CIR Affine Model: Structure & Applications

• CIR Affine Model is a continuous-time Markov process with affine drift and diffusion, enabling closed-form transition densities and exponential-affine pricing.
• Its extensions, including Lévy-driven, multifactor, and α-stable variants, capture jumps, heavy tails, and stochastic discontinuities in financial markets.
• The model’s structure facilitates efficient Riccati ODE solutions for bond pricing, risk management, and various applications in quantitative finance.

The CIR affine model comprises a class of continuous-time Markov processes and their Lévy-driven extensions, canonically used for modeling short rates, stochastic volatility, and other mean-reverting processes in quantitative finance. The prototypical Cox–Ingersoll–Ross (CIR) process is an affine diffusion—so named because its infinitesimal generator and conditional Laplace transform are affine in state—admitting closed-form transition and pricing results. Recent developments extend the CIR framework to multifactor, jump, α-stable, and stochastic discontinuity regimes while preserving core affine tractability and exponential-affine solution structure.

1. Formal Definition and Affine Structure

The classical one-factor CIR diffusion is governed by

$$dr_t = \kappa(\theta - r_t)\,dt + \sigma \sqrt{r_t}\,dW_t, \quad r_0 \ge 0,\,\kappa,\theta,\sigma > 0,$$

where $W_t$ is standard Brownian motion. Affine property follows as both drift and diffusion coefficients are affine functions of $r_t$, and the infinitesimal generator acts by

$$\mathcal{L} f(x) = \kappa(\theta-x)\,f'(x) + \frac{1}{2}\sigma^2 x\,f''(x).$$

The crucial Feller condition $2\kappa\theta \ge \sigma^2$ ensures non-negativity and avoids boundary hitting at zero (Kohatsu-Higa et al., 2 Oct 2025).

This structure generalizes to $d$-factor models and Lévy-driven extensions: $$dR(t) = F(R(t))\,dt + \sum_{i=1}^d G_i(R(t-))\,dZ_i(t),$$ where $F(x) = a x + b$, $G_i(x) = d_i x^{1/\alpha_i}$, and the driving $Z_i$ are independent Lévy martingales, typically α-stable with $W_t$0 (Barski et al., 2023, Barski et al., 2022, Barski et al., 2024).

Affineness in the term-structure context requires: (1) $W_t$1 affine, (2) diffusion and jump coefficients of a specific form, and (3) proportionality in the jump measures with state, so that the generator is affine in $W_t$2 and pricing ODEs remain solvable in Riccati-type form (2312.3661, Barski et al., 2019).

2. Classical and Extended Dynamics

2.1. One-Factor CIR Diffusion

The transition density of $W_t$3 is a scaled noncentral chi-squared, and bond prices admit closed-form exponential-affine solutions. The moments, stationary law, and ergodicity are explicit: as $W_t$4,

$W_t$5

if the Feller condition holds (Kohatsu-Higa et al., 2 Oct 2025, Ning et al., 15 Dec 2025).

2.2. Lévy and α-Stable Generalizations

Replacing the Brownian driver with a spectrally positive $W_t$6-stable Lévy process (so $W_t$7 is discontinuous yet self-similar and heavy-tailed) yields the so-called $W_t$8-CIR or “affine-root” process: $W_t$9 The generator becomes: $r_t$0 where $r_t$1 is the $r_t$2-stable Lévy measure (Jiao et al., 2016, Barski et al., 2023, Barski et al., 2019). The affine structure persists, with conditional Laplace transforms and bond prices again given by exponential-affine formulas involving generalized Riccati ODEs (Barski et al., 2024).

2.3. Multifactor and Dependent-Lévy Extensions

Affine CIR models extend to multivariate and multifactor (independent or dependent) Lévy drivers. For $r_t$3-dimensional noise with a spherical Lévy measure, under mild comparability, the process reduces to a canonical one-factor α-stable CIR model for pricing and limiting behavior (Barski et al., 2024, Barski et al., 2022).

3. Solvability: Exponential-Affine Pricing and Riccati ODEs

For all affine specifications (classical, jump, stable), the conditional expectation or Laplace transform

$r_t$4

satisfies a Riccati ODE system of the form

$r_t$5

where the sum captures multifactor stable components. The explicit solution exists for pure CIR ($r_t$6), while for general $r_t$7 or multiple indices, the system is numerically tractable (2312.3661, Barski et al., 2023, Jiao et al., 2016).

Forward rates and associated affine expectations retain exponential-affine form; bond options and related derivatives are computed using Laplace transforms or explicit density representations, with noncentral chi-squared and Kummer confluent hypergeometric functions in the two-factor case (Yilmaz et al., 31 Oct 2025).

4. Exact Distributions, Transition Densities, and Numerical Schemes

4.1. One-Factor Densities

For the classical CIR, transition densities are noncentral chi-squared, enabling likelihood-based calibration and explicit derivative pricing (Kohatsu-Higa et al., 2 Oct 2025).

4.2. Sums and Linear Combinations

For sums of two independent CIR processes (useful in multifactor short-rate or multi-Heston volatility models), the law is derived as a double Poisson-Gamma mixture, resulting in a kernel involving sums of confluent hypergeometric ($r_t$8) functions, enabling numerically stable evaluation of both the PDF and CDF (Yilmaz et al., 31 Oct 2025).

4.3. Numerical Implementation

Computations exploit analytic series truncation (for Poisson-gamma and Kummer series), vectorized calculation of weights and kernel values, and direct Laplace inversion if needed. These approaches provide $r_t$9 computational cost with high accuracy for parameter estimation and risk management (Yilmaz et al., 31 Oct 2025).

5. Major Applications in Mathematical Finance

• Short-rate modeling and bond pricing: CIR and multifactor affine extensions drive the Heath-Jarrow-Morton (HJM) term-structure models, with bond and option prices explicit modulo Riccati ODEs (2312.3661, Barski et al., 2023).
• Stochastic volatility: Affine CIR-type factors underlie the variance process in Heston and multi-Heston models, supporting closed-form or semi-analytical option pricing (Jin et al., 2016, Yilmaz et al., 31 Oct 2025).
• Risk management and credit analytics: Credit intensity processes modeled as affine CIR/α-CIR support explicit formulas for survival probabilities, credit valuation adjustments, and portfolio loss tails (Yilmaz et al., 31 Oct 2025).
• Insurance, reliability, biophysical modeling: Aggregated mean-reverting processes in risk and reliability theory admit analogous multifactor CIR representations (Yilmaz et al., 31 Oct 2025).

6. Theoretical Extensions and Nonstandard Regimes

• Parameter uncertainty (“nonlinear affine models”): Under parameter ambiguity, CIR-like dynamics are replaced by variational (supremum/infimum) Kolmogorov and Riccati equations, yielding bounds on bond prices and a robust framework for pricing under Knightian uncertainty (Fadina et al., 2018).
• Stochastic discontinuities: Affine CIR processes with state-dependent jumps at deterministic times admit a Riccati-ODE plus jump iteration solution; necessary and sufficient conditions for the affine property and infinite divisibility are established (Fontana et al., 19 Sep 2025).
• Delay and memory: Inclusion of fixed delay in the drift yields affine PDEs in an enlarged state space, with bond prices exponential-affine in the current state and the delayed history process (Flore et al., 2018).
• Data-driven/empirical:
  • Piecewise-calibrated/translated models: Extensions such as the CIR♯ framework fit time-varying volatility and negative rates by translation and segmentwise ARIMA calibration, preserving exponential-affine solvability (Orlando et al., 2018).
  • Integration-by-parts (Greeks): Explicit IBP formulas for sensitivities (e.g., Delta) exploit affine structure to avoid pathwise differentiation of the square-root diffusion (Kohatsu-Higa et al., 2 Oct 2025).

7. Calibration Results and Empirical Fit

Empirical studies, especially with ECB and Libor/swap data, show classical CIR fits yield material errors (1%–25%) in modern, low/noise/negative rate regimes. α-CIR and multifactor stable-CIR models typically achieve fitting errors reduced by up to 97%, with marginal improvements beyond two factors at the cost of parameter proliferation (Barski et al., 2023, Barski et al., 2024). α-stable (jump) extensions notably improve calibration in environments with heavy tails or observed “spikes” (Jiao et al., 2016, Barski et al., 2023).

Model	Data Type	Typical Fitting Error	Notes
CIR ($$\mathcal{L} f(x) = \kappa(\theta-x)\,f'(x) + \frac{1}{2}\sigma^2 x\,f''(x).$$0)	AAA EUR spot, Libor/swap	1–25%	Poor in low/negative, jump regimes
α-CIR, GCIR(2)	Same	0.4–1%	Drastic improvement, heavy tails
Higher factor GCIR	Same	~0.44%	Marginal additional improvement

These results highlight the practical necessity of flexibility in CIR affine modeling beyond the classical Gaussian paradigm for modern financial markets (Barski et al., 2023).

---

References: (Yilmaz et al., 31 Oct 2025, Kohatsu-Higa et al., 2 Oct 2025, Barski et al., 2023, Barski et al., 2022, Barski et al., 2024, Barski et al., 2019, Liu et al., 2013, Jiao et al., 2016, Jin et al., 2016, Orlando et al., 2018, Fontana et al., 19 Sep 2025, Ning et al., 15 Dec 2025, Fadina et al., 2018, Flore et al., 2018).