如何基于向量化方法计算带交易信号的DataFrame的PnL

问题说明 #

给定含价格与开平仓信号的DataFrame，需无循环计算配对交易的每日PnL：

• Entry (short)：做空A、做多B；Entry (long)：做多A、做空B
• Exit (short)/Exit (long)对应平仓对应头寸，信号时序有序，平仓均晚于开仓
• PnL规则：
  • 空头区间：(B.pct_change() - A.pct_change())/2
  • 多头区间：(-B.pct_change() + A.pct_change())/2

开平仓信号样例 #

price_df[price_df["Signal"].str.contains("E")]

              A        B        A/B Z score    Signal
2021-04-19  4926.1  148.2251    2.187012    Entry (short)
2021-05-31  4357.7  159.8565    0.030637    Exit (short)
2022-01-06  4658.9  137.4595    2.139689    Entry (short)
2022-02-07  3624.1  123.2928    -0.018995   Exit (short)

头寸区间样例（开平仓间标记为-1/1） #

A        B        A/B Z score    Signal
2021-04-19  4926.1  148.2251    2.187012    Entry (short)
2021-04-20  4726.9  145.8136    1.880473    -1
2021-04-21  4488.9  140.3698    1.715539    -1
...
2021-05-27  4333.4  157.0219    0.165915    -1
2021-05-28  4440.3  157.4703    0.417588    -1
2021-05-31  4357.7  159.8565    0.030637    Exit (short)

向量化实现步骤 #

1. 生成头寸标记列 #

通过ffill()向前填充，实现开仓到平仓区间内的头寸持续标记：

# 初始化头寸列为0
price_df['position'] = 0

# 映射开平仓信号到头寸值
price_df.loc[price_df['Signal'] == 'Entry (short)', 'position'] = -1
price_df.loc[price_df['Signal'] == 'Exit (short)', 'position'] = 0
price_df.loc[price_df['Signal'] == 'Entry (long)', 'position'] = 1
price_df.loc[price_df['Signal'] == 'Exit (long)', 'position'] = 0

# 向前填充，让开仓后到平仓前保持头寸方向
price_df['position'] = price_df['position'].ffill()

2. 计算每日PnL #

利用向量运算直接计算，无需循环：

# 计算A、B的日收益率
price_df['A_pct'] = price_df['A'].pct_change()
price_df['B_pct'] = price_df['B'].pct_change()

# 根据头寸方向计算每日PnL
# 多头头寸(1)对应(A_pct - B_pct)/2，空头(-1)对应(B_pct - A_pct)/2，无头寸(0)为0
price_df['PnL'] = price_df['position'] * (price_df['A_pct'] - price_df['B_pct']) / 2

3. 可选：处理开仓日NaN与累计盈亏 #

• 开仓当日的pct_change()为NaN，可填充为0：

price_df['PnL'] = price_df['PnL'].fillna(0)

• 计算累计盈亏：

price_df['cum_PnL'] = price_df['PnL'].cumsum()

内容的提问来源于stack exchange，提问作者Anonymous