Abstract
In this paper, we present CoMeTS-GAN (Correlated Multivariate Time Series GAN), a framework based on Conditional Generative Adversarial Networks (C-GANs), designed to generate mid-prices and volumes time series of correlated stocks. This tool provides a light and responsive solution for realistic stock market simulation. It is able to accurately learn and reproduce inter-asset correlations, a crucial aspect for achieving realness in multi-stock simulation environments. Our experimental campaign assesses the model using acknowledged stylized facts of stock markets as well as additional metrics capturing inter-asset correlations. We compare our model to leading architectures, highlighting our approach’s strengths. These findings suggest the potential of CoMeTS-GAN in realistically simulating correlated price movements, offering a responsive market environment and valuable input for trading strategy formulation.
Funding Information
  • JPMorgan Chase AI Research Faculty Award (Understanding interdependent market dynamics: vulnerabilities and opportunities)

This publication has 11 references indexed in Scilit: