The Vasicek model represents the instantaneous short rate as a one-factor Gaussian process with constant volatility and mean reversion.
The Vasicek interest rate model represents the instantaneous short rate as a one-factor Gaussian process that tends to revert toward a long-run level. It is used to study interest-rate dynamics and, under risk-neutral parameters, to derive arbitrage-consistent zero-coupon bond prices.
The model is analytically convenient, but its constant parameters, single source of risk, and normally distributed short rate are restrictive. In particular, the basic Vasicek model permits negative interest rates.
The standard Vasicek process is:
where:
| Symbol | Meaning | Required interpretation |
|---|---|---|
| (r_t) | Instantaneous short rate at time (t) | A model state variable, not necessarily an observed policy or quoted market rate |
| (\kappa>0) | Mean-reversion speed | Larger values pull the rate toward (\theta) faster |
| (\theta) | Long-run mean level | Depends on whether the process is under the real-world or pricing measure |
| (\sigma>0) | Instantaneous volatility | Constant in the basic model |
| (W_t) | Wiener process | Source of continuous Gaussian shocks |
The drift (\kappa(\theta-r_t)) is positive when the short rate is below (\theta) and negative when it is above (\theta).
For horizon (\tau=T-t), the conditional expected short rate is:
The conditional variance is:
As (\tau) increases, the conditional mean approaches (\theta). The variance approaches the long-run value (\sigma^2/(2\kappa)). Because the distribution is Gaussian, it has some probability of producing negative rates for any positive volatility, although that probability may be small for a particular parameter set.
Assume hypothetical annualized parameters:
| Input | Value |
|---|---|
| Current short rate, (r_0) | 5.00% |
| Long-run level, (\theta) | 3.00% |
| Mean-reversion speed, (\kappa) | 0.60 |
| Volatility, (\sigma) | 1.00 percentage point per square-root year |
| Horizon, (\tau) | 1 year |
The conditional mean after one year is:
The conditional standard deviation is:
The model therefore centers the one-year distribution near 4.10% with a standard deviation of about 0.76 percentage point. The 4.10% is a conditional model mean, not a guaranteed forecast. A different probability measure or parameter estimate can produce a different path.
The same process form can be written under different probability measures:
| Measure | Typical use | Parameter interpretation |
|---|---|---|
| Real-world, (\mathbb{P}) | Forecasting, scenario analysis, and statistical estimation | Drift reflects the estimated physical evolution of rates |
| Risk-neutral, (\mathbb{Q}) | Pricing bonds and interest-rate derivatives | Drift is adjusted so discounted tradable prices satisfy no-arbitrage conditions |
Under (\mathbb{Q}), write:
The risk-neutral long-run level (\theta_Q) need not equal the historical long-run estimate. The market price of interest-rate risk links the two measures under additional assumptions.
Using historical estimates directly in a pricing formula can therefore misprice bonds or derivatives. Conversely, parameters calibrated to market prices do not automatically provide unbiased real-world forecasts.
Under constant risk-neutral parameters, the price at time (t) of a default-free zero-coupon bond paying $1 at maturity (T) has the exponential-affine form:
where (\tau=T-t):
and:
The notation (B) in the final expression means (B(t,T)). The continuously compounded model yield is:
Bond valuation requires risk-neutral parameters and consistent compounding. It also assumes the modeled short rate is the relevant default-free discounting state variable. Credit, liquidity, collateral, and instrument-specific cash flows require separate treatment.
flowchart LR
A["Define use: forecast or pricing"] --> B["Choose real-world or risk-neutral measure"]
B --> C["Select short-rate and market data"]
C --> D["Estimate or calibrate kappa, theta, and sigma"]
D --> E["Generate rate paths or bond values"]
E --> F["Back-test, reprice, and stress model limits"]
For forecasting, analysts may estimate the process from a time series of rate observations. For pricing, they may calibrate risk-neutral parameters to a yield curve and derivative prices. The basic Vasicek model generally cannot fit every observed curve and volatility surface exactly with one constant parameter set.
Important implementation choices include:
| Model | Mean reversion | Volatility structure | Negative rates | Curve fit |
|---|---|---|---|---|
| Vasicek | Constant long-run level | Constant | Permitted | Restricted by constant parameters |
| Hull-White one-factor | Time-dependent drift | Often constant short-rate volatility | Permitted | Can be fitted to the initial term structure |
| Cox-Ingersoll-Ross | Constant long-run level | Proportional to (\sqrt{r_t}) | Designed to preserve nonnegative rates under standard conditions | Restricted without extensions |
The models answer different implementation needs. Preventing negative rates, fitting today’s curve, and reproducing option volatility can require different specifications or extensions.
This article provides general fixed-income and financial-modeling education. Model outputs are not guaranteed forecasts, executable valuations, hedge results, or personalized investment advice.