volatility or correlation appears, handle it separately from the target calculation. This order keeps an interpretation question from turning into an avoidable sign error.
Sources
- FRM Study Materials | GARP · garp.org · Retrieved
- How Important Is Sovereign Risk in Determining Corporate Credit Spreads in Emerging Markets? | BIS · bis.org · Retrieved



