193 lines
9.3 KiB
Python
193 lines
9.3 KiB
Python
import pandas as pd
|
||
import numpy as np
|
||
from scipy.stats import spearmanr # 用于因子3的原始思路，但实际简化了
|
||
|
||
epsilon = 1e-10
|
||
|
||
def _safe_divide(numerator, denominator, default_val=0.0):
|
||
"""安全除法"""
|
||
with np.errstate(divide='ignore', invalid='ignore'):
|
||
result = numerator / denominator
|
||
result[~np.isfinite(result)] = default_val
|
||
return result
|
||
|
||
# --- 修改后的因子计算函数 ---
|
||
|
||
def calculate_size_style_strength_factor(df: pd.DataFrame, N: int = 5, factor_name_suffix: str = '') -> pd.DataFrame:
|
||
"""
|
||
计算大小盘风格相对强度因子。
|
||
返回: 以 trade_date 为索引，因子值为列的 DataFrame。
|
||
"""
|
||
factor_name = f'size_style_strength_{N}{factor_name_suffix}'
|
||
print(f"Calculating {factor_name}...")
|
||
|
||
required_indices = ['399300.SZ', '000905.SH', '000852.SH']
|
||
if not all(idx in df['ts_code'].unique() for idx in required_indices):
|
||
print(f"Error: DataFrame 中缺少部分必需的指数代码 ({required_indices})。返回空因子 Series。")
|
||
return pd.DataFrame(index=df['trade_date'].unique(), columns=[factor_name]).rename_axis('trade_date')
|
||
|
||
# 1. 计算各指数N日收益率
|
||
df_copy = df.copy() # 操作副本，避免修改原始传入df
|
||
df_copy['_ret_N'] = df_copy.groupby('ts_code')['close'].pct_change(periods=N)
|
||
|
||
# 2. Pivot 以方便截面计算
|
||
pivot_ret_N = df_copy.pivot_table(index='trade_date', columns='ts_code', values='_ret_N')
|
||
|
||
# 确保列存在并获取
|
||
large_ret = pivot_ret_N.get('399300.SZ', pd.Series(np.nan, index=pivot_ret_N.index))
|
||
mid_ret = pivot_ret_N.get('000905.SH', pd.Series(np.nan, index=pivot_ret_N.index))
|
||
small_ret = pivot_ret_N.get('000852.SH', pd.Series(np.nan, index=pivot_ret_N.index))
|
||
|
||
# 3. 计算因子 (结果是每日一个标量值)
|
||
large_small_diff = large_ret - small_ret
|
||
avg_large_small_ret = (large_ret + small_ret) / 2
|
||
# 计算中盘偏离因子，处理NaN，如果中盘收益为NaN，则偏离因子不起调整作用（乘以1）
|
||
mid_deviation_raw = mid_ret - avg_large_small_ret
|
||
mid_deviation_factor = 1 + np.sign(mid_ret.fillna(0)) * np.abs(mid_deviation_raw.fillna(0))
|
||
|
||
daily_factor_values = large_small_diff * mid_deviation_factor
|
||
daily_factor_values.name = factor_name # 给 Series 命名
|
||
|
||
print(f"Finished {factor_name}.")
|
||
return daily_factor_values.to_frame() # 转换为 DataFrame 返回
|
||
|
||
def calculate_volatility_structure_factor(df: pd.DataFrame, N: int = 10, factor_name_suffix: str = '') -> pd.DataFrame:
|
||
"""
|
||
计算市场波动结构因子。
|
||
返回: 以 trade_date 为索引，因子值为列的 DataFrame。
|
||
"""
|
||
factor_name = f'vol_structure_idx_{N}{factor_name_suffix}'
|
||
print(f"Calculating {factor_name}...")
|
||
|
||
required_indices = ['399300.SZ', '000905.SH', '000852.SH']
|
||
if not all(idx in df['ts_code'].unique() for idx in required_indices):
|
||
print(f"Error: DataFrame 中缺少部分必需的指数代码 ({required_indices})。返回空因子 Series。")
|
||
return pd.DataFrame(index=df['trade_date'].unique(), columns=[factor_name]).rename_axis('trade_date')
|
||
|
||
if 'pct_chg' not in df.columns:
|
||
print(f"Error: DataFrame 缺少 'pct_chg' 列。将为 {factor_name} 填充 NaN。")
|
||
return pd.DataFrame(index=df['trade_date'].unique(), columns=[factor_name]).rename_axis('trade_date')
|
||
|
||
df_copy = df.copy()
|
||
# 1. 计算各指数N日波动率
|
||
df_copy['_vol_N'] = df_copy.groupby('ts_code')['pct_chg'].rolling(N, min_periods=max(1, N//2)).std().reset_index(level=0, drop=True)
|
||
|
||
# 2. Pivot
|
||
pivot_vol_N = df_copy.pivot_table(index='trade_date', columns='ts_code', values='_vol_N')
|
||
|
||
large_vol = pivot_vol_N.get('399300.SZ', pd.Series(np.nan, index=pivot_vol_N.index))
|
||
mid_vol = pivot_vol_N.get('000905.SH', pd.Series(np.nan, index=pivot_vol_N.index))
|
||
small_vol = pivot_vol_N.get('000852.SH', pd.Series(np.nan, index=pivot_vol_N.index))
|
||
|
||
# 3. 计算因子
|
||
daily_factor_values = _safe_divide((small_vol - mid_vol), large_vol)
|
||
daily_factor_values.name = factor_name
|
||
|
||
print(f"Finished {factor_name}.")
|
||
return daily_factor_values.to_frame()
|
||
|
||
def calculate_market_divergence_factor(df: pd.DataFrame, factor_name_suffix: str = '') -> pd.DataFrame:
|
||
"""
|
||
计算市场分化度因子 (基于每日三个指数收益率符号的一致性)。
|
||
返回: 以 trade_date 为索引，因子值为列的 DataFrame。
|
||
"""
|
||
factor_name = f'market_divergence_score{factor_name_suffix}'
|
||
print(f"Calculating {factor_name}...")
|
||
|
||
required_indices = ['399300.SZ', '000905.SH', '000852.SH']
|
||
if not all(idx in df['ts_code'].unique() for idx in required_indices):
|
||
print(f"Error: DataFrame 中缺少部分必需的指数代码 ({required_indices})。返回空因子 Series。")
|
||
return pd.DataFrame(index=df['trade_date'].unique(), columns=[factor_name]).rename_axis('trade_date')
|
||
|
||
if 'pct_chg' not in df.columns:
|
||
print(f"Error: DataFrame 缺少 'pct_chg' 列。将为 {factor_name} 填充 NaN。")
|
||
return pd.DataFrame(index=df['trade_date'].unique(), columns=[factor_name]).rename_axis('trade_date')
|
||
|
||
pivot_pct_chg = df.pivot_table(index='trade_date', columns='ts_code', values='pct_chg')
|
||
|
||
# 确保列存在
|
||
idx_large_col = '399300.SZ'
|
||
idx_mid_col = '000905.SH'
|
||
idx_small_col = '000852.SH'
|
||
|
||
# 使用 reindex 确保所有期望的列都存在，缺失的填充NaN
|
||
pivot_pct_chg = pivot_pct_chg.reindex(columns=[idx_large_col, idx_mid_col, idx_small_col])
|
||
|
||
def daily_divergence_score_calc(row):
|
||
# 当天只有这三个指数的收益率 Series
|
||
valid_returns = row.dropna() # 获取非 NaN 的收益率
|
||
if len(valid_returns) < 2: # 如果有效数据少于2个，无法判断分化
|
||
return np.nan
|
||
|
||
signs = np.sign(valid_returns)
|
||
unique_sign_count = len(signs.unique())
|
||
|
||
if unique_sign_count == 1: # 所有符号相同 (或都为0，sign后也是0)
|
||
return 0.0 # 分化度最低 (高度一致)
|
||
elif unique_sign_count == 2 and 0 in signs.unique(): # 一个方向，一个0
|
||
return 0.25 # 较低分化
|
||
elif unique_sign_count == 2: # 两个方向 (例如两正一负，或两负一正)
|
||
return 0.75 # 较高分化
|
||
elif unique_sign_count == 3: # 三个不同方向 (+, -, 0)
|
||
return 1.0 # 分化度最高
|
||
return np.nan # 其他未覆盖的情况 (理论上不应发生)
|
||
|
||
daily_factor_values = pivot_pct_chg[[idx_large_col, idx_mid_col, idx_small_col]].apply(daily_divergence_score_calc, axis=1)
|
||
daily_factor_values.name = factor_name
|
||
|
||
print(f"Finished {factor_name}.")
|
||
return daily_factor_values.to_frame()
|
||
|
||
# --- 整合所有因子计算到一个主函数 ---
|
||
def generate_daily_index_relation_factors(df_input: pd.DataFrame) -> pd.DataFrame:
|
||
"""
|
||
计算所有基于大中小盘指数关系的每日截面因子。
|
||
|
||
Args:
|
||
df_input (pd.DataFrame): 长格式的指数行情数据，包含 'ts_code', 'trade_date', 'close', 'pct_chg'。
|
||
|
||
Returns:
|
||
pd.DataFrame: 以 'trade_date' 为索引，各因子为列的 DataFrame。
|
||
"""
|
||
# 确保输入 df 不被修改
|
||
df = df_input.sort_values(['ts_code', 'trade_date']).reset_index(drop=True)
|
||
|
||
# 计算各个因子 (每个函数返回一个单列或多列的 DataFrame，索引为 trade_date)
|
||
factor1_df = calculate_size_style_strength_factor(df, N=5)
|
||
factor2_df = calculate_volatility_structure_factor(df, N=10)
|
||
factor3_df = calculate_market_divergence_factor(df)
|
||
|
||
# 还可以继续添加其他每日截面因子...
|
||
|
||
# 合并所有因子 DataFrame
|
||
# 使用 functools.reduce 和 pd.merge 来优雅地合并多个 DataFrame
|
||
from functools import reduce
|
||
daily_factors_list = [factor1_df, factor2_df, factor3_df]
|
||
# 过滤掉可能因错误产生的完全为空或只有NaN的DataFrame
|
||
daily_factors_list = [f_df for f_df in daily_factors_list if not f_df.empty and not f_df.iloc[:,0].isna().all()]
|
||
|
||
if not daily_factors_list:
|
||
print("警告: 未能成功计算任何因子。返回空 DataFrame。")
|
||
# 返回一个以日期为索引的空DataFrame，或者基于输入df的日期
|
||
return pd.DataFrame(index=df['trade_date'].unique()).rename_axis('trade_date')
|
||
|
||
# 使用 outer join 以保留所有日期，并确保索引是 trade_date
|
||
final_factors_df = reduce(lambda left, right: pd.merge(left, right, on='trade_date', how='outer'),
|
||
daily_factors_list)
|
||
|
||
final_factors_df = final_factors_df.sort_index() # 按日期排序
|
||
|
||
return final_factors_df
|
||
|
||
# --- 使用示例 ---
|
||
# 假设 all_indices_df 是你包含 '399300.SZ', '000905.SH', '000852.SH' 三个指数的长格式行情数据
|
||
# 确保它有 'ts_code', 'trade_date', 'open', 'high', 'low', 'close', 'vol', 'pct_chg' 列
|
||
# all_indices_df['trade_date'] = pd.to_datetime(all_indices_df['trade_date'])
|
||
# all_indices_df = all_indices_df.sort_values(['ts_code', 'trade_date'])
|
||
|
||
# daily_market_factors = generate_daily_index_relation_factors(all_indices_df)
|
||
# print("\n每日市场风格/情绪因子:")
|
||
# print(daily_market_factors.tail())
|
||
|
||
# 后续，你可以将 daily_market_factors 与你的个股数据 pdf 按 'trade_date' 合并
|
||
# pdf_with_market_factors = pd.merge(pdf, daily_market_factors, on='trade_date', how='left') |