""" 资金流因子模块 包含基于股票截面的资金流因子实现 """ import numpy as np import polars as pl from main.factor.operator_framework import StockWiseFactor class LGFlowFactor(StockWiseFactor): """大单净买量因子""" def __init__(self): super().__init__( name="lg_flow", parameters={}, required_factor_ids=["buy_lg_vol", "buy_elg_vol", "sell_lg_vol", "sell_elg_vol"] ) def calc_factor(self, group_df: pl.DataFrame) -> pl.Series: # 计算大单净买量 buy_lg = group_df["buy_lg_vol"] buy_elg = group_df["buy_elg_vol"] sell_lg = group_df["sell_lg_vol"] sell_elg = group_df["sell_elg_vol"] lg_net_flow = (buy_lg + buy_elg) - (sell_lg + sell_elg) return lg_net_flow.alias(self.factor_id) class FlowIntensityFactor(StockWiseFactor): """资金流强度因子""" def __init__(self): super().__init__( name="flow_intensity", parameters={}, required_factor_ids=["buy_lg_vol", "buy_elg_vol", "sell_lg_vol", "sell_elg_vol", "vol"] ) def calc_factor(self, group_df: pl.DataFrame) -> pl.Series: # 计算资金流强度 buy_lg = group_df["buy_lg_vol"] buy_elg = group_df["buy_elg_vol"] sell_lg = group_df["sell_lg_vol"] sell_elg = group_df["sell_elg_vol"] vol = group_df["vol"] lg_net_flow = (buy_lg + buy_elg) - (sell_lg + sell_elg) flow_intensity = lg_net_flow / (vol + 1e-8) # 避免除零 return flow_intensity.alias(self.factor_id) class FlowDivergenceFactor(StockWiseFactor): """资金流背离因子""" def __init__(self): super().__init__( name="flow_divergence", parameters={}, required_factor_ids=["buy_sm_vol", "sell_sm_vol", "buy_lg_vol", "buy_elg_vol", "sell_lg_vol", "sell_elg_vol"] ) def calc_factor(self, group_df: pl.DataFrame) -> pl.Series: # 计算资金流背离度 buy_sm = group_df["buy_sm_vol"] sell_sm = group_df["sell_sm_vol"] buy_lg = group_df["buy_lg_vol"] buy_elg = group_df["buy_elg_vol"] sell_lg = group_df["sell_lg_vol"] sell_elg = group_df["sell_elg_vol"] sm_net_flow = buy_sm - sell_sm lg_net_flow = (buy_lg + buy_elg) - (sell_lg + sell_elg) flow_divergence = sm_net_flow - lg_net_flow return flow_divergence.alias(self.factor_id) class FlowStructureFactor(StockWiseFactor): """资金流结构因子""" def __init__(self): super().__init__( name="flow_structure", parameters={}, required_factor_ids=["buy_sm_vol", "buy_lg_vol", "buy_elg_vol", "vol"] ) def calc_factor(self, group_df: pl.DataFrame) -> pl.Series: # 计算资金流结构 buy_sm = group_df["buy_sm_vol"] buy_lg = group_df["buy_lg_vol"] buy_elg = group_df["buy_elg_vol"] vol = group_df["vol"] total_buy = buy_sm + buy_lg + buy_elg lg_elg_buy_prop = (buy_lg + buy_elg) / (total_buy + 1e-8) # 避免除零 flow_structure = lg_elg_buy_prop.diff().alias(self.factor_id) return flow_structure class FlowAccelerationFactor(StockWiseFactor): """资金流加速度因子""" def __init__(self): super().__init__( name="flow_acceleration", parameters={}, required_factor_ids=["buy_lg_vol", "buy_elg_vol", "sell_lg_vol", "sell_elg_vol"] ) def calc_factor(self, group_df: pl.DataFrame) -> pl.Series: # 计算资金流加速度 buy_lg = group_df["buy_lg_vol"] buy_elg = group_df["buy_elg_vol"] sell_lg = group_df["sell_lg_vol"] sell_elg = group_df["sell_elg_vol"] lg_net_flow = (buy_lg + buy_elg) - (sell_lg + sell_elg) lg_net_flow_change = lg_net_flow.diff() flow_acceleration = lg_net_flow_change.diff() return flow_acceleration.alias(self.factor_id) class CostSqueeze(StockWiseFactor): factor_id = "factor_cost_squeeze" required_factor_ids = ["close", "cost_15pct", "cost_50pct", "cost_85pct", "vol"] def __init__(self): super().__init__( name=self.factor_id, parameters={}, required_factor_ids=self.required_factor_ids ) def calc_factor(self, g: pl.DataFrame) -> pl.Series: close = g["close"] cost15 = g["cost_15pct"] cost50 = g["cost_50pct"] cost85 = g["cost_85pct"] vol = g["vol"] cost_range = cost85 - cost15 median_cost = cost50 price_pos = (close - median_cost) / (cost_range + 1e-6) vol_5d = vol.rolling_mean(window_size=5, min_periods=1) vol_ratio = vol / (vol_5d + 1e-6) # 核心逻辑：成本区间窄 + 价格居中 + 量能萎缩 → 高挤压度 squeeze_score = ( (1.0 / (cost_range / (close + 1e-6) + 1e-6)) * (1.0 - price_pos.abs()) * (1.0 / (vol_ratio + 1e-6)) ) # 稳态化：对数变换 factor = (squeeze_score + 1.0).log() return factor.alias(self.factor_id) class HighCostSelling(StockWiseFactor): factor_id = "factor_high_cost_selling" required_factor_ids = ["close", "cost_85pct", "buy_sm_vol", "sell_lg_vol", "vol"] def __init__(self): super().__init__( name=self.factor_id, parameters={}, required_factor_ids=self.required_factor_ids ) def calc_factor(self, g: pl.DataFrame) -> pl.Series: close = g["close"] cost85 = g["cost_85pct"] buy_sm_vol = g["buy_sm_vol"] sell_lg_vol = g["sell_lg_vol"] vol = g["vol"] is_above_85 = (close > cost85).cast(pl.Float64) small_buy_ratio = buy_sm_vol / (vol + 1e-6) large_sell_ratio = sell_lg_vol / (vol + 1e-6) hcsp = is_above_85 * small_buy_ratio * large_sell_ratio # 稳态化：取对数（避免极端值） factor = (hcsp + 1e-6).log() return factor.alias(self.factor_id) class LowCostAccumulation(StockWiseFactor): factor_id = "factor_low_cost_accumulation" required_factor_ids = ["close", "his_low", "cost_15pct", "buy_lg_vol", "buy_elg_vol", "vol"] def __init__(self): super().__init__( name=self.factor_id, parameters={}, # 无参数，可扩展 required_factor_ids=self.required_factor_ids ) def calc_factor(self, g: pl.DataFrame) -> pl.Series: close = g["close"] his_low = g["his_low"] cost15 = g["cost_15pct"] buy_lg_vol = g["buy_lg_vol"] buy_elg_vol = g["buy_elg_vol"] vol = g["vol"] is_below_15 = (close < cost15).cast(pl.Float64) # 近5日最低价（含当日） rolling_min_5 = his_low.rolling_min(window_size=5, min_periods=1) # 注意：his_low 通常是历史最低，但这里我们用 close 的滚动最小更合理 # 修正：应使用 close 的滚动最小判断是否新低 close_rolling_min_5 = close.rolling_min(window_size=5, min_periods=1) not_new_low = (close >= close_rolling_min_5).cast(pl.Float64) big_buy_vol = buy_lg_vol + buy_elg_vol big_buy_ratio = big_buy_vol / (vol + 1e-6) lc_am = is_below_15 * not_new_low * big_buy_ratio # 稳态化 factor = (lc_am + 1e-6).log() return factor.alias(self.factor_id) class InstNetAccum(StockWiseFactor): factor_id = "inst_net_accum" required_factor_ids = ["close", "buy_lg_vol", "buy_elg_vol", "sell_lg_vol", "sell_elg_vol", "circ_mv"] def __init__(self): super(InstNetAccum, self).__init__( name=self.factor_id, parameters={}, required_factor_ids=self.required_factor_ids ) def calc_factor(self, g: pl.DataFrame) -> pl.Series: close = g["close"] buy_lg = g["buy_lg_vol"] buy_elg = g["buy_elg_vol"] sell_lg = g["sell_lg_vol"] sell_elg = g["sell_elg_vol"] circ_mv = g["circ_mv"] big_net_vol = (buy_lg + buy_elg) - (sell_lg + sell_elg) circ_shares = circ_mv / (close + 1e-6) ina = big_net_vol / (circ_shares + 1e-6) # 3日收益率（抑制大涨） ret3 = close / close.shift(3) - 1 ret3 = ret3.fill_null(strategy="forward").fill_null(0.0) ina = pl.when(ret3.abs() < 0.05).then(ina).otherwise(0.0) return ina.log1p().alias(self.factor_id) class ChipLockin(StockWiseFactor): factor_id = "chip_lockin" required_factor_ids = ["cost_5pct", "cost_95pct", "winner_rate"] def __init__(self): super(ChipLockin, self).__init__( name=self.factor_id, parameters={}, required_factor_ids=self.required_factor_ids ) def calc_factor(self, g: pl.DataFrame) -> pl.Series: cost5 = g["cost_5pct"] cost95 = g["cost_95pct"] winner_rate = g["winner_rate"] cost_width = cost95 - cost5 width_5d = cost_width.rolling_mean(window_size=5, min_periods=1) width_10d = cost_width.rolling_mean(window_size=10, min_periods=1) # 避免除零 width_contraction = (width_10d - width_5d) / (width_10d + 1e-6) winner_std_5 = winner_rate.rolling_std(window_size=5, min_periods=1).fill_null(1e-6) clm = width_contraction * (1.0 / (winner_std_5 + 1e-6)) return clm.log1p().alias(self.factor_id) class RetailOutInstIn(StockWiseFactor): factor_id = "retail_out_inst_in" required_factor_ids = ["close", "buy_sm_vol", "sell_sm_vol", "buy_lg_vol", "buy_elg_vol", "sell_lg_vol", "sell_elg_vol"] def __init__(self): super(RetailOutInstIn, self).__init__( name=self.factor_id, parameters={}, required_factor_ids=self.required_factor_ids ) def calc_factor(self, g: pl.DataFrame) -> pl.Series: close = g["close"] buy_sm = g["buy_sm_vol"] sell_sm = g["sell_sm_vol"] buy_lg = g["buy_lg_vol"] buy_elg = g["buy_elg_vol"] sell_lg = g["sell_lg_vol"] sell_elg = g["sell_elg_vol"] small_net_out = sell_sm - buy_sm big_net_in = (buy_lg + buy_elg) - (sell_lg + sell_elg) # 价格抗跌：近5日未破位（用 close 自身滚动最小） close_min_5 = close.rolling_min(window_size=5, min_periods=1) roii = small_net_out * big_net_in return roii.log1p().alias(self.factor_id) class AccumAccel(StockWiseFactor): factor_id = "accum_accel" required_factor_ids = ["buy_lg_vol", "buy_elg_vol", "sell_lg_vol", "sell_elg_vol", "vol"] def __init__(self): super(AccumAccel, self).__init__( name=self.factor_id, parameters={}, required_factor_ids=self.required_factor_ids ) def calc_factor(self, g: pl.DataFrame) -> pl.Series: buy_lg = g["buy_lg_vol"] buy_elg = g["buy_elg_vol"] sell_lg = g["sell_lg_vol"] sell_elg = g["sell_elg_vol"] vol = g["vol"] big_net_vol = (buy_lg + buy_elg) - (sell_lg + sell_elg) big_net_ratio = big_net_vol / (vol + 1e-6) net_5d = big_net_ratio.rolling_mean(window_size=5, min_periods=1) net_5d_lag = net_5d.shift(5).fill_null(strategy="forward").fill_null(0.0) acceleration = net_5d - net_5d_lag return acceleration.log1p().alias(self.factor_id)