多因子及AI量化选股框架

2025/3/29 1 2 多因子投资模型迭代 3 4  𝛽� 𝛼�  i 𝐸[𝑅𝑒]=𝛼�+𝛽𝑖� 𝛽�� *0 * * 𝛼� * * * ** ** 5 1. 6 数据清洗与标准化 7 8 2) 9 5) K-MeansDBSCAN 10 � 11 12 ForwardFill BackwardFill LinearInterpolation ARIMA 单因子研究 13 14 2 CAPM FF3 FF3 ROEROAROTC 15 16 17 18 5 RangeFactorWAdjCI 3.36% Wind 19 20 21 𝑋t−1 𝑟� 22 n di i IC -11 IC=1 IC=-1 IC=0 IC • • • Wind EPS_FY0_Num-IC EPS_FY0_Num-IC的移动平均 23 50.00% 30.00% 10 IC 10.00% -10.00% -30.00% IC t EPS_FY0_Num 2.32% 17.69% -47.00% 46.36% 13% 3.12 -50.00% 2016/1/82018/1/82020/1/8 2022/1/8 24 12 IC BP_LR 1212 ()TTM TTM/TTM/ 20 1200 120180 CapexSales 3 EPS1EPS3 EPS1 EP_Deducted_TTM FCFP_TTMOCFP_TTM Amount_Mean_20D Asset_Ln Employee_NumRevenues_LnAsset_Chg1YCurrentAsset_Chg1YCurrentDebt_Chg1YCurrentRatioPrice_Chg1200D Price_Chg120DPrice_Chg180D Capex2SalesNetIncome_Chg1YNetIncome_Chg3YEPS_FTTM_Chg1MEPS_FTTM_Chg3M EPS_FY0_Chg1M 线性多因子模型 25 ϵ • • • • •     30 AI与另类选股模型 31 32 • • • • • • 33 𝑀𝐴�𝑒𝑑𝑖𝑎��𝑀𝑒𝑑𝑖𝑎�� 34 35 1. 1 2 3 ROE RSI PCA 2. 1 2 3 4 XGBoostLightGBM K-means CNN t-SNE RNN/LSTM Transformer OpenFE CatBoost CategoricalFeatures Boosting 36 XGBoost GBDT LightGBM CatBoost XGBoost GBDT L1/L2 LightGBMHistogramLeaf-wise 37 GRU NN TCN Transformer 38 • • • 39 LLM 1. 2. 3. / LLM LLM FinBERT 40 41 1000 ChatGPT4 FinGPT BERT- TextCNN 13.95% 1.56 股评情感分类 千万别再跌了啊，再跌我就要赔一百元了啊消极 底仓涨停了积极 此股在过年之前应该会稳定在35左右，基本不会有长进，慎入中性 ChatGPT ChatGPT 2020202120222023 47.74%26.37% 202410 13.58%20.03% 9.29% 42 GPT-4o ChatGPT 风险模型框架 43 44 K riikfkk1 i 45 𝛽ik 𝑓� 𝜀� =0 corr(𝜀𝑖,𝜀j BBTD 46 47 投资组合管理 48 UMVwifi 1AwTw ii  2 49 50 51 52 53