Ultra-High Precision Generation of Financial Time Series Using Image Generation AI

Abstract

We propose a generative AI that applies the technique of diffusion models for image generation to replicate real financial time series data with extremely high precision. Financial time series data possess long-term memory processes, making it challenging to generate high-precision financial time series without setting the maximum sequence length in transformers to be very long. Therefore, by transforming time series data into frequency information through wavelet transformation and learning the long-term memory as images, we use a more lightweight diffusion model to generate real financial time series data with ultra-high precision.