Shenwan Hongyuan Securities·2025-05-09 09:15Quantitative Models and Construction Methods 1. Model Name: CSI 300 Index Adjustment Prediction Model - Model Construction Idea: The model predicts the adjustment of CSI 300 index constituents based on the official CSI 300 Index Compilation Scheme, using historical trading and market capitalization data as inputs [6][7] - Model Construction Process: 1. Select all A-shares listed for more than one quarter (one year for ChiNext and STAR Market stocks) 2. Filter stocks with the top 50% average daily trading volume over the past year 3. Rank the remaining stocks by average daily total market capitalization over the past year and select the top 300 stocks 4. Apply additional rules, including a 10% adjustment limit, priority retention of old samples, and a 20% buffer zone [6] 5. Adjustments are calculated using the formula: $ \text{Impact Coefficient} = \frac{\text{Passive Buy Amount} - \text{Passive Sell Amount}}{\text{Average Daily Trading Volume}} $ where the passive buy/sell amounts are estimated as: $ \text{Passive Buy Amount} = \text{Fund Size} \times \text{Estimated Weight} $ $ \text{Passive Sell Amount} = \text{Fund Size} \times \text{Original Weight} $ [7][10] - Model Evaluation: The model effectively identifies stocks with significant price impact due to passive fund rebalancing [7] 2. Model Name: CSI 500 Index Adjustment Prediction Model - Model Construction Idea: The model predicts adjustments to the CSI 500 index by excluding CSI 300 constituents and applying additional filters based on market capitalization and trading volume [11] - Model Construction Process: 1. Start with the CSI 300 sample space 2. Exclude CSI 300 constituents and stocks ranked in the top 300 by average daily market capitalization over the past year 3. Remove stocks in the bottom 20% by average daily trading volume over the past year 4. Select the top 500 stocks by average daily market capitalization from the remaining pool 5. Apply adjustment rules, including a 10% adjustment limit, priority retention of old samples, and a 10% buffer zone [11] 6. Impact coefficients are calculated similarly to the CSI 300 model [12] - Model Evaluation: The model captures the price impact of adjustments, particularly for stocks with high passive fund exposure [12] 3. Model Name: STAR 50 and STAR 100 Index Adjustment Prediction Models - Model Construction Idea: These models predict adjustments to STAR 50 and STAR 100 indices by filtering STAR Market stocks based on trading volume, market capitalization, and other criteria [18] - Model Construction Process: - STAR 50: 1. Select STAR Market stocks listed for more than six months (relaxed for high-market-cap stocks) 2. Exclude stocks with delisting risks or major violations 3. Remove the bottom 10% by average daily trading volume over the past year 4. Select the top 50 stocks by average daily market capitalization over the past year [18] - STAR 100: 1. Start with STAR Market stocks listed for more than six months 2. Exclude delisting-risk stocks, the bottom 10% by trading volume, STAR 50 constituents, and the top 40 by market capitalization 3. Select the top 100 stocks by average daily market capitalization from the remaining pool [18] - Impact coefficients are calculated using the same formula as the CSI 300 model [18][21] - Model Evaluation: The models are effective in identifying stocks with significant price impacts due to index adjustments, particularly for STAR 50 constituents [18] --- Model Backtesting Results 1. CSI 300 Model - Impact Coefficient: - Shanghai Electric: 1.29 - Sanan Optoelectronics: 4.41 - Liying Precision: 0.58 - Shanghai Rural Commercial Bank: 14.27 [8][9] 2. CSI 500 Model - Impact Coefficient: - Oriental Yuhong: -1.41 - Tianshan Cement: 2.59 - Zhengbang Technology: 2.21 - Hengxuan Technology: 3.04 [13][14][17] 3. STAR 50 and STAR 100 Models - Impact Coefficient: - STAR 50: - Dameng Data: 2.25 - Hengxuan Technology: 3.04 [21] - STAR 100: - Shanghai Crystal: 3.19 - Tianqi Lithium: 0.57 [22]