1Explain why options with many exercise dates need a model of how the whole curve moves, and why the short rate is the natural thing to model.
2Describe a mean-reverting short rate, its drift and volatility, and compute the half-life of a shock.
3Price a zero-coupon bond as the average discount factor over short-rate paths, by Monte Carlo and in closed form.
4Say what one factor implies: every yield moves with the short rate, by its loading, and all yields are perfectly correlated.
Why it matters
Module 5.2 priced each rate option on its own: one forward rate, one vol, Black or Bachelier. That works for an option with one exercise date. RiskAtoms Bank plc's mortgage borrowers can repay in any quarter, a callable bond can be called every year, and L5.2.08: Swaptions could only approximate the borrowers' option with a single swaption. To value a choice that can be made on many dates you need to know how the 2-year rate in three years relates to the 5-year rate in four: one model of how the whole curve moves. This module builds that model from one number, the short rate.
Real curveIllustrative dataThe sterling short rate at 31 December 2025
The curve is Module 5.2's: the Bank of England's fitted SONIA OIS curve at 31 December 2025, rebuilt from 23 quotes (L5.2.02: Bootstrapping a curve). 20 are derived from the Bank's curve or from SONIA derived; the 3 beyond 25 years (30Y, 40Y, 50Y) extend it illustrative. Today's short rate r₀ is SONIA's last fixing of 2025, 3.7257%, read on a secondary source secondary. Every model parameter on this page is either set here or fitted to that curve and to Module 5.2's illustrative vols, so every parameter is illustrative: a fit, not a market quote. Vasicek's σ is 90bp a year, the normal rate vol of Module 3.4's exposure model (L3.4.03: Exposure profiles of common trades) illustrative. Its κ and θ are a least-squares fit to the curve's zero rates, so they are derived from real data but belong to the model, not to the market. The simulations are seeded, so every visit prints the same numbers.
The short rate and the bank account
The short rateShort rate r(t): the interest rate for borrowing over the next instant at time t. SONIA, an overnight rate, is the closest rate the market quotes. r(t) is the rate for borrowing over the next instant. Nobody quotes it exactly, but SONIA, the overnight rate, is close: 3.7257% on 31 December 2025. Roll £1 over at the short rate and it grows to the bank accountBank account B(t): £1 deposited today and rolled over at the short rate every instant, worth exp(∫₀ᵗ r(s) ds) at t. It is the numeraire of the risk-neutral measure. B(T) = exp(∫₀ᵀ r). L5.2.05: Risk-neutral pricing and numeraires showed that a pound paid at T is worth today the risk-neutral average of 1/B(T): P(0, T) = E_ℚ[exp(−∫₀ᵀ r)]. So if we know how r can move under ℚ, every bond is an average over its paths, and so is every option on bonds or swap rates. That is a short-rate modelShort-rate model: a stochastic differential equation for r under the risk-neutral measure. Every bond and rate option is then an average over its paths..
Why the short rate and not, say, the 10-year rate? Because the bank account is built from it, and the bank account is the numeraire of ℚ. Model r and the discount factor along every path follows; the whole curve, today and in every future state, comes out of the same averages. The cost is that the model has to reproduce today's curve, which this module takes four lessons to do properly.
Drift, noise and mean reversion
An SDE (M-F02: Stochastic differential equations) gives r a drift and a random shock. Rates do not wander off like a stock price: central banks pull them back. So the drift is mean-revertingMean reversion: a drift that pulls a variable back towards a long-run level θ, faster the further away it is: κ(θ − r) in Vasicek., the model of Vasicek (1977):
dr=κ(θ−r)dt+σdW
Below θ the drift is positive, above it negative, and κ sets how hard it pulls. Take the expectation and the noise drops out: the mean m(t) solves the ODE dm/dt = κ(θ − m) of M-B06: Differential equations, so m(t) = θ + (r₀ − θ)e−κt. The gap to θ shrinks by the same fraction each year, and halves after the half-lifeHalf-life: the time for the expected gap to the long-run level to halve, ln 2/κ. A shock to today's rate is expected to be half gone after one half-life. ln 2/κ. With the parameters the next lesson fits to the SONIA curve, κ = 0.0549 and θ = 5.752%, the half-life is 12.6 years: after that long the expected short rate is 4.739%, halfway from 3.726% to θ. Twenty paths from today, monthly, with σ = 90bp:
Seed 53,002, exact steps. Solid: the mean, from 3.726% towards 5.752%; dashed: ± 2 standard deviations. At 10 years the mean is 4.581% and the standard deviation 2.22%, and 1.94% of the distribution is below zero (L5.3.02: The Vasicek model). illustrative
A bond as an average
Each path has its own area Y = ∫₀¹⁰ r and its own discount factor e−Y. Average them and you have the 10-year bond. Vasicek's r is a Gaussian process, so Y is normal and the average has a closed form; a simulation must agree with it, up to its standard errorStandard error: the standard deviation of a Monte Carlo average, the sample sd divided by √n. Halving it takes four times as many paths.:
Vasicek fitted to the curve (L5.3.02), σ = 90bp. Both simulations land within one standard error. Discounting at the average path misses: the average of e^(−Y) is above e^(−average Y).
The 10-year bond, P(0, 10)
Value
Standard error
SEs from closed form
Closed form, e^(−m + v/2)
0.663579
20,000 exact draws of Y (seed 53,001)
0.664002
0.000638
0.66
2,000 paths, monthly, Y by the trapezoid rule
0.664823
0.002038
0.61
The discount factor of the average path, e^(−m)
0.657549
The SONIA curve's P(0, 10)
0.670590
Two lessons sit in this table. The fourth row is below the first by a factor ev/2 = 1.0092: uncertainty in rates raises bond prices, the convexityConvexity (here): the average of exp(−Y) is above exp(−average Y), because exp(−x) curves upwards. For a normal Y the factor is exactly exp(v/2). of the discount factor. And the last row is not the first: the model's 10-year yield is 4.101%, the curve's 3.996%. Three numbers (κ, θ, σ) cannot reproduce a whole curve. L5.3.02: The Vasicek model measures how far off they are, and L5.3.04: Fitting today's curve: Ho-Lee fixes it.
One factor: every yield moves together
In Vasicek every zero yield is a straight line in r: y(τ) = a(τ) + (B(τ)/τ) r, with B(τ) = (1 − e−κτ)/κ. When r moves, each yield moves by its loadingLoading: how far a yield moves when the short rate moves by 1. In Vasicek the τ-year zero yield moves by B(τ)/τ, with B(τ) = (1 − exp(−κτ))/κ. B(τ)/τ. Long yields move less, because the market expects a shock to revert before they mature. The house Hull-White model of L5.3.05: The Hull-White model has much weaker mean reversion (1.60%), so its loadings stay near 1 for longer:
Zero yields, continuously compounded. Every row moves in step with r: the correlation between any two yield changes is 1.
Maturity τ
1y
2y
5y
10y
30y
Vasicek loading, κ = 0.0549
0.973
0.947
0.875
0.770
0.490
Hull-White loading, a = 1.60%
0.992
0.984
0.961
0.924
0.794
Vasicek yield move, r +50bp
48.7bp
47.4bp
43.7bp
38.5bp
24.5bp
That is the price of a one-factor modelOne-factor model: one source of randomness drives the whole curve. Every yield is a function of the short rate, so all yields move together.: the correlation between changes in the 1-year and 10-year yields is exactly 1. Real curves twist: the two-factor G2++ model fitted in L5.3.10: Two-factor models gives 0.153. For a Bermudan swaption or a callable bond, where one rate decides the exercise and another the payoff, that matters; for most of this module one factor is enough.
Sort itWhat does one factor imply?
For each statement about a one-factor short-rate model such as Vasicek, decide whether it is always true, true only of some one-factor models, or false.
Every zero-coupon bond price is a function of today's short rate alone
The 2-year and 30-year yields can move in opposite directions on the same day
Rates can go below zero
Today's curve is reproduced exactly
A bond price is the risk-neutral average of exp(−∫r) over the paths
The formal version
A bond as an expectation in Vasicek
Under ℚ, with r₀ today's short rate (Vasicek 1977; Brigo and Mercurio 2006, chapter 3):
κ is the speed of mean reversion (a year), θ the long-run level, σ the normal volatility (decimal a year), W a ℚ-Brownian motion. The second line solves the SDE exactly (M-F02: Stochastic differential equations, the Ornstein-Uhlenbeck process); its first two terms are the mean, the solution of the ODE dm/dt = κ(θ − m) of M-B06: Differential equations. Y is an integral of a Gaussian process, hence normal, and E[e−Y] = e−m + v/2 for any normal Y. The full bond formula P(t, T) = A(τ)e−B(τ)r(t) is the next lesson's; here a(τ) is −ln A(τ)/τ.
Try it
The model starts at today's SONIA with the fitted κ, θ and σ = 90bp. Move κ and watch the band narrow and the half-life fall; move the maturity and compare the simulated bond with the closed form and with the curve.
Try itShort-rate paths and a bond as their average
Paths
Twenty paths of r (one highlighted), the mean (solid), ± 2 standard deviations (dashed) and θ (dotted). The red line is zero.
P(0, 10) closed form0.66358
Path by path, ∫r by trapezoid ± SE0.66482 ± 0.00204
SEs from the closed form0.61
Exact draws of ∫r ± SE0.66486 ± 0.00200
The curve's P(0, T)0.67059
Half-life ln 2/κ12.6 years
Mean of r at 10y4.581%
Sd of r at 10y2.218%
P(r < 0) at 10y1.94%
10y yield loading B(T)/T0.770
Set σ to zero: every path is the mean, and the bond is one discount factor. Raise κ and the paths are pulled to θ faster, the band narrows and the loading of long yields falls: a shock to today's rate is forgotten before the bond matures. Fewer paths widen the error bars; neither estimate drifts from the closed form. The model's bond and the curve's differ: three numbers cannot fit the whole curve, as the next lesson shows.
Worked example
The 10-year bond as an expectation
Vasicek from SONIA at 31 December 2025, the 10-year zero-coupon bond
Step 1 of 6
1
The model
r₀ = 3.7257% (SONIA), κ = 0.054862, θ = 5.7517%, σ = 90bp a year: the fit of L5.3.02: The Vasicek model. T = 10 years. illustrative
Regulatory lens · looking ahead
A short-rate model used to value positions is a model in the sense of the PRA's supervisory statement SS1/23, Model risk management principles for banks (published 17 May 2023, effective 17 May 2024; the current version is dated 23 April 2026). It applies to UK banks with permission to use internal models for capital, and covers every model they use to inform business decisions, including models used for financial reporting, under five principles: model identification and model risk classification, governance, model development, implementation and use, independent model validation, and model risk mitigants. A one-factor model that misses today's curve by tens of basis points, as this one does, is exactly the kind of limitation validation must find and record. Validation itself is M6.7: Model risk, validation and machine learning; the exposure simulations behind XVA, which run short-rate models like this one forward, are M6.5: Counterparty risk and XVA.
Spark Atoms
✦Spark AtomThe same Vasicek, three times+
Oldřich Vašíček, born in Czechoslovakia in 1942, joined the management science department of Wells Fargo Bank in January 1969, after leaving his country following the 1968 invasion. His 1977 paper gave this module its first model. You have met him twice already: the precision-weighted beta of L3.1.10: Using and testing CAPM and the one-factor credit model of L3.3.11: Unexpected loss and default correlation. In 1989 he co-founded KMV, the credit-risk firm Moody's bought in 2002.
✦Spark AtomUncertainty raises bond prices+
Discounting at the average path gives 0.6575 for the 10-year bond; averaging the discount factors gives 0.6636. The gap, a factor of 1.0092, is worth 9.1bp of yield. Its logarithm, v/2, grows roughly with σ² and, while κT is small, with the cube of maturity: one reason long yields sit below where expected rates alone would put them.
Key takeaways
Options with many exercise dates need one model of how the whole curve moves; a short-rate model gives it, because the bank account is built from r.
A bond is the risk-neutral average of exp(−∫r) over the short rate's paths; every rate option is an average of the same kind.
Mean reversion pulls r towards θ; the mean solves dm/dt = κ(θ − m), and a shock halves in ln 2/κ: 12.6 years for the fitted Vasicek.
Monte Carlo agrees with Vasicek's closed form within its standard error (0.66400 ± 0.00064 against 0.66358), but the model misses the curve's 0.67059.
One factor: every yield moves with r by its loading B(τ)/τ, so all yields are perfectly correlated.
Reaction
Check what stuck
Question 1 · objective 2
A short-rate model has mean reversion κ = 0.10 a year. After how many years is a shock to today's rate expected to be half gone?
years
Question 2 · objective 2
Today's short rate is 3.7257%. In a model with θ = 5% and κ = 0.10, what is the expected short rate in 5 years, in %?
%
Question 3 · objective 3
20,000 draws give the 10-year bond as 0.664002 with a standard error of 0.000638. The closed form is 0.663579. How many standard errors is the simulation from it?
Question 4 · objective 4
In a one-factor short-rate model such as Vasicek, what is the correlation between daily changes in the 1-year and the 10-year zero yields?
Question 5 · objective 4
In the fitted Vasicek model the 10-year zero yield has a loading B(10)/10 = 0.7697. If the short rate jumps by 50bp, how far does the 10-year yield move, in bp?
bp
Glossary
Short rate
Short rate r(t): the interest rate for borrowing over the next instant at time t. SONIA, an overnight rate, is the closest rate the market quotes.
Bank account
Bank account B(t): £1 deposited today and rolled over at the short rate every instant, worth exp(∫₀ᵗ r(s) ds) at t. It is the numeraire of the risk-neutral measure.
Short-rate model
Short-rate model: a stochastic differential equation for r under the risk-neutral measure. Every bond and rate option is then an average over its paths.
Mean reversion
Mean reversion: a drift that pulls a variable back towards a long-run level θ, faster the further away it is: κ(θ − r) in Vasicek.
Half-life
Half-life: the time for the expected gap to the long-run level to halve, ln 2/κ. A shock to today's rate is expected to be half gone after one half-life.
Loading
Loading: how far a yield moves when the short rate moves by 1. In Vasicek the τ-year zero yield moves by B(τ)/τ, with B(τ) = (1 − exp(−κτ))/κ.
One-factor model
One-factor model: one source of randomness drives the whole curve. Every yield is a function of the short rate, so all yields move together.
Standard error
Standard error: the standard deviation of a Monte Carlo average, the sample sd divided by √n. Halving it takes four times as many paths.
Convexity
Convexity (here): the average of exp(−Y) is above exp(−average Y), because exp(−x) curves upwards. For a normal Y the factor is exactly exp(v/2).
J. Hull and A. White, “Pricing interest-rate-derivative securities”, Review of Financial Studies 3(4), 1990: its opening sections survey the short-rate models of this module and their drawbacks. Intermediate.
D. Brigo and F. Mercurio, Interest Rate Models: Theory and Practice, 2nd ed., Springer, 2006: chapter 3, one-factor short-rate models. Advanced.
Origins
O. A. Vasicek, “An equilibrium characterization of the term structure”, Journal of Financial Economics 5(2), 177–188, 1977.
J. Hull and A. White, “Pricing interest-rate-derivative securities”, Review of Financial Studies 3(4), 573–592, 1990.
Prudential Regulation Authority, SS1/23 Model risk management principles for banks, May 2023 (current version 23 April 2026).
Wikipedia, “Oldřich Vašíček” (biography), read 7 November 2026.
Bank of England, UK OIS yield curve daily data, 31 December 2025.
Watch this lesson
Video version
Prefer to watch? The same lesson as a 14-minute video will sit here, made from this page's script.