Abstract:Stock price prediction is of significant importance in quantitative investment. Existing approaches encounter two primary issues: First, they often overlook the crucial role of capturing short-term stock fluctuations for predicting high-volatility returns. Second, mainstream methods, relying on graphs or attention mechanisms, inadequately explore the temporal relationships among stocks, often blurring distinctions in their characteristics over time and the causal relationships before and after. However, the high volatility of stocks and the intricate market correlations are crucial to accurately predicting stock prices. To address these challenges, we propose a Dual-branch Framework of Fluctuation and Trend (DFT), which decomposes stocks into trend and fluctuation components. By employing a carefully design decomposition module, DFT effectively extracts short-term fluctuations and trend information from stocks while explicitly modeling temporal variations and causal correlations. Our extensive experiments demonstrate that DFT outperforms existing methods across multiple metrics, including a 300% improvement in ranking metrics and a 400% improvement in portfolio-based indicators. Through detailed experiments, we provide valuable insights into different roles of trends and fluctuations in stock price prediction.

What problem does this paper attempt to address?

This paper attempts to solve two main problems in stock price prediction: 1. **Ignoring the importance of short - term fluctuations**: Existing methods often overlook the importance of capturing short - term stock fluctuations for predicting high - volatility returns. Short - term fluctuations contain important market information, which is crucial for accurately predicting stock prices. 2. **Insufficient modeling of temporal relationships**: Mainstream methods rely on graphs or attention mechanisms and fail to fully explore the temporal relationships between stocks. This leads to the blurring of the characteristics of different stocks in different periods and the confusion of causal relationships. The high volatility of the stock market and complex market correlations are key factors for accurately predicting stock prices. To address these problems, the authors propose a dual - branch framework (Dual - branch Framework of Fluctuation and Trend, DFT), which decomposes stocks into trend and fluctuation components. Through a carefully designed decomposition module, DFT can effectively extract short - term fluctuation and trend information from stocks and explicitly model temporal changes and causal relationships. ### Main contributions - **Proposing a new dual - branch stock prediction framework**: This framework effectively captures the correlations of time and stocks from both fluctuation and trend aspects. To the best of the authors' knowledge, this is the first stock price prediction model that uses a learning method to simultaneously capture long - term trend and short - term fluctuation information. - **Considering the role of differential representations of different time steps in time - series correlations and causal correlations**, fully exploiting the complex time - dependence in stock data. - **Verifying the design of the proposed method through experiments** and demonstrating its superiority over baseline methods. Through comprehensive analysis, it provides valuable insights into the different roles of trend and fluctuation in stock price prediction and the influence of temporal causal relationships. ### Method overview The DFT framework mainly includes three steps: 1. **Stock decomposition**: A decomposition module is designed to obtain the overall trend information and daily fluctuation information of each stock within the observation window \(T\). 2. **Correlation modeling**: Considering the different characteristics of fluctuation and trend information, independent branches containing time and stock correlations are designed for modeling. 3. **Output prediction**: Aggregate the time - step information output by the two branches and input it into the prediction layer to obtain the prediction result. ### Experimental results - **Overall performance comparison**: On three representative stock datasets, CSI300, CSI800, and S&P500, DFT has achieved an absolute lead over classical time - series prediction models and the latest stock price prediction SOTA methods. DFT has improved by 250% - 500% over SOTA on rank - based metrics and by 400% - 1000% on combination - based metrics. - **Ablation experiments**: The effectiveness of each component is verified by removing different modules (such as the decomposition module, fluctuation branch, trend branch, stock correlation module, and time correlation module). The experimental results show that each module has a significant impact on the model performance, especially the decomposition module and the fluctuation branch. - **Hyperparameter sensitivity**: Hyperparameter sensitivity experiments were carried out on the CSI800 dataset to explore the influence of different settings on the framework performance, including the average pooling kernel size, market convolution kernel size, RWKV head number, backtracking window length, and prediction interval. Through these experiments, the authors have demonstrated the effectiveness and robustness of the DFT framework, especially in dealing with the high volatility and complex correlations in the stock market.

DFT: A Dual-branch Framework of Fluctuation and Trend for Stock Price Prediction

A Dual-Attention-Based Stock Price Trend Prediction Model with Dual Features

A Stock Prediction Method Based on Multidimensional and Multilevel Feature Dynamic Fusion

Stock Trend Prediction Using Candlestick Charting and Ensemble Machine Learning Techniques with a Novelty Feature Engineering Scheme

A deep learning integrated framework for predicting stock index price and fluctuation via singular spectrum analysis and particle swarm optimization

Parallel Extraction of Long-term Trends and Short-term Fluctuation Framework for Multivariate Time Series Forecasting

Intraday Return Forecasts and High-Frequency Trading of Stock Index Futures: A Hybrid Wavelet-Deep Learning Approach

A novel data-driven stock price trend prediction system

Enhanced forecasting of stock prices based on variational mode decomposition, PatchTST, and adaptive scale-weighted layer

A novel stock forecasting model based on High-order-fuzzy-fluctuation Trends and Back Propagation Neural Network

A Hybrid Framework Based on Extreme Learning Machine, Discrete Wavelet Transform, and Autoencoder with Feature Penalty for Stock Prediction

HATR-I: Hierarchical Adaptive Temporal Relational Interaction for Stock Trend Prediction

Forecasting Based on High-Order Fuzzy-Fluctuation Trends and Particle Swarm Optimization Machine Learning

EMD2FNN: A strategy combining empirical mode decomposition and factorization machine based neural network for stock market trend prediction

A Hybrid Method Based on Extreme Learning Machine and Wavelet Transform Denoising for Stock Prediction

ALERTA-Net: A Temporal Distance-Aware Recurrent Networks for Stock Movement and Volatility Prediction

Self-FTS: A Self-Supervised Learning Method for Financial Time Series Representation in Stock Intraday Trading.

Enhancing Stock Price Trend Prediction Via a Time-Sensitive Data Augmentation Method

Deep learning framework for stock price prediction using long short-term memory

A Dual-View Model for Stock Price Prediction of Internet-of-Thing Enterprises

Multi-task Recurrent Neural Networks and Higher-order Markov Random Fields for Stock Price Movement Prediction