Quantitative Models and Construction Methods - **Model Name**: Convertible Bond K-line Shape Factor **Construction Idea**: Quantify K-line shapes to extract predictive information for convertible bonds, combining price trends and volume trends for enhanced forecasting ability [15][33][96] **Construction Process**: 1. Define single K-line shapes based on absolute values of body, upper shadow, and lower shadow, dividing them into 16 types [22][33][34] 2. Extend to multi-K-line shapes by incorporating relationships between adjacent K-lines, such as gap openings (e.g., high/low opening) [23][24][33] 3. Combine K-line shapes with price trends (e.g., position of closing price in historical range) and volume trends (e.g., shrinking, fluctuating, or increasing volume) for more complete predictive signals [25][33][96] **Formula**: - Single K-line scoring: $score_{p} = \frac{mean(r_{p})}{std(r_{p})}$ where $r_{p}$ represents the future 20-day returns of K-line shape $p$ [34][35] - Multi-K-line scoring: $KP_{c,T} = \sum_{t=T-40}^{T} w_{t} * score_{c,t}$ $w_{t} = 0.5^{\frac{T-t}{\lambda}}$ where $w_{t}$ is the exponential decay weight, $\lambda$ is half the window period [36][37] **Evaluation**: Effective in predicting convertible bond returns, especially post-2022, with stable IC and IR values [97][93][87] Model Backtesting Results - **Convertible Bond K-line Shape Factor**: - IC Mean: 0.11 [87][93][97] - IC Win Rate: 66.31% [87][93][97] - IR: 0.43 [87][93][97] - Group 1 Annualized Return: 6.34% [88][89][97] - Excess Return Win Rate: 70.08% [88][89][97] Quantitative Factors and Construction Methods - **Factor Name**: Single K-line Shape Factor **Construction Idea**: Quantify single K-line shapes to identify predictive patterns for convertible bonds [22][33][96] **Construction Process**: 1. Divide K-line body, upper shadow, and lower shadow into short/long categories based on absolute thresholds (e.g., body > 2%, shadow > 1%) [22][33][34] 2. Combine these components to form 16 distinct single K-line shapes [22][33][34] **Evaluation**: Certain shapes (e.g., A6, A7) show high predictive power for short-term returns [42][46][54] - **Factor Name**: Multi-K-line Shape Factor **Construction Idea**: Incorporate relationships between adjacent K-lines to enhance predictive accuracy [23][33][96] **Construction Process**: 1. Define relationships such as gap openings (e.g., high/low opening) between adjacent K-lines [23][24][33] 2. Combine these relationships with single K-line shapes to form 768 distinct 2K shapes [24][33][66] **Evaluation**: Certain combinations (e.g., B5 + B3) are strong signals for upward trends [66][69][72] Factor Backtesting Results - **Single K-line Shape Factor**: - A6 Shape: Future 5-day excess return 48.04%, win rate > 60% [54][56][58] - A7 Shape: Future 5-day excess return 67.40%, win rate > 60% [54][56][58] - **Multi-K-line Shape Factor**: - B5 + B3 Shape: Future 5-day excess return 214.74% [66][69][71] - A5 + A6 Shape: Future 5-day excess return 208.06% [66][69][70] Factor Correlation Analysis - **Convertible Bond K-line Shape Factor**: - Correlation with other factors (e.g., implied volatility, YTM) is low, indicating unique information contribution [93][94][97]