Affine Process¶
A Markov process whose conditional transition characteristic function is exponential-affine in its initial state.
Core Idea¶
An affine process is a Markov process whose conditional transition characteristic function has an exponential-affine dependence on its initial state: for appropriate \(t,u\) and every initial state \(x\) in the declared state space, \(\mathbb E_x[e^{i\langle u,X_t\rangle}]=e^{\phi(t,u)+\langle\psi(t,u),x\rangle}\) in the conservative case. If killing is allowed, the expectation is restricted to survival before \(t\). This transform identity is the defining criterion, not an assumption that paths or every model coefficient are linear.[^ref-16e681e4326a]
Scope of Application¶
Gaussian Ornstein–Uhlenbeck dynamics on real states and nonnegative branching-type or CIR square-root diffusions are different cases sharing the transform form. The frozen Wikipedia topic, Basic affine jump diffusion, is one narrower subtype. Jumps, a state-independent jump-size law, time-integral transforms and bond or option pricing are not necessary to be an affine process.[ref-16e681e4326a][ref-e4fc96883dd5]
Clarity¶
The test ranges over initial states, horizons and transform arguments, not one observed path or fitted state. Duffie, Filipović and Schachermayer characterize generator coefficients and generalized Riccati equations for regular affine processes under separate assumptions; those theorems should not be made constitutive of every affine process. The state space and admissible parameters must be stated.[^ref-16e681e4326a]
Manages Complexity¶
The conditional law's dependence on initial state is reduced to functions \(\phi\) and \(\psi\). Under regularity, Riccati equations can compute them. A modeler can test the transform criterion first, then the regularity needed for a computation method, and only later any added pricing, intensity or payoff assumptions.[ref-16e681e4326a][ref-e4fc96883dd5]
Abstract Reasoning¶
For an OU solution \(X_t=e^{-\kappa t}x+Z_t\) with innovation \(Z_t\) independent of \(x\), the characteristic transform factors as \(\mathbb E[e^{iuZ_t}]e^{iu e^{-\kappa t}x}\). A square-root/CBI process has different dynamics and state constraints but the same abstract exponential-affine initial-state form. A general Markov process lacking that form is outside this subclass.[^ref-16e681e4326a]
Knowledge Transfer¶
Ask of a proposed model: What is its Markov state space, and does its conditional characteristic transform have the specified exponential-affine form throughout that space? This recognizes a shared class across continuous and jump models without confusing its mathematical identity with financial applications. The proposed DAG parent is live Markov Process; staged Ornstein–Uhlenbeck Process is a narrower example.[ref-16e681e4326a][ref-e4fc96883dd5]
[^ref-16e681e4326a]: Darrell Duffie, Damir Filipović and Walter Schachermayer, “Affine Processes and Applications in Finance,” Annals of Applied Probability 13 (2003), 984–1053, author-hosted original full text directly checked at Definition 2.1/Eq. (2.2), Definitions 2.4–2.5, Theorem 2.7, Corollary 2.10 and §12. [^ref-e4fc96883dd5]: Darrell Duffie, Jun Pan and Kenneth Singleton, “Transform Analysis and Asset Pricing for Affine Jump-Diffusions,” author-hosted 1999 draft of Econometrica 68 (2000), abstract and introduction PDF pp. 1–2 directly checked.
Relationships to Other Abstractions¶
Current abstraction Affine Process Domain-specific
Parents (1) — more general patterns this builds on
-
Affine Process is a kind of Markov Process Prime
Every affine process is a Markov process with the additional exponential-affine transition-transform condition.
Hierarchy paths (4) — routes to 4 parentless roots
- Affine Process → Markov Process → Stochastic Process
- Affine Process → Markov Process → State and State Transition → Phase Space
- Affine Process → Markov Process → Probability → Measure → Set and Membership
- Affine Process → Markov Process → Probability → Measure → Aggregation → Micro Macro Linkage
Neighborhood in Abstraction Space¶
Affine Process sits in a sparse region of the domain-specific corpus (77th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Unclustered & Miscellaneous (2551 abstractions)
Nearest neighbors
- Kolmogorov Equations for Continuous-Time Markov Chains — 0.85
- Control-Theoretic Orbit — 0.85
- Metropolis Algorithm — 0.82
- Differential equation — 0.82
- Differential Inclusion — 0.82
Computed from structural-signature embeddings · 2026-10-08