Abstract
This paper presents a novel framework for computing the characteristic functions of complex stochastic processes, specifically multi-dimensional affine jump-diffusions. By conceptualizing these transforms as "analytic vectors" of the process's generator, the authors derive convergent functional series representations. This approach effectively bypasses the traditional requirement to solve complex systems of generalized Riccati differential equations, offering a robust alternative for high-dimensional modeling in finance and physics.
Introduction
In the fields of mathematical finance and theoretical physics, the Fourier transform (or characteristic function) is an essential tool for pricing derivatives and analyzing system dynamics. For the class of affine processes, determining these transforms typically involves solving a system of generalized Riccati differential equations. While these equations are well-characterized, finding closed-form solutions in high-dimensional settings is often computationally prohibitive or impossible.
Furthermore, standard numerical fallbacks, such as "small-time" Taylor expansions, frequently fail because they possess a zero radius of convergence at the starting time ($t=0$). This research addresses these challenges by developing a method to represent transforms as convergent series that remain valid across the entire time horizon, transforming a differential problem into an algebraic one.
Theoretical Framework and Methodology
The authors bridge stochastic calculus and functional analysis by utilizing the concept of analytic vectors. The methodology is built upon several key pillars:
- Pseudo-Generators: The generator of an Itô-Lévy process—the operator describing the infinitesimal movement of the process—is defined as an infinite-order differential operator.
- Analytic Vectors: The researchers identify a set of functions, such as the Fourier basis, that serve as analytic vectors for this generator. When a function is an analytic vector, its transformation under the process can be described via a power series.
- Holomorphic Extension: A primary innovation of this work is the proof that these transforms can be extended holomorphically (smoothly in the complex plane) into a "strip" that encompasses the positive real time axis. This ensures the series remains valid for long-term horizons rather than just infinitesimal increments.
- Stirling Number Expansions: To ensure convergence over a wide domain, the authors replace standard Taylor series with a specific expansion involving unsigned Stirling numbers of the first kind.
- Recursive Logic: For affine processes, the transform is known to be "log-affine," meaning the exponent is a linear function of the state. The paper provides recursive algebraic formulas to calculate the coefficients of this exponent, eliminating the need for Riccati integration.
Key Insights
- Operator-Centric View: The fundamental shift in this paper is viewing the Fourier transform not merely as an integral, but as the result of an operator acting on an analytic vector. This transition from calculus to operator theory provides new tools for approximation.
- Bypassing Riccati Equations: The recursive series allow for the approximation of characteristic functions in high-dimensional affine processes to any desired degree of accuracy without solving differential equations.
- Generality: While the focus is on affine processes (such as the Heston or Bates models used in finance), the framework is designed for general Itô-Lévy processes, suggesting broad applicability in stochastic modeling.
Principal Results
- Theorem 3.3: This theorem establishes the formal equivalence between a transform possessing a holomorphic extension and its ability to be represented by a convergent series based on Stirling-like coefficients.
- Theorems 4.1 & 5.4: These provide explicit bounds and estimates for the radius of convergence for affine generators, proving that the Fourier basis effectively forms a set of analytic vectors for these processes.
- Algebraic Recursion: The paper details explicit recursive procedures (Section 6) to compute the terms of the log-affine expansion for both one-dimensional and multi-dimensional cases, providing a clear path for numerical implementation.
Limitations and Future Directions
Despite the power of the framework, the authors note that the set of generators for which the Fourier basis works perfectly as analytic vectors is relatively "thin" (Proposition 4.8). This implies that while the method is highly effective for affine processes, adapting it for highly non-linear or non-affine processes may require further theoretical refinement.
Additionally, some global convergence results depend on the existence of a limiting stationary distribution. If a process is prone to "explosion" or lacks a steady state, the series convergence may be limited to a specific time window. Future research may focus on optimizing these recursive formulas to reduce the computational demand when high accuracy is required in very high-dimensional systems.
Summary updated