信号研究

min_data数据下载：

链接：https://pan.baidu.com/s/1OBTX-zc2GfsGutlEeoznJg
提取码：uz91

1. 获取数据

清洗数据

import warnings
warnings.filterwarnings('ignore')
from jaqs_fxdayu.data.dataservice import LocalDataService
ds = LocalDataService()
from time import time
## 加freq参数
start = time()
path = './min_data/VnTrader_1Min_Db'
props = {'fields': 'open,high,low,close,volume','symbol': 
         'EOSUSDT:binance',
         'freq': '15Min,25Min',
         'start_date':20180601000000}
Time_dict = ds.bar_reader(path,props) #读取数据
print(time()-start)

6.14521598815918

df = Time_dict['15Min'].dropna().set_index("datetime")

画图看数据是否缺失

import matplotlib.pyplot as plt
plt.figure(figsize=(20,10))
plt.plot(df.close)
plt.show()

2. 信号研究

提出假设：借鉴经典的海龟策略思想，当价格突破一定周期的最高价则看多，反之看空

编写信号：

（1）读取数据，计算16个k线周期的最高最低价格指标

import pandas as pd
import numpy as np
import os
import talib as ta
import matplotlib.pyplot as plt
df1 = Time_dict['15Min'].dropna().set_index("datetime")

df1['upband'] = ta.MAX(df1['close'].values,16)
df1['downband'] = ta.MIN(df1['close'].values,16)

（2）将指标往前推一个k线，产生突破信号

df1['upband'] = df1['upband'].shift(1)
df1['downband'] = df1['downband'].shift(1)

（3）可视化指标

fig, ax1 = plt.subplots(figsize=(15, 7))
plt.plot(df1[['close','upband','downband']].loc['20180901':'20180911',])
plt.grid(True)
plt.axis('tight')
plt.xlabel('date')
plt.ylabel('Price 1nd')
plt.title('two stock')
plt.show()

（4）编写信号，若价格突破16周期最高价则看多，siganl为1，突破最低价则看空，signal为-1，若在两轨中间，则为0

可视化数据与信号

df1['signal1'] = np.where(df1['close']>df1['upband'],1,np.where(df1['close']<df1['downband'],-1,0))

fig, ax1 = plt.subplots(figsize=(15, 7))
plt.plot(df1[['close']].loc['20180901':'20180911',])
plt.grid(True)
plt.axis('tight')
plt.xlabel('date')
plt.ylabel('Price 1nd')
plt.title('two stock')
ax2 = ax1.twinx()  #双坐标
plt.plot(df1[['signal1']].loc['20180901':'20180911',],'r')
# plt.plot(df1[['signal6']].loc['20180801':'20180811',],'g')
plt.ylabel('Price 2nd')
plt.show()

发现问题：发现虚假信号过多，特别是在震荡的行情中开仓次数太频繁

解决方案：加入ATR波动率指标过滤信号

df1['atr20'] = ta.ATR(df1['high'].values,df1['low'].values,df1['close'].values,14)
df1['atr50'] = ta.ATR(df1['high'].values,df1['low'].values,df1['close'].values,76)

当14周期atr > 76周期atr，允许开仓，signal2 为 1，否则signal2 为 0 ，不允许开仓。将两个指标结合，得出signal3信号

df1['signal2'] = np.where(df1['atr20']>=df1['atr50'],1,0)
df1['signal3'] = np.where(df1['signal2']==1,df1['signal1'],0)

加入atr信号之后可以看到过滤掉了很多震荡行情的假信号

fig, ax1 = plt.subplots(figsize=(15, 7))
plt.plot(df1[['close','upband','downband']].loc['20180901':'20180911',])
plt.grid(True)
plt.axis('tight')
plt.xlabel('date')
plt.ylabel('Price 1nd')
plt.title('two stock')
ax2 = ax1.twinx()  #双坐标
plt.plot(df1[['signal3']].loc['20180901':'20180911',],'r')
# plt.plot(df1[['signal2']].loc['20180901':'20180911',],'g')
plt.ylabel('Price 2nd')
plt.show()

思路验证

通过vnpy回测策略思路得到回测绩效，验证策略可行性