FMamba: A Highly Scalable Financial Time-Series Prediction Model

Temporal information holds a pivotal role in the financial world, exemplified by stock market open and close prices, daily transaction volumes, and more. However, the complex fluctuations in this data present significant challenges to professionals. Accurate analysis and forecasting of financial time series trends are crucial for a deeper understanding of financial markets and for promoting stable economic development. Historically, a vast amount of research has focused on time series models, which often fall short in capturing high-dimensional features, resulting in suboptimal outcomes. Additionally, these models tend to have poor scalability and do not efficiently handle the vast amounts of data characteristic of today's era of big data. This paper introduces a highly scalable and robust framework, FMamba, which incorporates the recently proposed structured state space sequence model, Mamba, along with attention mechanisms. FMamba not only delves deeply into high-dimensional features of financial time series to provide precise predictions, enhancing professionals' grasp of financial markets, but also confidently handles large volumes of data with its scalability, ensuring that substantial datasets are no longer neglected due to model limitations.