Simulating financial time series using attention

TL;DR

This paper introduces attention-based GANs, including convolutional and transformer models, to simulate financial time series data, capturing complex dependencies and stylized facts for improved data generation.

Contribution

It presents novel attention-enhanced GAN architectures specifically designed for financial data simulation, improving the replication of statistical properties and long-range dependencies.

Findings

Attention-based GANs reproduce stylized facts effectively.

They smooth the autocorrelation of returns.

Outperform pure convolutional GANs in key metrics.

Abstract

Financial time series simulation is a central topic since it extends the limited real data for training and evaluation of trading strategies. It is also challenging because of the complex statistical properties of the real financial data. We introduce two generative adversarial networks (GANs), which utilize the convolutional networks with attention and the transformers, for financial time series simulation. The GANs learn the statistical properties in a data-driven manner and the attention mechanism helps to replicate the long-range dependencies. The proposed GANs are tested on the S&P 500 index and option data, examined by scores based on the stylized facts and are compared with the pure convolutional GAN, i.e. QuantGAN. The attention-based GANs not only reproduce the stylized facts, but also smooth the autocorrelation of returns.