Vasicek Interest Rate Model

Why use it?
– Analytic tractability: closed‑form bond prices and transition density make calibration, simulation and pricing simpler.
– Good as a first model for interest‑rate dynamics and for pedagogical purposes.
– Useful for pricing interest‑rate futures, zero‑coupon bonds and some derivatives under simplifying assumptions.

Important formulas (practical)
1) Transition (exact) distribution (for Δ = T − t):
r(t+Δ) = r(t) e^{−aΔ} + b (1 − e^{−aΔ}) + ε
with ε ~ N(0, Var) and
Var = (σ^2 / (2a)) (1 − e^{−2aΔ}).

This exact update is extremely useful for simulation and calibration.

2) Zero‑coupon bond price (risk‑neutral measure)
Under the risk‑neutral measure Q (market price of risk λ adjusts the drift), the price at t of a zero‑coupon bond maturing at T is
P(t,T) = A(t,T) exp[ − B(t,T) r(t) ],
where
B(t,T) = (1 − e^{−a (T−t)}) / a,
and
ln A(t,T) = (b_Q − σ^2/(2 a^2)) ( B(t,T) − (T−t) ) − (σ^2 B(t,T)^2 ) / (4 a).

Here b_Q is the long‑run mean under Q (b_Q = b − λ σ / a if market price of risk λ is constant). See Vasicek (1977) and standard term‑structure texts for derivations.

Practical steps: model selection, calibration, simulation, pricing
Follow these numbered, actionable steps when implementing Vasicek in practice.

Step 0 — Data and objective
– Decide objectives: yield curve fitting, risk management (scenario simulation), pricing specific derivatives, or academic analysis.
– Gather data: short‐rate series r_t (if available), or use a proxy (overnight rate, money‑market rates), and market prices of liquid bonds/notes for cross‑sectional fitting.

Step 1 — Choose measure (P vs Q)
– Physical/real world measure P: useful for forecasting and risk modelling (parameters a, b, σ estimated on historical rates).
– Risk‑neutral measure Q: required for arbitrage‑free pricing of bonds/derivatives (use market instruments to infer b_Q or market price of risk λ).

Step 2 — Estimate parameters (calibration)
A. Time‑series (historical) estimation (estimating a, b, σ under P)
– Use the exact discretization implied by the OU process:
r_{t+Δ} = r_t e^{−aΔ} + b(1 − e^{−aΔ}) + η_t, η_t ~ N(0, Var)
– This is a linear Gaussian model → you can estimate via Maximum Likelihood (closed form) or OLS after rearranging into an AR(1) form:
r_{t+Δ} = φ r_t + µ + η_t, where φ = e^{−aΔ}, µ = b(1 − φ)
From estimated φ, µ, and residual variance s^2, recover a = −ln φ / Δ, b = µ / (1 − φ), and σ from Var formula:
σ = sqrt( 2 a s^2 / (1 − φ^2) ) (use exact variance mapping).
– Kalman filter: when short rate is unobservable (you observe yields), cast as state‑space and run Kalman filter/MLE (common in practice).

B. Cross‑sectional calibration to market bond prices (risk‑neutral)
– Fit A(t,T) and B(t,T) to observed P(t,T) across maturities; infer r(t) and b_Q (or jointly fit a, b_Q, σ and instantaneous r(t)). Use nonlinear optimization to minimize pricing errors.
– If market prices (or yields) at multiple maturities are available, calibration under Q is straightforward due to closed‑form formulas.

Step 3 — Simulation
– Use the exact transition formula rather than Euler‑Maruyama for stability:
r_{t+Δ} = r_t e^{−aΔ} + b(1 − e^{−aΔ}) + sqrt( (σ^2 / (2a)) (1 − e^{−2aΔ}) ) * Z, Z ~ N(0,1)
– Simulate many paths for scenario analysis, VaR, stress tests or Monte Carlo pricing.

Step 4 — Pricing bonds and simple derivatives
– Use closed form P(t,T) = A e^{−B r(t)} with A,B above.
– Coupon bond: price as sum of discounted cash flows using P(t,ti).
– Options on bonds: may require numerical integration or transformation; for some vanilla cases you can use Black‑type approximations or Monte Carlo. For bond options with Gaussian short rate, closed‑form (normal) approximations exist.

Step 5 — Pricing interest‑rate derivatives (practical)
– Caps/floors/futures: Vasicek can be used but may mis‑price if caplet volatilities are nonconstant or if rates cannot go negative under model assumptions you prefer.
– Use Monte Carlo or analytic formulas under Q when available. For complex payoffs, simulate short‑rate paths and discount under Q.

Step 6 — Diagnostics and model validation
– Backtest: Compare model‑implied yields and volatilities to observed ones.
– Residuals: Check normality, heteroskedasticity and autocorrelation of residuals from time‑series fit.
– Sensitivity: Examine how pricing changes with parameter shifts; check if negative rates implied are plausible for your market.

Common calibration pitfalls and fixes
– Short‑rate unobservability: when the “short rate” is not directly observable, use a state‑space approach (Kalman filter) and include measurement equations linking r(t) to observed yields.
– Time‑varying parameters: if a, b or σ are clearly nonstationary, consider time‑dependent versions (Hull‑White) or multi‑factor models.
– Negative‑rate implications: if you need strictly positive rates, consider Cox–Ingersoll–Ross (CIR) or shifted models.

Strengths and limitations
Strengths
– Tractable: closed‑form bond prices and Gaussian transition density.
– Easy to calibrate and simulate (exact discretization).
– Good for pedagogical and initial modelling.

Limitations
– Allows negative rates (which may be unrealistic historically, though negative rates have occurred in some markets).
– Constant volatility σ — can’t reproduce observed term‑structure of volatilities or implied volatility smiles.
– Single factor — cannot capture richer correlation dynamics of yields of different maturities.
– Gaussianity implies symmetric rate changes and thin tails relative to some markets’ behavior.

Common alternatives and extensions
– Cox–Ingersoll–Ross (CIR) model: square‑root diffusion, enforces positivity.
– Hull–White (extended Vasicek): time‑dependent mean level (b(t)) or volatility σ(t) for exact fit to initial yield curve.
– Multi‑factor models (e.g., two‑factor Gaussian G2++): capture richer dynamics of the term structure.
– Heath‑Jarrow‑Morton (HJM) framework: models the forward curve directly.

Practical Python implementation sketch
– Data preprocessing: obtain short rate series or use proxy (O/N), interpolate yields.
– Estimation using AR(1) mapping

1) Estimate φ and µ by regressing r_{t+Δ} on r_t and a constant.
2) Compute a = − ln φ / Δ, b = µ/(1 − φ).
3) Compute σ from residual variance s^2:
σ = sqrt( 2 a s^2 / (1 − φ^2) ).

• Simulation (exact update):
for each step:
phi = exp(−a Δ)
mean = r * phi + b * (1 − phi)
var = (σ^2/(2*a)) * (1 − phi^2)
r_new = mean + sqrt(var) * np.random.normal()

(Use Kalman filter if r is latent and you have cross‑sectional yields.)

Use cases and practical advice
– Use Vasicek when you want an analytically tractable baseline, fast simulation, or closed‑form bond pricing.
– For trading/pricing liquid interest‑rate options or when positivity is essential, consider more complex models (Hull‑White, CIR).
– Always validate by comparing model implied yields and volatilities to market data; treat Vasicek as a parsimonious starting point, not the final answer for all markets.

References and further reading
– Vasicek, O. (1977). “An equilibrium characterization of the term structure.” Journal of Financial Economics, 5, 177–188.
– Cox, J., Ingersoll, J., & Ross, S. (1985). “A Theory of the Term Structure of Interest Rates.” Econometrica, 53(2), 385–407.
– Hull, J., & White, A. (1990/2000). Hull–White extensions and calibration techniques.
– Lesniewski, A. (2019). Lecture notes on Short Rate Models (Baruch College).
– Investopedia. “Vasicek Model” (overview).
– World Economic Forum. “Negative Interest Rates” (context about negative rates in practice).

Summary checklist (one‑page)
– Define objective: forecasting vs pricing.
– Choose measure: P for history, Q for pricing.
– Estimate a, b, σ (and λ if price of risk matters) via AR(1)/MLE/Kalman.
– Use exact transition for simulation.
– Price bonds via P(t,T)=A e^{−B r(t)}.
– Validate: backtest yields and residuals; consider model upgrades (CIR, Hull‑White, multi‑factor) if required.

Editor’s note: The following topics are reserved for upcoming updates and will be expanded with detailed examples and datasets.