A Rolling Standardized-Spread Mean-Reversion Strategy for Two Futures
Summary
This script describes a mean-reversion strategy for two related futures contracts. It collects daily closes over a rolling window, standardizes each contract’s prices separately, and defines the spread as the difference between those standardized series. It opens a long spread when the spread falls sufficiently below its rolling mean, and a short spread when it rises sufficiently above it. Positions pair opposite directions in the two contracts, with the second leg’s quantity based on the current price ratio.
The strategy exits when the spread approaches its mean, a maximum holding period is reached, or a loss threshold is breached. It also tracks trade counts, wins, profit, and account balances during a historical simulation. The document provides code and parameter settings but no reported performance analysis. Its rolling normalization, changing position ratio, fixed contract sizing, and simplified profit calculation may affect how closely the signals and accounting represent an economically hedged pair; no evidence here establishes profitability or robustness.
Key ideas
- The strategy trades the difference between separately standardized prices of two futures contracts.
- It enters long or short spreads when the current spread moves beyond a threshold from its rolling mean.
- The paired legs use opposite directions, with the second leg sized from the current price ratio.
- Exit conditions include mean reversion, a holding time limit, and a loss threshold.
- The document supplies a simulation script but no results establishing performance.
Tags
Full text
# distance_based
# distance_based
## Source (Apache-2.0)
```python
#!/usr/bin/env python
# -*- coding: utf-8 -*-
__author__ = 'Chaos'
from tqsdk import TqApi, TqAuth, TargetPosTask, TqBacktest, BacktestFinished
from datetime import date, datetime
import numpy as np
# === 全局参数 ===
SYMBOL1 = "SHFE.rb2305"
SYMBOL2 = "SHFE.hc2305"
WINDOW = 30
K_THRESHOLD = 2.0
CLOSE_THRESHOLD = 0.5
MAX_HOLD_DAYS = 10
STOP_LOSS_PCT = 0.05
POSITION_LOTS1 = 200 # 合约1固定手数
POSITION_RATIO = 1.0 # 合约2与合约1的数量比例
# === 全局变量 ===
price_data1, price_data2 = [], []
position_long = False
position_short = False
entry_price1 = 0
entry_price2 = 0
position_time = None
trade_ratio = 1
entry_spread = 0
trade_count = 0
win_count = 0
total_profit = 0
# === API初始化 ===
api = TqApi(backtest=TqBacktest(start_dt=date(2023, 2, 1),end_dt=date(2023, 4, 27)),
auth=TqAuth("快期账户", "快期密码")
)
quote1 = api.get_quote(SYMBOL1)
quote2 = api.get_quote(SYMBOL2)
klines1 = api.get_kline_serial(SYMBOL1, 24*60*60)
klines2 = api.get_kline_serial(SYMBOL2, 24*60*60)
target_pos1 = TargetPosTask(api, SYMBOL1)
target_pos2 = TargetPosTask(api, SYMBOL2)
print(f"策略开始运行，交易品种: {SYMBOL1} 和 {SYMBOL2}")
try:
while True:
api.wait_update()
if api.is_changing(klines1.iloc[-1], "datetime") or api.is_changing(klines2.iloc[-1], "datetime"):
price_data1.append(klines1.iloc[-1]["close"])
price_data2.append(klines2.iloc[-1]["close"])
if len(price_data1) <= WINDOW:
continue
if len(price_data1) > WINDOW:
price_data1 = price_data1[-WINDOW:]
price_data2 = price_data2[-WINDOW:]
data1 = np.array(price_data1)
data2 = np.array(price_data2)
norm1 = (data1 - np.mean(data1)) / np.std(data1)
norm2 = (data2 - np.mean(data2)) / np.std(data2)
spread = norm1 - norm2
mean_spread = np.mean(spread)
std_spread = np.std(spread)
current_spread = spread[-1]
price_ratio = quote2.last_price / quote1.last_price
trade_ratio = round(price_ratio * POSITION_RATIO, 2)
position_lots2 = int(POSITION_LOTS1 * trade_ratio)
current_time = datetime.fromtimestamp(klines1.iloc[-1]["datetime"] / 1e9)
# === 平仓逻辑 ===
if position_long or position_short:
days_held = (current_time - position_time).days
if position_long:
current_profit = (quote1.last_price - entry_price1) * POSITION_LOTS1 - (quote2.last_price - entry_price2) * position_lots2
else:
current_profit = (entry_price1 - quote1.last_price) * POSITION_LOTS1 - (entry_price2 - quote2.last_price) * position_lots2
profit_pct = current_profit / (entry_price1 * POSITION_LOTS1)
close_by_mean = abs(current_spread - mean_spread) < CLOSE_THRESHOLD * std_spread
close_by_time = days_held >= MAX_HOLD_DAYS
close_by_stop = profit_pct <= -STOP_LOSS_PCT
if close_by_mean or close_by_time or close_by_stop:
target_pos1.set_target_volume(0)
target_pos2.set_target_volume(0)
trade_count += 1
if profit_pct > 0:
win_count += 1
total_profit += current_profit
reason = "均值回归" if close_by_mean else "时间限制" if close_by_time else "止损"
print(f"平仓 - {reason}, 盈亏: {profit_pct:.2%}, 持仓天数: {days_held}")
position_long = False
position_short = False
# === 开仓逻辑 ===
else:
if current_spread < mean_spread - K_THRESHOLD * std_spread:
target_pos1.set_target_volume(POSITION_LOTS1)
target_pos2.set_target_volume(-position_lots2)
position_long = True
position_time = current_time
entry_price1 = quote1.last_price
entry_price2 = quote2.last_price
entry_spread = current_spread
print(f"开仓 - 多价差, 合约1: {POSITION_LOTS1}手, 合约2: {-position_lots2}手, 比例: {trade_ratio}")
elif current_spread > mean_spread + K_THRESHOLD * std_spread:
target_pos1.set_target_volume(-POSITION_LOTS1)
target_pos2.set_target_volume(position_lots2)
position_short = True
position_time = current_time
entry_price1 = quote1.last_price
entry_price2 = quote2.last_price
entry_spread = current_spread
print(f"开仓 - 空价差, 合约1: {-POSITION_LOTS1}手, 合约2: {position_lots2}手, 比例: {trade_ratio}")
# 每日统计
if api.is_changing(klines1.iloc[-1], "datetime"):
account = api.get_account()
print(f"日期: {current_time.date()}, 账户权益: {account.balance:.2f}, 可用资金: {account.available:.2f}")
if trade_count > 0:
print(f"交易统计 - 总交易: {trade_count}, 胜率: {win_count/trade_count:.2%}, 总盈亏: {total_profit:.2f}")
except BacktestFinished as e:
print("回测结束")
api.close()
```Shown in full with attribution under the source's licence. Licence: Apache-2.0
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.