SINOLINK SECURITIES(SH:600109) SINOLINK SECURITIES·2025-10-13 08:48Report Industry Investment Rating - Not provided in the given content Core Viewpoints of the Report - The proposed multi - objective - oriented Tikhonov regression model can effectively measure the multi - asset positions of "Fixed Income +" funds, with high accuracy in identifying significant position adjustments and providing key value for fund evaluation and investment practice [3]. - The model can track the flow of funds in "Fixed Income +" funds and identify significant position - adjustment behaviors, helping investors capture market signals and select products with sustainable timing capabilities [3]. Summary by Relevant Catalogs I. Research Background and Model Introduction of "Fixed Income +" Fund Position Measurement 1. Background and Market - mainstream Model Methods Review - Measuring fund positions is necessary for understanding the asset structure of "Fixed Income +" funds, enhancing investment transparency, and strengthening risk management. In a low - interest - rate environment, "Fixed Income +" funds are important for capital inflow and stable value - added, and high - frequency position monitoring can provide basis for portfolio optimization [7]. - The current market - mainstream method for measuring fund positions is to construct a regression model based on fund net value and market indices and estimate positions through a portfolio optimization framework, which can provide high - frequency and dynamic position tracking [7]. 2. Systemic Dilemmas of Mainstream Measurement Models in "Fixed Income +" Fund Applications - Mainstream models cannot measure multi - asset portfolios, have low interpretability, significantly underestimate pure - bond positions, and have regression variable setting biases [8][9][10]. 3. Tikhonov Regression: A New Method for Solving Multi - asset Measurement Problems - Tikhonov regression is an extension of ridge regression, which can solve the multi - collinearity problem of multi - asset regression by designing a specific regularization matrix [12][14]. 4. Optimization of Tikhonov Regression Loss Function - Instead of using hard constraints, the loss function is optimized by introducing L1 and L2 regularization to achieve soft constraints, which can maintain numerical optimality and practical investment restrictions [17][18]. 5. Definition of Tikhonov Regression Equation - A unified modeling framework under 1/2 regularization conditions is defined, providing a basis for solving the optimal position [19][20]. 6. Solution of Regression Problem with L1/L2 Regularization - The gradient/ sub - gradient descent method is used to iteratively update parameters to approximate the optimal solution, obtaining the optimal position estimation of funds in various assets [21]. II. Implementation Process and Result Presentation of "Fixed Income +" Fund Position Measurement Model 1. Construction of Regression Indices - The regression indices of the fund pool are finely divided. The convertible bond and stock ends combine "individual characteristics" and "group characteristics" to construct indices, while the pure - bond end uses ChinaBond indices for regression according to the fund's position structure [26]. 2. Index Correlation - The average correlation coefficients within the pure - bond and stock index pools are relatively reasonable. By classifying funds and mapping industries, the correlation between indices is further reduced, improving the stability and reliability of regression results [32]. 3. Multi - objective - oriented Tikhonov Regression Algorithm - The data is constructed and pre - processed, the regression model is defined, the augmented matrix is built, and the solver is used to calculate. The results are visualized and analyzed to verify the rationality and interpretability of the model [34]. 4. Model Results - The model has high accuracy in measuring the positions of "Fixed Income +" funds. In August, the overall equity - containing position of "Fixed Income +" funds increased slightly, with a structural differentiation between stock and convertible bond positions [37]. 5. Error Distribution - The measurement errors of pure - bond and stock positions are relatively small and concentrated, while the error of convertible bond positions is relatively controllable. The model can be further optimized to improve the measurement accuracy of convertible bond positions [40]. 6. Classification by Equity - containing Exposure - The model can accurately track the asset allocation levels of various funds. For funds with different equity - containing exposures, the position centers and absolute errors of pure - bond, stock, and convertible bond positions vary [42]. 7. Stability of Large - scale Fund Position Measurement - The model has better stability in measuring the positions of large - scale funds, effectively tracking their dynamic position adjustments [46]. III. Applications of "Fixed Income +" Fund Position Measurement Model 1. High - frequency Tracking of Fund Flows - The model can effectively track the flow of funds in the stock positions of "Fixed Income +" funds. Most industries showed slight increases in positions this month, and the concentration of the top three industries increased, indicating a strengthening of market consensus. Leading funds have a more active equity allocation strategy [52]. 2. Identification of Significant Position - adjustment Behaviors - The model can identify significant position - adjustment behaviors with a win - rate of nearly 80%, which is helpful for evaluating the timing ability of fund managers and providing reference for asset allocation [54].