Abstract:This paper introduces FourNet, a novel single-layer feed-forward neural network (FFNN) method designed to approximate transition densities for which closed-form expressions of their Fourier transforms, i.e. characteristic functions, are available. A unique feature of FourNet lies in its use of a Gaussian activation function, enabling exact Fourier and inverse Fourier transformations and drawing analogies with the Gaussian mixture model. We mathematically establish FourNet's capacity to approximate transition densities in the $L_2$-sense arbitrarily well with finite number of neurons. The parameters of FourNet are learned by minimizing a loss function derived from the known characteristic function and the Fourier transform of the FFNN, complemented by a strategic sampling approach to enhance training. We derive practical bounds for the $L_2$ estimation error and the potential pointwise loss of nonnegativity in FourNet for $d$-dimensions ($d\ge 1$), highlighting its robustness and applicability in practical settings. FourNet's accuracy and versatility are demonstrated through a wide range of dynamics common in quantitative finance, including Lévy processes and the Heston stochastic volatility models-including those augmented with the self-exciting Queue-Hawkes jump process.

What problem does this paper attempt to address?

This paper attempts to solve the problem of how to effectively and accurately estimate transition density functions without closed - form expressions in the financial field. Specifically: 1. **Problem Background**: - In quantitative finance, the transition density functions of many stochastic models are unknown, but their Fourier transforms (i.e., characteristic functions) usually have explicit expressions. These transition density functions are crucial for describing the dynamics of stochastic processes. - Traditional numerical methods such as kernel density estimation can be used to estimate these density functions, but they perform poorly when dealing with complex and non - smooth characteristics and lack a control mechanism for non - negativity. 2. **Limitations of Existing Methods**: - Existing neural network methods are mainly used to solve high - dimensional Kolmogorov partial differential equations (PDEs), but most of these methods are black - box in nature and require continuous reformulation of Kolmogorov PDEs for different stochastic models. - These methods may be too complex in practical applications and have limited analysis of estimation errors and non - negativity losses. 3. **Goals of the Paper**: - Propose a method based on a single - layer feed - forward neural network (FFNN) - FourNet, for approximating transition density functions with known Fourier transforms. - Four main goals include: 1. **Simplify Implementation**: By training in the Fourier domain, the implementation is simplified compared to deep neural network structures. 2. **Error Analysis**: Rigorously analyze the FourNet estimation error in the L2 sense and prove that it remains invariant under Fourier transformation. 3. **Verify Accuracy**: Verify the accuracy and generality of FourNet in various stochastic financial models through extensive numerical experiments. 4. **Control Non - negativity**: Provide practical L2 estimation error and potential non - negativity loss bounds to ensure its reliability and robustness in practical applications. 4. **Innovations**: - Use the Gaussian activation function, enabling FourNet to perform accurate Fourier and inverse Fourier transforms. - Utilize the known characteristic function and the Fourier transform of the FFNN to construct the loss function and enhance the training effect through strategic sampling methods. - Through strict mathematical derivations, prove that FourNet can approximate the transition density function arbitrarily well in the L2 sense and that the error remains invariant under Fourier transformation. 5. **Application Prospects**: - FourNet shows excellent robustness in handling complex scenarios such as ultra - short maturity times and asymmetric heavy - tailed distributions. - Future work will be extended to more complex stochastic control problems, such as portfolio optimization, further broadening its application range. In summary, this paper aims to address the challenges of transition density function estimation in the financial field by introducing the FourNet method, providing a new method that is efficient, accurate, and easy to implement.

Fourier Neural Network Approximation of Transition Densities in Finance

Fourier Neural Networks for Function Approximation

Approximating Numerical Fluxes Using Fourier Neural Operators for Hyperbolic Conservation Laws

Neural Fourier Modelling: A Highly Compact Approach to Time-Series Analysis

Fourier neural networks as function approximators and differential equation solvers

Deep learning of transition probability densities for stochastic asset models with applications in option pricing

Deep Learning without Global Optimization by Random Fourier Neural Networks

Fourier warm start for physics-informed neural networks

Simo Fourier Neural Networks Research

Physics-embedded Fourier Neural Network for Partial Differential Equations

Neural network stochastic differential equation models with applications to financial data forecasting

Fourier Head: Helping Large Language Models Learn Complex Probability Distributions

Discretization Error of Fourier Neural Operators

Fourier-Net: Fast Image Registration with Band-limited Deformation

Robust Fourier Neural Networks

Neural network representation of the probability density function of diffusion processes

Finite Neural Networks as Mixtures of Gaussian Processes: From Provable Error Bounds to Prior Selection

Understanding Training and Generalization in Deep Learning by Fourier Analysis.

Physics-informed Neural Networks with Fourier Features for Seismic Wavefield Simulation in Time-Domain Nonsmooth Complex Media

Fourier Basis Density Model

Enforcing asymptotic behavior with DNNs for approximation and regression in finance