海龟交易法--参数调优(Python)

注意到在上一篇文章的代码实现中，我们设置了一些参数，这些参数也影响了我们实际交易中信号触发的条件和周期。为此，设计了网格调优的程序，寻找能产生最大收益或者控制最大回撤的对应参数。我们希望年均回报率能够大于15%，而希望最大回撤能控制在-20%以内。

首先导入需要的module，并读取所需的外汇数据：

import datetime as dt
import matplotlib.pyplot as plt
plt.rcParams['font.sans-serif']=['Kaiti']
plt.rcParams['axes.unicode_minus']=False
import mplfinance as mpf
import pandas as pd
import numpy as np

import warnings
warnings.filterwarnings("ignore")

df1 = pd.read_csv('eurusd_daily.csv', parse_dates=["Date"], index_col="Date")
df1.drop(['Change'], axis=1, inplace=True)

 将之前写的海龟交易法回测程序主体写成一个自定义函数

def turtle_optimization(n1=30,n2=15,atr_parameter=30,start_date='1990-01-01',end_date='2020-12-31',df=df1):
  
    df['DonHigh_n1'] = df['High'].rolling(n1).max().shift(1)
    df['DonLow_n1'] = df['Low'].rolling(n1).min().shift(1)
    df['DonHigh_n2'] = df['High'].rolling(n2).max().shift(1)
    df['DonLow_n2'] = df['Low'].rolling(n2).min().shift(1)
    df['H-L'] = abs(df['High'] - df['Low'])
    df['H-PC'] = abs(df['High'] - df['Close'].shift(1))
    df['L-PC'] = abs(df['Low'] - df['Close'].shift(1))
    df['TR'] = df[['H-L', 'H-PC', 'L-PC']].max(axis=1, skipna=False)
    df['ATR'] = df['TR'].rolling(atr_parameter).mean()
    df.drop(['H-L','H-PC','L-PC','TR'], axis=1, inplace=True)
    df.dropna(inplace=True)

    st = dt.datetime.strptime(start_date, '%Y-%m-%d')
    en = dt.datetime.strptime(end_date, '%Y-%m-%d')
    df = df[st:en]

    # 仓位和量价的初始化
    position = 0
    df['position'] = np.nan
    last_price = 0
    df['profit'] = np.nan
    volume = 0

    def DCn1upcross(i):
        if df['High'][i] > df['DonHigh_n1'][i] and df['High'][i-1] <= df['DonHigh_n1'][i-1]:
            return True
        else:
            return False

    def DCn1downcross(i):
        if df['Low'][i] < df['DonLow_n1'][i] and df['Low'][i-1] >= df['DonLow_n1'][i-1]:
            return True
        else:
            return False

    def DCn2upcross(i):
        if df['High'][i] > df['DonHigh_n2'][i] and df['High'][i-1] <= df['DonHigh_n2'][i-1]:
            return True
        else:
            return False

    def DCn2downcross(i):
        if df['Low'][i] < df['DonLow_n2'][i] and df['Low'][i-1] >= df['DonLow_n2'][i-1]:
            return True
        else:
            return False

    # 股票仓位：0, 1, -1分别代表不持仓，持多头仓位和持空头仓位
    for i in range(len(df)-1):
        if position == 0:
            if DCn1upcross(i):
                position = 1
                last_price = df['Close'][i]  #入场价格等于当天收盘价，未考虑滑点
                volume = df['ATR'][i]/0.0001*10  #0.0001为外汇的最小基点

            if DCn1downcross(i):
                position = -1
                last_price = df['Close'][i]  #入场价格等于当天收盘价，未考虑滑点
                volume = df['ATR'][i]/0.0001*10  #0.0001为外汇的最小基点

        if position == 1:
            if DCn2downcross(i):
                position = 0
                df['position'][i] = 0  #将多头全部平掉
                df['profit'][i] = (df['Close'][i]-last_price)/0.0001*volume

        if position == -1:
            if DCn2upcross(i):
                position = 0
                df['position'][i] = 0  #将空头全部平掉
                df['profit'][i] = (last_price-df['Close'][i])/0.0001*volume

        if i == len(df)-2:
            if position == 1:
                position = 0
                df['position'][i] = 0  #将多头全部平掉
                df['profit'][i] = (df['Close'][i]-last_price)/0.0001*volume

            if position == -1:
                position = 0
                df['position'][i] = 0  #将空头全部平掉
                df['profit'][i] = (last_price-df['Close'][i])/0.0001*volume

    # 结果计算
    # df['profit'].dropna()
    # df['profit'].dropna().sum()

    account = 1000000  #初始账户的金额（计算回报率用）
    df['return'] = df['profit']/account  #每次操作的回报率
    df['return'].dropna()
    sum = df['return'].sum()
#     print ("总回报率倍数：", round(sum,2))
    
    return round(sum,2)

调用海龟交易法程序，并设置相应的参数为n1=40，n2=20，atr_parameter=30

turtle_optimization(n1=40, n2=20, atr_parameter=30)

此时总体回报率达到：

7.27 (对应727%)

通过for循环，让不同的n1(取从20到40的所有整数)和n2(取从20到70的所有证书)自由组合，计算所有组合情况下的最大回报率。

n1=np.array(range(20,41,1))
n2=np.array(range(20,71,1))

best_return = 0
best_n1 = 0
best_n2 = 0

from tqdm import *

for num1 in tqdm(n1):
    for num2 in n2:
        result = turtle_optimization(num1, num2)
        if result > best_return:
            best_return = result
            best_n1 = num1
            best_n2 = num2

best_return, best_n1, best_n2

得到最大回报率所对应的n1和n2的值分别为：

20, 21

此时总体回报率达到

11.64 (1,164%)

同理，还可以针对n1, n2及atr_parameter等参数一起进行网格调参，也还可以针对最大回撤率进行求解最优参数等。