1. Report Industry Investment Rating No relevant information provided. 2. Core View of the Report The report systematically reviews the evolution of financial time - series prediction models and constructs a prediction model for China's 10 - year Treasury bond yield using the Long Short - Term Memory (LSTM) neural network, with historical time - series as the single input variable, to explore the application of this deep - learning model in the fixed - income quantitative field [10]. 3. Summary by Directory 3.1 Financial Time - Series Prediction and Neural Network Models - **Evolution of Financial Time - Series Prediction Models**: Financial time - series prediction has gone through three main stages: traditional econometric models (e.g., ARIMA, GARCH), traditional machine - learning models (e.g., SVM, RF), and deep - learning models. Traditional econometric models have clear forms but struggle with nonlinear and complex dynamic relationships. Traditional machine - learning models can perform nonlinear fitting but need manual feature extraction. Deep - learning models can automatically extract features and capture long - term patterns, with RNN and its variants like LSTM being mainstream methods [11][12]. - **Neural Network Models and LSTM Models**: Neural network models mimic the human brain's neuron connection structure. RNN is designed for sequence data but has issues with long - term memory. LSTM solves the long - term dependency problem of RNN through a "gating mechanism" and memory units, enhancing robustness to irregular data. It is suitable for bond yield prediction due to its ability to handle long - term time series and filter noise [13][17][18]. 3.2 Treasury Bond Yield Prediction Based on Stacked LSTM Model - **Stacked LSTM Model**: Stacked LSTM connects multiple LSTM layers sequentially, offering advantages in long - sequence processing and multi - dimensional feature extraction, making it more suitable for complex financial time - series prediction [23]. - **Construction of Treasury Bond Yield Prediction Model**: - **Data Processing and Sample Construction**: The data is the yield of the 10 - year Treasury bond from the beginning of 2021 to December 12, 2025. First - order differences are calculated and standardized. Samples are constructed with the first - order differences of the past 60 trading days as input features and the first - order differences of the next week as the prediction target. The samples are divided into a training set (72%), a validation set (8%), and a test set (20%) [27]. - **Model Design and Evaluation**: The model architecture consists of LSTM, Dropout, and Dense layers. The training strategy involves 200 iterations with an early - stopping mechanism. Evaluation metrics include MSE, MAE, and RMSE [28][29]. - **Model Results**: A medium - complexity LSTM neural network model with about 130,000 adjustable parameters is built. The optimal model is obtained at the 27th iteration, and the early - stopping mechanism is triggered at the 77th iteration. The average absolute error for the test set is 1.43BP. The 10 - year Treasury bond yield is predicted to decline from December 15 - 19, 2025, with the predicted value on December 19, 2025, being 1.8330%, slightly lower than 1.8396% on December 12, 2025 [30]. 3.3 Follow - up Optimization Directions - **Model Design Optimization**: Adjust and optimize relevant designs such as time windows, data processing, network architecture, and training strategies [3][36]. - **Input Multi - Dimensional Variables**: Expand input variables from a single yield sequence to multi - dimensional variables such as macroeconomic, market, and sentiment variables to make the model more in line with economic logic and capture more comprehensive information [3][36]. - **Construct Hybrid Models**: Combine the LSTM model with traditional econometric models or other machine - learning models to build hybrid models like ARIMAX - LSTM and CNN - LSTM - ATT, leveraging different model advantages and improving prediction accuracy [3][36]. - **Introduce Rolling Back - testing Mechanism**: Use a rolling time - window back - testing mechanism to update the model dynamically and make continuous predictions, enhancing the model's adaptability to market changes [3][36].