GUOTAI HAITONG SECURITIES·2026-03-21 13:37Quantitative Models and Construction Methods 1. Model Name: Sentiment Model - Model Construction Idea: The sentiment model is designed to measure the strength of market sentiment using factors related to limit-up and limit-down events[4][13] - Model Construction Process: The model uses factors such as net limit-up ratio, next-day return after limit-down, limit-up ratio, limit-down ratio, and high-frequency board-hitting returns to quantify sentiment. Each factor is scored, and the overall sentiment score is calculated on a scale of 0 to 5[13][17] - Model Evaluation: The sentiment model score is currently 0, indicating weak market sentiment[13][17] 2. Model Name: Trend Model - Model Construction Idea: The trend model identifies market trends by analyzing technical indicators and price movements[4][13] - Model Construction Process: The model uses the SAR (Stop and Reverse) indicator to detect trend reversals. For example, the Wind All-A Index broke below the SAR indicator on March 3, signaling a negative trend[13][14] - Model Evaluation: The trend model currently emits a negative signal, indicating a bearish market trend[13] 3. Model Name: Weighted Model - Model Construction Idea: The weighted model combines multiple signals, including sentiment and trend, to provide a comprehensive market outlook[4][13] - Model Construction Process: The weighted model aggregates signals from the sentiment and trend models, assigning weights to each component. The current weighted model signal is negative, reflecting overall bearishness in the market[13] - Model Evaluation: The model effectively integrates multiple signals but currently indicates a negative market outlook[13] 4. Model Name: High-Frequency Capital Flow Model - Model Construction Idea: This model uses high-frequency capital flow data to generate buy/sell signals for major broad-based indices[4][13] - Model Construction Process: The model tracks capital flow trends for indices such as CSI 300, CSI 500, CSI 1000, and CSI 2000. Signals are categorized into aggressive long, aggressive short, conservative long, and conservative short. For the week, all indices (CSI 300, CSI 500, CSI 1000, CSI 2000) showed negative signals (-1 for aggressive short and conservative short)[13][17] - Model Evaluation: The model provides timely signals for broad-based indices but currently reflects a bearish sentiment across all indices[13][17] --- Model Backtesting Results 1. Sentiment Model - Sentiment score: 0 (out of 5)[13][17] 2. Trend Model - SAR indicator: Negative signal for Wind All-A Index[13][14] 3. Weighted Model - Weighted signal: Negative[13] 4. High-Frequency Capital Flow Model - CSI 300: Aggressive short (-1), Conservative short (-1)[13][17] - CSI 500: Aggressive short (-1), Conservative short (-1)[13][17] - CSI 1000: Aggressive short (-1), Conservative short (-1)[13][17] - CSI 2000: Aggressive short (-1), Conservative short (-1)[13][17] --- Quantitative Factors and Construction Methods 1. Factor Name: Small-Cap Factor - Factor Construction Idea: Measures the performance of small-cap stocks and their crowding level[18][19] - Factor Construction Process: The factor's crowding level is calculated using metrics such as valuation spread, pairwise correlation, market volatility, and return reversal. The composite crowding score is 0.09[18][19] - Factor Evaluation: The small-cap factor shows a slight increase in crowding, indicating moderate interest from investors[18][19] 2. Factor Name: Low-Valuation Factor - Factor Construction Idea: Tracks the performance of low-valuation stocks and their crowding level[18][19] - Factor Construction Process: The factor's crowding level is calculated using the same metrics as the small-cap factor. The composite crowding score is -0.31[18][19] - Factor Evaluation: The low-valuation factor shows a decrease in crowding, suggesting reduced investor interest[18][19] 3. Factor Name: High-Profitability Factor - Factor Construction Idea: Measures the performance of high-profitability stocks and their crowding level[18][19] - Factor Construction Process: The factor's crowding level is calculated using the same metrics as the small-cap factor. The composite crowding score is 0.24[18][19] - Factor Evaluation: The high-profitability factor shows moderate crowding, indicating stable investor interest[18][19] 4. Factor Name: High-Growth Factor - Factor Construction Idea: Tracks the performance of high-growth stocks and their crowding level[18][19] - Factor Construction Process: The factor's crowding level is calculated using the same metrics as the small-cap factor. The composite crowding score is 0.25[18][19] - Factor Evaluation: The high-growth factor shows moderate crowding, reflecting steady investor interest[18][19] --- Factor Backtesting Results 1. Small-Cap Factor - Composite crowding score: 0.09[18][19] 2. Low-Valuation Factor - Composite crowding score: -0.31[18][19] 3. High-Profitability Factor - Composite crowding score: 0.24[18][19] 4. High-Growth Factor - Composite crowding score: 0.25[18][19]