Interest Rate Models
Why Interest Rate Models?
Interest rate modeling is fundamentally different from equity modeling due to the term structure — the relationship between interest rates and time to maturity. Unlike a single stock price, we must model an entire curve of rates simultaneously while ensuring no-arbitrage relationships between different maturities.
Interest rate models are essential for pricing bonds, swaps, caps, floors, swaptions, and other fixed-income derivatives. They also drive
risk management for banks, insurance companies, and pension funds with significant interest rate exposure.
Key Challenges in Interest Rate Modeling
1. Mean Reversion
Interest rates exhibit strong mean reversion — very high or very low rates tend to revert toward long-run averages. This contrasts with stock prices, which can trend indefinitely.
2. Term Structure Consistency
Models must simultaneously fit the entire yield curve and ensure that bond prices satisfy no-arbitrage conditions across all maturities.
3. Multiple Factors
The yield curve's complex dynamics require multi-factor models to capture parallel shifts, steepening/flattening, and butterfly movements.
4. Non-Negative Rates
Interest rates should generally remain non-negative (though recent experience with negative rates has challenged this assumption).
Short Rate Models
Framework
Short rate models specify the dynamics of the instantaneous short rate
r t r_t r t , from which bond prices and other derivatives are derived.
Bond prices satisfy:
P ( t , T ) = E Q [ e − ∫ t T r s d s ∣ F t ] P(t,T) = \mathbb{E}^{\mathbb{Q}}\left[e^{-\int_t^T r_s ds} | \mathcal{F}_t\right] P ( t , T ) = E Q [ e − ∫ t T r s d s ∣ F t ]
Vasicek Model (1977)
SDE :
d r t = κ ( θ − r t ) d t + σ d W t dr_t = \kappa(\theta - r_t)dt + \sigma dW_t d r t = κ ( θ − r t ) d t + σ d W t
Features :
Linear mean reversion toward long-run rate θ \theta θ
Gaussian rates (can be negative)
Analytical bond pricing formulas
Bond Price :
P ( t , T ) = A ( t , T ) e − B ( t , T ) r t P(t,T) = A(t,T)e^{-B(t,T)r_t} P ( t , T ) = A ( t , T ) e − B ( t , T ) r t
where:
B ( t , T ) = 1 − e − κ ( T − t ) κ B(t,T) = \frac{1-e^{-\kappa(T-t)}}{\kappa} B ( t , T ) = κ 1 − e − κ ( T − t )
A ( t , T ) = exp ( ( B ( t , T ) − T + t ) ( σ 2 / 2 κ 2 − θ κ ) κ − σ 2 B ( t , T ) 2 4 κ ) A(t,T) = \exp\left(\frac{(B(t,T) - T + t)(\sigma^2/2\kappa^2 - \theta\kappa)}{\kappa} - \frac{\sigma^2 B(t,T)^2}{4\kappa}\right) A ( t , T ) = exp ( κ ( B ( t , T ) − T + t ) ( σ 2 /2 κ 2 − θ κ ) − 4 κ σ 2 B ( t , T ) 2 )
Cox-Ingersoll-Ross (CIR) Model (1985)
SDE :
d r t = κ ( θ − r t ) d t + σ r t d W t dr_t = \kappa(\theta - r_t)dt + \sigma\sqrt{r_t} dW_t d r t = κ ( θ − r t ) d t + σ r t d W t
Features :
Square-root diffusion ensures non-negative rates (under Feller condition)
Mean reversion like Vasicek
Analytical tractability via affine structure
Feller Condition :
2 κ θ ≥ σ 2 2\kappa\theta \geq \sigma^2 2 κ θ ≥ σ 2 prevents the process from reaching zero.
Bond Price : Similar affine form as Vasicek but with different
A ( t , T ) A(t,T) A ( t , T ) and
B ( t , T ) B(t,T) B ( t , T ) .
Hull-White Model (1990)
SDE :
d r t = [ θ ( t ) − κ r t ] d t + σ d W t dr_t = [\theta(t) - \kappa r_t]dt + \sigma dW_t d r t = [ θ ( t ) − κ r t ] d t + σ d W t
Features :
Time-dependent mean reversion level θ ( t ) \theta(t) θ ( t )
Perfect fit to initial yield curve
Analytical formulas for bonds and options
Calibration :
θ ( t ) \theta(t) θ ( t ) is chosen to match the observed forward curve:
θ ( t ) = ∂ f ( 0 , t ) ∂ t + κ f ( 0 , t ) + σ 2 2 κ ( 1 − e − 2 κ t ) \theta(t) = \frac{\partial f(0,t)}{\partial t} + \kappa f(0,t) + \frac{\sigma^2}{2\kappa}(1-e^{-2\kappa t}) θ ( t ) = ∂ t ∂ f ( 0 , t ) + κ f ( 0 , t ) + 2 κ σ 2 ( 1 − e − 2 κ t )
Multi-Factor Models
Two-Factor Vasicek
d r t = ( ϕ 1 Y 1 , t + ϕ 2 Y 2 , t ) d t d Y 1 , t = − κ 1 Y 1 , t d t + σ 1 d W 1 , t d Y 2 , t = − κ 2 Y 2 , t d t + σ 2 d W 2 , t \begin{aligned}
dr_t &= (\phi_1 Y_{1,t} + \phi_2 Y_{2,t}) dt \\
dY_{1,t} &= -\kappa_1 Y_{1,t} dt + \sigma_1 dW_{1,t} \\
dY_{2,t} &= -\kappa_2 Y_{2,t} dt + \sigma_2 dW_{2,t}
\end{aligned} d r t d Y 1 , t d Y 2 , t = ( ϕ 1 Y 1 , t + ϕ 2 Y 2 , t ) d t = − κ 1 Y 1 , t d t + σ 1 d W 1 , t = − κ 2 Y 2 , t d t + σ 2 d W 2 , t
Features :
Two factors capture different aspects of yield curve dynamics
Factor loadings ϕ 1 , ϕ 2 \phi_1, \phi_2 ϕ 1 , ϕ 2 determine level and slope sensitivity
Correlation between factors via d W 1 d W 2 = ρ d t dW_1 dW_2 = \rho dt d W 1 d W 2 = ρ d t
Principal Component Models
Based on empirical analysis of yield curve movements:
Level : Parallel shifts (85-90% of variance)
Slope : Steepening/flattening (8-12% of variance)
Curvature : Butterfly movements (1-3% of variance)
Forward Rate Models
Heath-Jarrow-Morton (HJM) Framework
Instead of modeling the short rate, HJM models the evolution of the entire forward curve .
SDE for Forward Rates :
d f ( t , T ) = α ( t , T ) d t + σ ( t , T ) d W t df(t,T) = \alpha(t,T)dt + \sigma(t,T)dW_t df ( t , T ) = α ( t , T ) d t + σ ( t , T ) d W t
No-Arbitrage Condition :
α ( t , T ) = σ ( t , T ) ∫ t T σ ( t , u ) d u \alpha(t,T) = \sigma(t,T) \int_t^T \sigma(t,u) du α ( t , T ) = σ ( t , T ) ∫ t T σ ( t , u ) d u
This drift restriction ensures consistency with bond pricing relationships.
LIBOR Market Model (BGM)
Models forward LIBOR rates directly:
d L ( t , T ) = μ ( t , T ) d t + σ ( t , T ) L ( t , T ) d W t dL(t,T) = \mu(t,T)dt + \sigma(t,T)L(t,T)dW_t d L ( t , T ) = μ ( t , T ) d t + σ ( t , T ) L ( t , T ) d W t
Features :
Log-normal forward rates (always positive)
Market observables (LIBOR rates)
Swaption pricing via Black's formula
Drift Condition : Under the forward measure:
μ ( t , T ) = σ ( t , T ) ∑ j : T j > T τ j σ ( t , T j ) L ( t , T j ) ρ T , T j 1 + τ j L ( t , T j ) \mu(t,T) = \sigma(t,T) \sum_{j: T_j > T} \frac{\tau_j \sigma(t,T_j) L(t,T_j) \rho_{T,T_j}}{1 + \tau_j L(t,T_j)} μ ( t , T ) = σ ( t , T ) j : T j > T ∑ 1 + τ j L ( t , T j ) τ j σ ( t , T j ) L ( t , T j ) ρ T , T j
Affine Term Structure Models
General Framework
Affine models assume bond prices have the form:
P ( t , T ) = e A ( t , T ) + B ( t , T ) ⋅ X t P(t,T) = e^{A(t,T) + B(t,T) \cdot X_t} P ( t , T ) = e A ( t , T ) + B ( t , T ) ⋅ X t
where X t X_t X t is a vector of state variables.
State Variable Dynamics :
d X t = κ ( θ − X t ) d t + Σ S t d W t dX_t = \kappa(\theta - X_t)dt + \Sigma\sqrt{S_t} dW_t d X t = κ ( θ − X t ) d t + Σ S t d W t
where S t S_t S t is diagonal with S i i , t = α i + β i ⋅ X t S_{ii,t} = \alpha_i + \beta_i \cdot X_t S ii , t = α i + β i ⋅ X t .
Examples
Vasicek : One-factor affine (X t = r t X_t = r_t X t = r t )
CIR : One-factor affine with square-root diffusion
Multi-factor CIR : Vector generalization
Advantages
Analytical tractability : Bond prices in closed form
Flexible correlation structure : Rich factor interactions
Calibration efficiency : Linear optimization problems
Negative Interest Rates
Shifted Models
To handle negative rates, modify traditional models:
Shifted Vasicek :
d r t = κ ( θ − ( r t + λ ) ) d t + σ d W t dr_t = \kappa(\theta - (r_t + \lambda))dt + \sigma dW_t d r t = κ ( θ − ( r t + λ )) d t + σ d W t
Shifted CIR :
d ( r t + λ ) = κ ( θ − ( r t + λ ) ) d t + σ r t + λ d W t d(r_t + \lambda) = \kappa(\theta - (r_t + \lambda))dt + \sigma\sqrt{r_t + \lambda} dW_t d ( r t + λ ) = κ ( θ − ( r t + λ )) d t + σ r t + λ d W t
where λ > 0 \lambda > 0 λ > 0 is a shift parameter.
Gaussian Models
Return to Gaussian models (Vasicek, Hull-White) that naturally allow negative rates.
Term Structure Fitting
Nelson-Siegel Model
Parametric representation of the yield curve:
y ( t , T ) = β 0 + β 1 1 − e − λ ( T − t ) λ ( T − t ) + β 2 ( 1 − e − λ ( T − t ) λ ( T − t ) − e − λ ( T − t ) ) y(t,T) = \beta_0 + \beta_1 \frac{1-e^{-\lambda(T-t)}}{\lambda(T-t)} + \beta_2 \left(\frac{1-e^{-\lambda(T-t)}}{\lambda(T-t)} - e^{-\lambda(T-t)}\right) y ( t , T ) = β 0 + β 1 λ ( T − t ) 1 − e − λ ( T − t ) + β 2 ( λ ( T − t ) 1 − e − λ ( T − t ) − e − λ ( T − t ) )
Interpretation :
β 0 \beta_0 β 0 : Long-term level
β 1 \beta_1 β 1 : Slope factor
β 2 \beta_2 β 2 : Curvature factor
λ \lambda λ : Decay parameter
Svensson Extension
Adds a fourth parameter for additional flexibility:
y ( t , T ) = Nelson-Siegel + β 3 ( 1 − e − λ 2 ( T − t ) λ 2 ( T − t ) − e − λ 2 ( T − t ) ) y(t,T) = \text{Nelson-Siegel} + \beta_3 \left(\frac{1-e^{-\lambda_2(T-t)}}{\lambda_2(T-t)} - e^{-\lambda_2(T-t)}\right) y ( t , T ) = Nelson-Siegel + β 3 ( λ 2 ( T − t ) 1 − e − λ 2 ( T − t ) − e − λ 2 ( T − t ) )
Calibration Methods
Market Observable Instruments
Government bonds : Risk-free curve construction
Interest rate swaps : Liquid medium to long-term rates
Interest rate futures : Short to medium-term rates
Caps/floors : Volatility information
Swaptions : Correlation and volatility structure
Objective Functions
Least squares on prices :
min θ ∑ i = 1 N w i ( P i market − P i model ( θ ) ) 2 \min_{\theta} \sum_{i=1}^N w_i (P_i^{\text{market}} - P_i^{\text{model}}(\theta))^2 θ min i = 1 ∑ N w i ( P i market − P i model ( θ ) ) 2
Least squares on yields :
min θ ∑ i = 1 N w i ( y i market − y i model ( θ ) ) 2 \min_{\theta} \sum_{i=1}^N w_i (y_i^{\text{market}} - y_i^{\text{model}}(\theta))^2 θ min i = 1 ∑ N w i ( y i market − y i model ( θ ) ) 2
Regularization
Smoothness penalties : Prevent over-fitting
Parameter bounds : Economic constraints
Stability requirements : Ensure mean reversion
Interest Rate Derivatives
Caps and Floors
Cap payoff :
∑ i = 1 n τ i max ( L ( T i , T i + 1 ) − K , 0 ) \sum_{i=1}^n \tau_i \max(L(T_i, T_{i+1}) - K, 0) ∑ i = 1 n τ i max ( L ( T i , T i + 1 ) − K , 0 )
Floor payoff :
∑ i = 1 n τ i max ( K − L ( T i , T i + 1 ) , 0 ) \sum_{i=1}^n \tau_i \max(K - L(T_i, T_{i+1}), 0) ∑ i = 1 n τ i max ( K − L ( T i , T i + 1 ) , 0 )
Swaptions
Payer swaption : Option to enter into a swap as fixed-rate payer
Receiver swaption : Option to enter into a swap as fixed-rate receiver
Black's Formula : Standard market formula assumes log-normal forward swap rates.
Bond Options
Options on government bonds, corporate bonds, or bond futures.
Risk Management Applications
Duration and Convexity
Modified Duration :
D = − 1 P ∂ P ∂ y D = -\frac{1}{P}\frac{\partial P}{\partial y} D = − P 1 ∂ y ∂ P
Convexity :
C = 1 P ∂ 2 P ∂ y 2 C = \frac{1}{P}\frac{\partial^2 P}{\partial y^2} C = P 1 ∂ y 2 ∂ 2 P
Price Approximation :
Δ P ≈ − D ⋅ P ⋅ Δ y + 1 2 C ⋅ P ⋅ ( Δ y ) 2 \Delta P \approx -D \cdot P \cdot \Delta y + \frac{1}{2}C \cdot P \cdot (\Delta y)^2 Δ P ≈ − D ⋅ P ⋅ Δ y + 2 1 C ⋅ P ⋅ ( Δ y ) 2
Key Rate Durations
Sensitivity to specific points on the yield curve:
K R D i = − 1 P ∂ P ∂ y i KRD_i = -\frac{1}{P}\frac{\partial P}{\partial y_i} K R D i = − P 1 ∂ y i ∂ P
Value at Risk (VaR)
Interest rate VaR using:
Historical simulation : Historical rate movements
Monte Carlo : Simulate rate paths from calibrated models
Analytical methods : Assume normal distribution
Numerical Methods
Tree Methods
Binomial/trinomial trees for short rate models:
Discretize time and state space
Ensure no-arbitrage conditions
Backward induction for option pricing
Monte Carlo Simulation
Simulate rate paths under risk-neutral measure
Compute payoffs along each path
Discount and average to get present value
Finite Difference Methods
Solve the bond pricing PDE :
∂ P ∂ t + 1 2 σ 2 ( r ) ∂ 2 P ∂ r 2 + μ Q ( r ) ∂ P ∂ r − r P = 0 \frac{\partial P}{\partial t} + \frac{1}{2}\sigma^2(r)\frac{\partial^2 P}{\partial r^2} + \mu^{\mathbb{Q}}(r)\frac{\partial P}{\partial r} - rP = 0 ∂ t ∂ P + 2 1 σ 2 ( r ) ∂ r 2 ∂ 2 P + μ Q ( r ) ∂ r ∂ P − r P = 0
Credit Risk and Interest Rates
Credit Spread Models
Structural models : Link default to firm value
Reduced-form models : Model hazard rates directly
Credit-Interest Rate Correlation
Default probability often increases when rates rise (corporate stress) or fall (economic weakness).
Inflation Modeling
Real vs Nominal Rates
Fisher equation :
1 + r nominal = ( 1 + r real ) ( 1 + π ) 1 + r_{\text{nominal}} = (1 + r_{\text{real}})(1 + \pi) 1 + r nominal = ( 1 + r real ) ( 1 + π )
Inflation-Protected Securities
Treasury Inflation-Protected Securities (TIPS) : Principal adjusts with CPI
Inflation swaps : Exchange fixed for realized inflation
Connection to Other Topics
Interest rate models integrate many quantitative concepts:
Built on Stochastic Differential Equations
Use Martingale pricing for no-arbitrage
Apply Itô's Lemma for bond pricing PDEs
Connect to Brownian Motion and mean-reverting processes
Extend to Jump-Diffusion for crisis modeling
Foundation for fixed-income option pricing
Critical for risk management and ALM