Comparison between transformer, informer, autoformer and non-stationary transformer in financial market

Abstract

This paper delves into the significance of predicting stock prices and carries out comparative experiments using a variety of models, including Support Vector Regression, Long Short-Term Memory model, Transformer, Informer, Autoformer, and Non-Stationary Transformer. These models are used to train and forecast the China Securities Index, Hang Seng Index, and S&P 500 Index. The results of the experiments are measured using indicators such as Mean Absolute Error and Root Mean Square Error. The findings show that the Non-Stationary Transformer model has the highest prediction accuracy. Additionally, a simple trading strategy is designed for each model and their Sharpe and Calmar ratios are compared. Since Autoformer has the highest Sharpe and Calmar, it can be concluded that Autoformer is the most practical in financial market among the four models. This research contributes to the field of stock price prediction by providing an empirical study on the application of Transformer and its derivative models which have been less explored in this domain. In conclusion, this paper offers valuable insights and recommendations for data scientists and financial engineer and introduces new methods for predicting stock prices.