债券信用利差套利策略全解析：从理论到实战代码实现

# 计算信用利差历史分位
def calc_spread_percentile(df, window=252):
    df['spread'] = df['corp_yield'] - df['treasury_yield']
    df['spread_rank'] = df['spread'].rolling(window).apply(
        lambda x: (x[-1] - x.min()) / (x.max() - x.min())
    )
    return df

import pandas as pd
import tushare as ts

# 获取企业债与国债数据
def fetch_bond_data(bond_code):
    pro = ts.pro_api()
    df = pro.bond_daily(ts_code=bond_code, 
                       start_date='20200101', 
                       end_date='20241231',
                       fields='trade_date,close,yield')
    return df.set_index('trade_date')

corp_bond = fetch_bond_data('1380001.IB')  # 某城投债
treasury_bond = fetch_bond_data('019627.IB') # 10年期国债
merged_df = pd.merge(corp_bond, treasury_bond, 
                    left_index=True, right_index=True,
                    suffixes=('_corp', '_treasury'))

# 动态阈值策略
def generate_signals(df):
    df = calc_spread_percentile(df)
    
    # 做空信号（利差过高）
    df['short_signal'] = (df['spread_rank'] > 0.8) & \
                        (df['spread'].diff() > 0)
    
    # 做多信号（利差过低）  
    df['long_signal'] = (df['spread_rank'] < 0.2) & \
                       (df['spread'].diff() < 0)
    
    return df

# 多空组合构建
def execute_trades(df, capital=1e6):
    df['position'] = 0
    
    # 做空利差（卖企业债/买国债）
    df.loc[df['short_signal'], 'position'] = -1
    
    # 做多利差（买企业债/卖国债）
    df.loc[df['long_signal'], 'position'] = 1
    
    # 交易成本计算（双边0.02%）
    trade_dates = df[df['position'].diff() != 0].index
    df['cost'] = capital * 0.0002 * (len(trade_dates)//2)
    
    return df

def dynamic_stoploss(df):
    # 波动率止损（20日ATR）
    df['atr'] = df['spread'].rolling(20).apply(
        lambda x: max(x) - min(x)
    )
    df['stoploss'] = df['spread'] + np.where(
        df['position'] == 1, -2*df['atr'], 2*df['atr']
    )
    
    # 触发止损
    df['position'] = np.where(
        (df['position'] == 1) & (df['spread'] > df['stoploss']), 0,
        np.where((df['position'] == -1) & (df['spread'] < df['stoploss']), 0, df['position'])
    )
    return df

# 基于凯利公式的仓位计算
def kelly_position(df):
    win_rate = 0.6  # 历史胜率
    win_loss_ratio = 2.5  # 盈亏比
    f = (win_rate * win_loss_ratio - (1 - win_rate)) / win_loss_ratio
    df['position'] = df['position'] * f
    return df

from sklearn.ensemble import GradientBoostingRegressor

# 特征工程
features = ['spread', 'treasury_yield', 'corp_rating', 
           'industry_risk', 'liquidity_ratio']
target = 'spread_1m'  # 未来1月利差变化

# GBDT模型训练
model = GradientBoostingRegressor()
model.fit(X_train[features], y_train)
df['pred_spread'] = model.predict(df[features])

# 跨境利差套利（中资美元债）
def cross_border_spread(cny_bond, usd_bond, fx_rate):
    spread = (cny_bond.yield - usd_bond.yield) - fx_rate.forward_rate
    return spread

# 舆情监控系统
from transformers import pipeline
classifier = pipeline("text-classification", model="finiteautomata/bertweet-base-sentiment-analysis")

def news_sentiment(news_text):
    result = classifier(news_text)
    return result[0]['score']