Multivariate Financial Time Series Forecasting Model Based on Transformer Architecture
Ruibo Wu · 2024
This paper focuses on the design and application of a multivariate financial time series prediction model based on the Transformer, with the goal of enhancing forecast accuracy and reliability in financial market time series information. This passage begins by introducing the fundamental concept behind the Transformer model, with a particular emphasis on the utilization of the self-attention mechanism and the multi-head self-attention mechanism. Experiments will be done next in order to confirm that the model works with effectiveness in predicting stock prices and analyzing the foreign exchange market. The results show that the model can accurately track the complex dynamics in the financial market and outperforms traditional methods on different data sets. The findings of this study show that the Transformer model has great potential in predicting financial time series and offers significant backing for financial decision-making.