1. Report Industry Investment Rating No industry investment rating is provided in the report. 2. Core Viewpoints of the Report - The prediction of the ten - year Treasury bond yield faces complex challenges from the data, model, and market ends, making high - precision prediction difficult to achieve. - XGBoost is the preferred model for predicting the ten - year Treasury bond yield due to its adaptability to bond market characteristics and technical advantages. - A prediction model for the ten - year Treasury bond yield is constructed using XGBoost, which can provide guidance for investment strategies [3][6][7]. 3. Summary by Related Catalogs 3.1 Deconstructing the Full Process of Treasury Yield Prediction - **Data End**: The factors affecting Treasury bond yields include short - term real interest rates, inflation expectations, and term premiums. The relevant data has a "low signal - to - noise ratio" and limited quantity, which may lead to the model capturing false patterns [6][21][27]. - **Model End**: From traditional machine learning to deep learning models, they have formed a complementary technology matrix. However, traditional machine learning has difficulties in dealing with non - linear relationships, and deep learning has problems such as the "black - box" feature and high data requirements [31][33][43]. - **Market End**: The "financial market uncertainty principle" exists in the financial market. Precise observation of the current state may interfere with long - term trend judgment, and trend prediction turning into collective action will reshape the market [54][55]. - **Coupling Resonance of the Three Ends**: The technical logic of the model relying on historical data and pursuing quantitative accuracy conflicts with the "financial market uncertainty principle", making it difficult to achieve accurate prediction [56]. 3.2 Reasons for Choosing the XGBoost Model - **Existing Model Ecology and Core Challenges**: Current models for predicting long - term Treasury bond yields have limitations. The prediction of the ten - year Treasury bond yield faces challenges such as market non - linearity and low volatility [64][65]. - **Model Foundation**: XGBoost is an efficient implementation of GBDT. Its core advantages include second - order Taylor expansion, sparse perception, feature parallelization, explicit regularization, and column sampling [70][74]. - **Comparison with Deep Learning Models**: XGBoost has advantages in data requirements, feature engineering, training efficiency, and interpretability when dealing with medium - and small - scale tabular data [75]. - **Scenario Adaptability**: XGBoost can output the "rise - fall direction", which is suitable for the practical logic of "direction first". It can efficiently process core indicators and ensure robustness and timeliness [77]. 3.3 Building Logic of the Prediction Model for the Ten - Year Treasury Bond Yield Change - **Data Processing**: The data from 2010 to the present is selected. The data is divided according to the "rolling window" principle. More than 200 indicators are selected, and cleaning and pre - processing are carried out, including extreme value truncation, missing value filling, standardization, and one - hot encoding [83][87][93]. - **Sample Weighting**: A "one - model - for - one - category" strategy is adopted to deal with the problem of category imbalance, improving the prediction accuracy of minority categories [97][98]. - **Parameter Tuning**: A "coarse - tuning + fine - tuning" method is used to balance model complexity and generalization ability, and the optimal parameters are determined [100][102][103]. - **Experience Backup**: When the number of samples predicted to be in a volatile market by the model is less than 50%, the prediction results of samples with the top 50% probability of the volatile market predicted by the volatile - market - specialized model are changed to volatile, reducing extreme risks [104]. 3.4 Model Win - rate Analysis and Guidance for Investment Strategies - **Model Win - rate**: The overall accuracy of the model reaches 92.0%, and the Cohen's kappa coefficient is 78.35%, indicating that the model can effectively capture interest rate change rules [9][115]. - **Investment Strategy Guidance**: When the model predicts that the ten - year Treasury bond yield will go "bullish", it is advisable to lengthen the duration; when it predicts "bearish", shorten the duration; when it predicts a "volatile" market, maintain the current investment portfolio [110][112][113].