多因子策略

import numpy as np
import pandas as pd

def winsorize(series, lower=0.01, upper=0.99):
    q_low = series.quantile(lower)
    q_high = series.quantile(upper)
    return series.clip(q_low, q_high)

def standardize(series):
    return (series - series.mean()) / series.std()

def neutralize(factor, market_cap, industry_dummy):
    from sklearn.linear_model import LinearRegression
    X = pd.concat([np.log(market_cap), industry_dummy], axis=1)
    model = LinearRegression().fit(X, factor)
    residual = factor - model.predict(X)
    return standardize(residual)

# 示例：对PE因子进行完整预处理
pe_raw = df['pe_ratio']
pe_clean = winsorize(pe_raw)
pe_std = standardize(pe_clean)
pe_neutral = neutralize(pe_std, df['market_cap'], df[['is_bank', 'is_tech']])

from scipy.stats import spearmanr

def calc_ic(factor, forward_return):
    ic, _ = spearmanr(factor, forward_return)
    return ic

def分层回测(df, factor_col, return_col='forward_1m_return', n_groups=5):
    df['group'] = pd.qcut(df[factor_col], n_groups, labels=False)
    group_returns = df.groupby('group')[return_col].mean()
    return group_returns

# 计算某因子过去12个月的月度IC
ics = []
for month_end in pd.date_range('2020-01', '2020-12', freq='M'):
    snapshot = df[df['date'] == month_end]
    ic = calc_ic(snapshot['factor_value'], snapshot['forward_return'])
    ics.append(ic)

print(f"IC均值: {np.mean(ics):.3f}")
print(f"IR: {np.mean(ics)/np.std(ics):.3f}")

def build_multi_factor_portfolio(df, factor_weights, n_stocks=50):
    df['score'] = 0
    for factor, weight in factor_weights.items():
        df['score'] += df[factor] * weight
    selected = df.nlargest(n_stocks, 'score')
    selected['weight'] = 1.0 / n_stocks
    return selected[['code', 'name', 'score', 'weight']]

weights = {
    'pe_z': -0.20, 'pb_z': -0.10, 'roe_z': 0.25,
    'momentum_3m_z': 0.20, 'volatility_z': -0.15, 'turnover_z': -0.10
}
portfolio = build_multi_factor_portfolio(stock_pool, weights, n_stocks=100)

我有一份A股全市场股票的月度数据，包含PE_TTM、PB、ROE_TTM、毛利率、过去1月/3月/6月收益率、20日波动率、20日平均换手率。
请帮我写Python代码：
1. 计算每个因子的月度IC（与未来1个月收益率的Spearman相关系数）
2. 画出IC的时间序列图和分布直方图
3. 筛选出IC均值>0.03且IR>0.3的因子
4. 检查筛选后因子的两两相关性，剔除高度冗余的因子

请用Python和Backtrader实现一个多因子选股策略：
- 每月初调仓，范围沪深300成分股
- 因子1（价值）：PE_TTM倒数标准化，权重30%
- 因子2（质量）：ROE_TTM标准化，权重40%
- 因子3（动量）：过去60日收益率（排除最近20日），权重30%
- 每个因子做市值和行业中性化
- 选综合得分最高的30只，等权配置
- 回测2018-01-01至2023-12-31
- 输出年化收益、夏普比率、最大回撤

小市值因子在2017年之前表现极好，但2017年后持续失效甚至反转。
请分析：
1. 从市场结构、投资者结构、政策环境解释失效原因
2. 写Python代码检测该因子的结构断裂点
3. 提出自适应因子权重方案：当因子近期IC为负时自动降权

"""多因子选股策略 - 简化版 需要: pip install akshare pandas numpy"""
import akshare as ak
import pandas as pd
import numpy as np

def get_stock_data():
    df = ak.stock_zh_a_spot_em()
    df = df[['代码','名称','最新价','市盈率-动态','市净率',
             '总市值','换手率','涨跌幅']].copy()
    df.columns = ['code','name','price','pe','pb',
                  'market_cap','turnover','change_pct']
    df = df[df['pe'] > 0]
    df = df[df['market_cap'] > 1e8]
    return df

def calc_factors(df):
    df['ep'] = 1 / df['pe']
    df['bp'] = 1 / df['pb']
    df['liq'] = -df['turnover']
    for col in ['ep','bp','liq']:
        df[col] = df[col].clip(df[col].quantile(0.01),
                               df[col].quantile(0.99))
        df[f'{col}_z'] = (df[col]-df[col].mean())/df[col].std()
    return df

def build_portfolio(df, n=50):
    df['score'] = df['ep_z']*0.4 + df['bp_z']*0.4 + df['liq_z']*0.2
    selected = df.nlargest(n, 'score')
    selected['weight'] = 1.0 / n
    return selected

if __name__ == '__main__':
    raw = get_stock_data()
    factored = calc_factors(raw)
    portfolio = build_portfolio(factored, n=50)
    print("=== 多因子组合持仓 ===")
    print(portfolio[['code','name','score','weight']].head(10))
    print(f"\n组合平均PE: {portfolio['pe'].mean():.2f}")
    print(f"组合平均PB: {portfolio['pb'].mean():.2f}")
    print(f"组合平均市值: {portfolio['market_cap'].mean()/1e8:.2f}亿")