Abstract:We propose a machine learning method to solve a mean-field game price formation model with common noise. This involves determining the price of a commodity traded among rational agents subject to a market clearing condition imposed by random supply, which presents additional challenges compared to the deterministic counterpart. Our approach uses a dual recurrent neural network architecture encoding noise dependence and a particle approximation of the mean-field model with a single loss function optimized by adversarial training. We provide a posteriori estimates for convergence and illustrate our method through numerical experiments.

What problem does this paper attempt to address?

The paper attempts to address the problem of how to use machine learning methods to solve the price formation model in Mean Field Games (MFGs) in the presence of common noise. Specifically, the paper focuses on determining the price \( \varpi \) of goods traded among multiple rational agents under random supply \( Q(t) \) within a finite time horizon \( T > 0 \). Here, the supply function \( Q(t) \) satisfies a Stochastic Differential Equation (SDE) with common noise: \[ \begin{cases} dQ(t) = b_S(Q(t), t) dt + \sigma_S(Q(t), t) dW(t), \\ Q(0) = q_0, \end{cases} \] where \( q_0 \in \mathbb{R} \) is the initial supply, and \( W(t) \) is a one-dimensional Brownian motion representing the common noise. The coefficients \( b_S \) and \( \sigma_S \) satisfy the usual Lipschitz conditions to ensure the existence and uniqueness of the solution. The paper proposes a dual-loop Recurrent Neural Network (RNN) architecture to encode noise dependency and transforms the mean field model into a single loss function for optimization through a particle approximation method. Additionally, the paper provides posterior convergence estimates and validates the effectiveness of the method through numerical experiments. In summary, the main contributions of the paper are: 1. **Introduction of machine learning methods**: Utilizing an RNN architecture to solve the price formation model with common noise. 2. **Handling infinite-dimensional state space**: Addressing the issue of the state space becoming infinite-dimensional due to common noise through a particle approximation method. 3. **Providing posterior convergence estimates**: Ensuring the method can provide accurate results even in the absence of a benchmark solution. 4. **Numerical experiment validation**: Demonstrating the effectiveness and accuracy of the method through numerical experiments under a linear-quadratic framework.

Machine Learning architectures for price formation models with common noise

A mean-field game price model with noise

The Potential Method For Price-Formation Models

Exploration Noise for Learning Linear-Quadratic Mean Field Games

A random-supply Mean Field Game price model

A Mean-Field Game Approach to Price Formation

Deep Learning for Mean Field Games and Mean Field Control with Applications to Finance

Common Noise by Random Measures: Mean-Field Equilibria for Competitive Investment and Hedging

Stochastic Delay Differential Games: Financial Modeling and Machine Learning Algorithms

A Deep Learning Approach to Data-Driven Model-Free Pricing and to Martingale Optimal Transport

Machine learning for option pricing: an empirical investigation of network architectures

Bayesian Learning in Mean Field Games

A Machine Learning Method for Stackelberg Mean Field Games

Deep Learning for Population-Dependent Controls in Mean Field Control Problems with Common Noise

Reinforcement Learning for Intra-and-Inter-Bank Borrowing and Lending Mean Field Control Game

A machine learning framework for solving high-dimensional mean field game and mean field control problems

Multiagent Model and Mean Field Theory of Complex Auction Dynamics

Mean field games with absorption and common noise with a model of bank run

Machine Learning Methods for Pricing Financial Derivatives

Reinforcement Learning for Mean Field Game