[4L3-GS-10] AI application: General

[4L3-GS-10-01] Generation of Financial Time Series with Diffusion Models

〇Tomonori Takahashi¹, Takayuki Mizuno^2,1 (1. The Graduate University for Advanced Studies, SOKENDAI, 2. National Institute of Informatics)

Keywords:finance, time series, generative models

Despite its practical significance, generating realistic synthetic financial time series is a challenging task because of their statistical properties known as 'stylized facts' such as fat tails, volatility clustering, and autocorrelation. Various generative models including generative adversarial networks (GANs) and variational autoencoders (VAEs) have been employed to address this challenge, but no model satisfies all of stylized facts yet. As an alternative approach, we propose the utilization of diffusion models, specifically, denoising diffusion probabilistic models (DDPMs), for synthetic financial time series. This approach allows for the simultaneous generation of multiple time series such as stock prices, volumes, and spreads. We demonstrate the proposed approach can reproduce the statistical properties observed in financial markets.