量化投资 — 简单动量策略（Momentum Strategy）

动量策略 - Momentum Strategy

0. 引库

import numpy as np
import pandas as pd
import tushare as ts
import matplotlib.pyplot as plt
import seaborn
plt.style.use('seaborn')      
import matplotlib as mpl
%matplotlib inline
mpl.rcParams['font.family'] = 'serif'
import warnings; warnings.simplefilter('ignore')     # 忽略警告信息

1. 数据准备 & 回测准备

# tushare 获取沪深300股票数据
data = ts.get_k_data('hs300', start = '2010-01-01', end='2017-06-30')[['date','close']]
data.rename(columns={'close': 'price'}, inplace=True)
data.set_index('date', inplace = True)

data.head()

	price
date
2010-01-04	3535.229
2010-01-05	3564.038
2010-01-06	3541.727
2010-01-07	3471.456
2010-01-08	3480.130

2. 策略开发思路

# 计算连续收益率
data['returns'] = np.log(data['price'] / data['price'].shift(1))

data.head()

	price	returns
date
2010-01-04	3535.229	NaN
2010-01-05	3564.038	0.008116
2010-01-06	3541.727	-0.006280
2010-01-07	3471.456	-0.020040
2010-01-08	3480.130	0.002496

# np.sign()符号函数支持向量化
data['position'] = np.sign(data['returns'])     

data.head()

	price	returns	position
date
2010-01-04	3535.229	NaN	NaN
2010-01-05	3564.038	0.008116	1.0
2010-01-06	3541.727	-0.006280	-1.0
2010-01-07	3471.456	-0.020040	-1.0
2010-01-08	3480.130	0.002496	1.0

# 计算Momentum策略收益（注意未来函数问题）
data['strategy'] = data['position'].shift(1) * data['returns']  

data.head(10)

	price	returns	position	strategy
date
2010-01-04	3535.229	NaN	NaN	NaN
2010-01-05	3564.038	0.008116	1.0	NaN
2010-01-06	3541.727	-0.006280	-1.0	-0.006280
2010-01-07	3471.456	-0.020040	-1.0	0.020040
2010-01-08	3480.130	0.002496	1.0	-0.002496
2010-01-11	3482.052	0.000552	1.0	0.000552
2010-01-12	3534.916	0.015068	1.0	0.015068
2010-01-13	3421.144	-0.032715	-1.0	-0.032715
2010-01-14	3469.051	0.013906	1.0	-0.013906
2010-01-15	3482.738	0.003938	1.0	0.003938

3. 策略可视化

# 计算策略最终累计收益
data[['returns', 'strategy']].cumsum().apply(np.exp).plot(figsize=(10, 6))；

4. 策略优化之思路——参数优化和穷举

# 计算5日内平均信号
data['position_5'] = np.sign(data['returns'].rolling(5).mean())
data['strategy_5'] = data['position_5'].shift(1) * data['returns']
data[['returns', 'strategy_5']].dropna().cumsum().apply(np.exp).plot(figsize=(10, 6))

参数寻优——使用离散Return计算方法

data['returns_dis'] = data['price'] / data['price'].shift(1) - 1
# 也可这样计算离散收益率：data['returns_dis'] = data['price'].pct_change()
data['returns_dis_cum'] = (data['returns_dis'] + 1).cumprod()
data.head()

	price	returns	position	strategy	position_5	strategy_5	returns_dis	returns_dis_cum
date
2010-01-04	3535.229	NaN	NaN	NaN	NaN	NaN	NaN	NaN
2010-01-05	3564.038	0.008116	1.0	NaN	NaN	NaN	0.008149	1.008149
2010-01-06	3541.727	-0.006280	-1.0	-0.006280	NaN	NaN	-0.006260	1.001838
2010-01-07	3471.456	-0.020040	-1.0	0.020040	NaN	NaN	-0.019841	0.981961
2010-01-08	3480.130	0.002496	1.0	-0.002496	NaN	NaN	0.002499	0.984414

# 建立一个绘制图形的 list
price_plot = ['returns_dis_cum']    
type(price_plot)

list

# 参数寻优（入门方法）
for days in [10,20,30,60]:
    price_plot.append('sty_cumr_%dd' % days)
    data['position_%dd' % days] = np.where(data['returns'].rolling(days).mean()>0, 1, -1)
    data['strategy_%dd' % days] = data['position_%dd' % days].shift(1) * data['returns']
    data['sty_cumr_%dd' % days] = (data['strategy_%dd' % days] + 1).cumprod()

data.head()

	price	returns	position	strategy	position_5	strategy_5	returns_dis	returns_dis_cum	position_10d	strategy_10d	sty_cumr_10d	position_20d	strategy_20d	sty_cumr_20d	position_30d	strategy_30d	sty_cumr_30d	position_60d	strategy_60d	sty_cumr_60d
date
2010-01-04	3535.229	NaN	NaN	NaN	NaN	NaN	NaN	NaN	-1	NaN	NaN	-1	NaN	NaN	-1	NaN	NaN	-1	NaN	NaN
2010-01-05	3564.038	0.008116	1.0	NaN	NaN	NaN	0.008149	1.008149	-1	-0.008116	0.991884	-1	-0.008116	0.991884	-1	-0.008116	0.991884	-1	-0.008116	0.991884
2010-01-06	3541.727	-0.006280	-1.0	-0.006280	NaN	NaN	-0.006260	1.001838	-1	0.006280	0.998113	-1	0.006280	0.998113	-1	0.006280	0.998113	-1	0.006280	0.998113
2010-01-07	3471.456	-0.020040	-1.0	0.020040	NaN	NaN	-0.019841	0.981961	-1	0.020040	1.018115	-1	0.020040	1.018115	-1	0.020040	1.018115	-1	0.020040	1.018115
2010-01-08	3480.130	0.002496	1.0	-0.002496	NaN	NaN	0.002499	0.984414	-1	-0.002496	1.015574	-1	-0.002496	1.015574	-1	-0.002496	1.015574	-1	-0.002496	1.015574

# 查看绘制图形的 list
price_plot

['returns_dis_cum',
 'sty_cumr_10d',
 'sty_cumr_20d',
 'sty_cumr_30d',
 'sty_cumr_60d']

data[price_plot].dropna().plot(
    title='HS300 Multi Parameters Momuntum Strategy',
    figsize=(10, 6), style=['--', '--', '--', '--','--'])

5. 策略优化思路之—— High Frequency Data用于Momentum策略

hs300_hf = ts.get_k_data('hs300', ktype='5')

hs300_hf.head(10)

	date	open	close	high	low	volume	amount	turnoverratio	code
0	2019-03-28 14:55	3724.82	3725.07	3725.68	3724.42	3806742.0	{}	0.0000	hs300
1	2019-03-28 15:00	3725.07	3728.40	3728.40	3724.90	2450800.0	{}	0.0000	hs300
2	2019-03-29 09:35	3739.77	3754.43	3754.51	3739.77	10164739.0	{}	0.0000	hs300
3	2019-03-29 09:40	3754.44	3763.63	3763.94	3748.90	6341654.0	{}	0.0000	hs300
4	2019-03-29 09:45	3764.66	3763.26	3764.84	3759.35	5622416.0	{}	0.0000	hs300
5	2019-03-29 09:50	3763.44	3762.66	3763.89	3754.12	5955572.0	{}	0.0000	hs300
6	2019-03-29 09:55	3762.87	3766.84	3769.46	3762.87	4600542.0	{}	0.0000	hs300
7	2019-03-29 10:00	3766.66	3758.50	3766.66	3757.00	5533989.0	{}	0.0000	hs300
8	2019-03-29 10:05	3758.20	3765.39	3765.39	3758.20	3587029.0	{}	0.0000	hs300
9	2019-03-29 10:10	3766.19	3775.06	3775.15	3766.19	3993704.0	{}	0.0000	hs300

hs300_hf.set_index('date',inplace = True)
hs300_hf.index = hs300_hf.index.to_datetime()
hs300_hf.info()          

<class 'pandas.core.frame.DataFrame'>
DatetimeIndex: 482 entries, 2019-03-28 14:55:00 to 2019-04-12 15:00:00
Data columns (total 8 columns):
open             482 non-null float64
close            482 non-null float64
high             482 non-null float64
low              482 non-null float64
volume           482 non-null float64
amount           482 non-null object
turnoverratio    482 non-null object
code             482 non-null object
dtypes: float64(5), object(3)
memory usage: 33.9+ KB

# 如果自己实现，请注意具体日期范围，否则可能没有数据
hs300_hf['2019-04-01':'2019-04-02'].head()

	open	close	high	low	volume	amount	turnoverratio	code
2019-04-01 09:35:00	3901.17	3923.72	3929.06	3901.17	25618175.0	{}	0.0000	hs300
2019-04-01 09:40:00	3923.72	3952.36	3952.74	3921.96	14481548.0	{}	0.0000	hs300
2019-04-01 09:45:00	3952.36	3952.94	3955.06	3951.27	13020862.0	{}	0.0000	hs300
2019-04-01 09:50:00	3952.94	3965.55	3967.12	3952.94	11968217.0	{}	0.0000	hs300
2019-04-01 09:55:00	3965.55	3953.47	3965.55	3951.80	9892385.0	{}	0.0000	hs300

hs300_hf['returns'] = np.log(hs300_hf['close'] / hs300_hf['close'].shift(1))
hs300_hf['position'] = np.sign(hs300_hf['returns'].rolling(10).mean()) 
hs300_hf['strategy'] = hs300_hf['position'].shift(1) * hs300_hf['returns']

# 画图
hs300_hf[['returns', 'strategy']].dropna().cumsum().apply(np.exp).plot(figsize=(10, 6), style=['--', '--'])