A Labeling Method for Financial Time Series Prediction Based on Trends.

Area of Science:

• * Financial econometrics and time series analysis.
• * Machine learning applications in finance.

Background:

• * Time series prediction is crucial for financial applications like stock and commodity price forecasting.
• * Current financial time series labeling methods often fail to capture continuous trends due to data non-linearity and short-term randomness.
• * Existing methods compare current data with short future periods, creating a gap between labeled results and real market trends.

Purpose of the Study:

• * To introduce a novel 'continuous trend labeling' method for financial time series.
• * To address the limitations of existing labeling techniques in accurately reflecting market dynamics.
• * To improve the prediction accuracy of machine learning models in financial forecasting.

Main Methods:

• * Development of a 'continuous trend labeling' method to extract long-term features.
• * Implementation of a novel feature preprocessing technique to avoid look-ahead bias.
• * Creation of an automatic labeling algorithm for continuous trend extraction.

Main Results:

• * The proposed continuous trend labeling method significantly outperforms existing state-of-the-art methods.
• * Demonstrated superior classification accuracy and other evaluation metrics on Chinese stock market data (Shanghai and Shenzhen Indices).
• * Validated the suitability of deep learning models like LSTM and GRU for financial time series prediction using the new labeling method.

Conclusions:

• * Continuous trend labeling offers a more accurate approach to labeling financial time series data.
• * The method effectively captures essential trend features missed by traditional techniques.
• * This advancement enhances the reliability of machine learning models for financial forecasting and investment decision-making.