如何对因子算法做参数优化？
步骤
Step_1： 设计编写因子算法的类，逐行计算合成DataFrame，返回MultiIndex
Step_2： 股票池与数据读取
Step_3： 实例化因子的类，并获取因子的数值
Step_4： 可视化与计算IC绩效
Step_5： 优化因子的参数
Step_6： 最优绩效结果可视化与保存Excel
Step_1： 设计编写因子算法的类，逐行计算合成DataFrame，返回MultiIndex
- 因子的类：父类， 子类， 继承， 实例化
- 参数与算法设计编写
- 数据格式与合成
- 返回MultiIndex
# -*- coding: utf-8 -*-
import pandas as pd
from fxdayu_alphaman.factor import Factor
class Factor_Volume001(Factor):
c = 3
def calculate_factor_by_column(self, data):
# 逐支股票计算volume001因子
candle_data = data[1].dropna()
if len(candle_data) == 0:
return
high = candle_data["high"]
volume = candle_data["volume"]
adv_s = self.ts_mean(volume, 10)
result = - self.correlation(high, adv_s, self.c) #计算因子值
result.index = candle_data.index
result = pd.DataFrame(result)
result.columns = [data[0],]
return result
def factor_calculator(self, pn_data):
# volume001
factor = map(self.calculate_factor_by_column, pn_data.iteritems())
factor = pd.concat(factor, axis=1)
factor = self.winsorize(factor) #去极值
factor = self.standardize(factor) #标准化
factor = self.factor_df_to_factor_mi(factor) #转化成MuitiIndex格式(相当与stack()方法)
return factor
Step_2： 股票池与数据读取
- 选取沪深300的股票代号，需要元组类型
- 获取因子的Panel数据与价格的DataFrame数据
from datetime import datetime
from fxdayu_data import DataAPI
# 时间设置
start = datetime(2015, 1, 1)
end = datetime(2017, 8, 1, 15)
periods = (1, 5, 10)
# 获取数据
codes = DataAPI.info.codes('hs300')
PN = DataAPI.candle(codes, 'D', ('high','close', 'volume'), start=start, end=end, adjust='after')
prices = PN.minor_xs('close')
print (PN)
print (prices.head())
<class 'pandas.core.panel.Panel'>
Dimensions: 300 (items) x 629 (major_axis) x 3 (minor_axis)
Items axis: 000001.XSHE to 603993.XSHG
Major_axis axis: 2015-01-05 15:00:00 to 2017-08-01 15:00:00
Minor_axis axis: close to volume
000001.XSHE 000002.XSHE 000008.XSHE 000009.XSHE \
datetime
2015-01-05 15:00:00 889.136424 1340.204221 30.280252 NaN
2015-01-06 15:00:00 875.850859 1290.735544 30.462517 NaN
2015-01-07 15:00:00 859.160869 1279.118634 31.161198 71.912931
2015-01-08 15:00:00 830.264765 1221.614928 32.151504 72.015559
2015-01-09 15:00:00 836.990582 1209.029942 32.333768 70.021643
000027.XSHE 000039.XSHE 000060.XSHE 000061.XSHE \
datetime
2015-01-05 15:00:00 62.456065 491.222155 244.407901 285.692765
2015-01-06 15:00:00 62.569829 487.457473 253.067838 273.653467
2015-01-07 15:00:00 65.090124 474.786104 256.338847 272.370415
2015-01-08 15:00:00 62.456065 461.012876 250.032154 268.991713
2015-01-09 15:00:00 60.259558 454.103307 243.466604 262.020468
000063.XSHE 000069.XSHE ... 601933.XSHG \
datetime ...
2015-01-05 15:00:00 213.049186 323.423454 ... 18.016449
2015-01-06 15:00:00 225.373661 310.671992 ... 17.974898
2015-01-07 15:00:00 220.888270 306.537427 ... 17.895951
2015-01-08 15:00:00 224.034945 294.520139 ... 17.638336
2015-01-09 15:00:00 219.094113 290.385574 ... 17.185432
601939.XSHG 601958.XSHG 601985.XSHG 601988.XSHG \
datetime
2015-01-05 15:00:00 8.443892 13.505666 NaN 5.533117
2015-01-06 15:00:00 8.321436 13.165430 NaN 5.708088
2015-01-07 15:00:00 8.089190 13.349342 NaN 5.683707
2015-01-08 15:00:00 7.748563 13.047202 NaN 5.421250
2015-01-09 15:00:00 7.809087 12.851467 NaN 5.596221
601989.XSHG 601998.XSHG 603000.XSHG 603885.XSHG \
datetime
2015-01-05 15:00:00 15.737842 9.711167 42.404787 NaN
2015-01-06 15:00:00 16.123040 9.436754 44.042705 NaN
2015-01-07 15:00:00 15.989058 9.413886 43.324463 NaN
2015-01-08 15:00:00 15.287328 8.973047 43.123029 NaN
2015-01-09 15:00:00 14.803319 9.044191 42.748648 NaN
603993.XSHG
datetime
2015-01-05 15:00:00 3.302398
2015-01-06 15:00:00 3.398134
2015-01-07 15:00:00 3.514912
2015-01-08 15:00:00 3.418123
2015-01-09 15:00:00 3.394978
[5 rows x 300 columns]
Step_3： 实例化因子的类，并获取因子与收益的数据
- 用get_factor方法
- 输入Panel数据
- 用alphalens的utils.get_clean_factor_and_forward_returns获取未来收益
volume001 = Factor_Volume001()
factor = volume001.get_factor(PN)
factor.head()
| factor | ||
|---|---|---|
| date | asset | |
| 2015-01-20 15:00:00 | 000001.XSHE | -0.604585 |
| 000002.XSHE | -0.618333 | |
| 000008.XSHE | -0.632032 | |
| 000027.XSHE | -0.615691 | |
| 000039.XSHE | -0.627686 |
Step_4: 可视化与计算IC绩效
- 用alphalens的plotting可视化IC
- 用calculate_performance计算绩效
from alphalens import utils,performance,plotting
import matplotlib.pyplot as plt
def plot_performance(factor, prices):
factor_data = utils.get_clean_factor_and_forward_returns(factor,prices,quantiles=5,periods=(1,5,10))
ic = performance.factor_information_coefficient(factor_data)
plotting.plot_ic_hist(ic)
plotting.plot_ic_ts(ic)
mean_ic = performance.mean_information_coefficient(factor_data,by_time="M")
plotting.plot_monthly_ic_heatmap(mean_ic)
# 按quantile区分的持股平均收益（减去了总体平均值）
mean_return_by_q = performance.mean_return_by_quantile(factor_data, by_date=True, demeaned=True)[0]
# 按quantile画出累积持有收益
for i in [1, 5, 10]:
plotting.plot_cumulative_returns_by_quantile(mean_return_by_q, period=i)
plt.show()
plot_performance(factor,prices)
from fxdayu_alphaman.factor.admin import Admin
factor_admin = Admin()
original_perf = factor_admin.calculate_performance('Volume001',
factor,
start,
end,
periods=(1,5,10),
quantiles=5,
price=prices)
original_perf.mean_ic
| 0 | |
|---|---|
| 1 | 0.014715 |
| 5 | 0.018427 |
| 10 | 0.015767 |
Step_5： 优化因子的参数，并读取最优参数的因子值与绩效
- 用enumerate_parameter枚举优化参数，返回因子值与参数值的列表
- 用show_factors_performance计算绩效，返回每个参数绩效的对象
- 用rank_performance对绩效排序，输出第一个的绩效与参数
- 用instantiate_factor_and_get_factor_value实例化因子并返回最优因子值
- 用calculate_performance计算获取最优绩效
# 参数优化
para_range_dict = {"c":range(3,11,1)}
# 1.枚举参数空间
factor_value_list,para_dict_list = factor_admin.enumerate_parameter('Factor_Volume001',
para_range_dict,
codes,start,
end,
Factor=volume001,data=PN)
factor_name_list = []
for para_dict in para_dict_list:
factor_name_list.append("Factor_Volume001+" + str(para_dict))
factors_dict = dict(zip(factor_name_list,factor_value_list))
# 2.获得不同参数下计算得出的因子值的表现
performance_list = factor_admin.show_factors_performance(factors_dict,
start,
end,
periods=(1,5,10),
quantiles=5,
price=prices)
# print("#####################################################################################")
# 按绩效指标对结果排序（寻优） 本例按10天持有期的mean_IC降序排列了所有结果。
performance_list = factor_admin.rank_performance(performance_list,
target_period=10,
ascending=False)
# 输出最优因子的ic和名称
print (performance_list[0].mean_ic)
print (performance_list[0].factor_name)
0
1 0.015963
5 0.019791
10 0.026054
Factor_Volume001+{'c': 6}
factor_opt = factor_admin.instantiate_factor_and_get_factor_value('Factor_Volume001',
start=start,
end=end,
pool=codes,
data=PN,
Factor=volume001,
para_dict={'c':6})
perf = factor_admin.calculate_performance('Volume001',
factor_opt,
start,
end,
periods=(1,5,10),
quantiles=5,
price=prices)
perf.mean_ic
| 0 | |
|---|---|
| 1 | 0.015963 |
| 5 | 0.019791 |
| 10 | 0.026054 |
Step_6： 最优绩效结果可视化与保存Excel
plot_performance(factor_opt,prices)
import alphalens
import numpy as np
quantile = alphalens.utils.quantize_factor(factor_opt,quantiles=5)
factor_sheet = quantile[quantile==5].unstack().replace(np.nan, 0).replace(5, 1)
factor_sheet.to_excel('factor_opt.xlsx')