@Channelchan
2018-01-11T11:50:50.000000Z
字数 14429
阅读 24040
参数优化器
- 优化IC
- 优化Ann RET
- 优化IR
1_优化IC
- 读取本地相应数据
- 设置Optimizer()
- 优化IC启动枚举enumerate_optimizer()
- 获取最优结果的signal_data与performance
1. 读取本地相应数据
import numpy as npfrom jaqs.data import DataViewfrom jaqs.research import SignalDiggerimport warningswarnings.filterwarnings("ignore")dataview_folder = 'JAQS_Data/hs300'dv = DataView()dv.load_dataview(dataview_folder)
D:\Anaconda3\lib\site-packages\statsmodels\compat\pandas.py:56: FutureWarning: The pandas.core.datetools module is deprecated and will be removed in a future version. Please use the pandas.tseries module instead.
from pandas.core import datetools
Dataview loaded successfully.
def mask_index_member():df_index_member = dv.get_ts('index_member')mask_index_member = df_index_member ==0 #定义信号过滤条件-非指数成分return mask_index_memberdef limit_up_down():# 定义可买卖条件——未停牌、未涨跌停trade_status = dv.get_ts('trade_status')mask_sus = trade_status == u'停牌'# 涨停dv.add_formula('up_limit', '(close - Delay(close, 1)) / Delay(close, 1) > 0.095', is_quarterly=False)# 跌停dv.add_formula('down_limit', '(close - Delay(close, 1)) / Delay(close, 1) < -0.095', is_quarterly=False)can_enter = np.logical_and(dv.get_ts('up_limit') < 1, ~mask_sus) # 未涨停未停牌can_exit = np.logical_and(dv.get_ts('down_limit') < 1, ~mask_sus) # 未跌停未停牌return can_enter,can_exit
mask = mask_index_member()can_enter,can_exit = limit_up_down()price = dv.get_ts('close_adj')price_bench = dv.data_benchmark
Add formula failed: name [up_limit] exist. Try another name.
Add formula failed: name [down_limit] exist. Try another name.
2.初始化Optimizer()
'''
:param dataview: 包含了计算公式所需要的所有数据的jaqs.data.DataView对象
:param formula: str 需要优化的公式：如'(open - Delay(close, l1)) / Delay(close, l2)'
:param params: dict 需要优化的参数范围：如{"LEN1"：range(1,10,1),"LEN2":range(1,10,1)}
:param name: str (N) 信号的名称
:param price: dataFrame (N) 价格与ret不能同时存在
:param ret: dataFrame (N) 收益
:param high: dataFrame (N) 最高价 用于计算上行收益空间
:param low: dataFrame (N) 最低价 用于计算下行收益空间
:param benchmark_price: dataFrame (N) 基准价格 若不为空收益计算模式为相对benchmark的收益
:param period: int (5) 选股持有期
:param n_quantiles: int (5)
:param mask: 过滤条件 dataFrame (N)
:param can_enter: dataFrame (N) 是否能进场
:param can_exit: dataFrame (N) 是否能出场
:param forward: bool(True) 是否forward return
:param commission: float(0.0008) 手续费率
:param is_event: bool(False) 是否是事件(0/1因子)
:param is_quarterly: bool(False) 是否是季度因子
'''
# 优化因子from jaqs.research import Optimizeralgorithm_opt = Optimizer(dataview=dv,formula='- Correlation(vwap_adj, volume, LEN)',params={"LEN":range(2,15,1)},name='Cor_Vwap_Volume',price=price,benchmark_price=None,#=None求绝对收益 #=price_bench求相对收益period=5,n_quantiles=5,mask=mask,can_enter=can_enter,can_exit=can_exit,commission=0.0008,#手续费 默认0.0008is_event=False,#是否是事件(0/1因子)is_quarterly=False)#是否是季度因子 默认为False
3. 枚举优化enumerate_optimizer()
'''
:param target_type: 目标种类t/top_quantile_ret/bottom_quantile_ret/tmb_ret
:param target: 优化目标
:param ascending: bool(False)升序or降序排列
:param in_sample_range: [date_start(int),date_end(int)] (N) 定义样本内优化范围.
'''
target_type:
- ic类:
return_ic/upside_ret_ic/downside_ret_ic - 持有收益类
long_ret/short_ret/long_short_ret/top_quantile_ret/bottom_quantile_ret/tmb_ret - 收益空间类
long_space/short_space/long_short_space/top_quantile_space/bottom_quantile_space/tmb_space
target:
- ic类
"IC Mean", "IC Std.", "t-stat(IC)", "p-value(IC)", "IC Skew", "IC Kurtosis", "Ann. IR" - 持有收益类
't-stat', "p-value", "skewness", "kurtosis", "Ann. Ret", "Ann. Vol", "Ann. IR", "occurance" - 收益空间类的'upside_space_mean','upside_space_std','upside_space_mean/std','upside_space_max','upside_space_min','upside_space_percentile25','upside_space_percentile50','upside_space_percentile75','upside_space_occurance','downside_space_mean','downside_space_std','downside_space_mean/std','downside_space_max',downside_space_min','downside_space_percentile25','downside_space_percentile50','downside_space_percentile75','downside_space_occurance','up&down_space_mean_sum',
algorithm_ic_opt = algorithm_opt.enumerate_optimizer(target_type="return_ic",target="IC Mean",in_sample_range=[20170101,20170701],ascending=False)
Nan Data Count (should be zero) : 0; Percentage of effective data: 52%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
4. 获取最优结果的signal_data与performance
algorithm_data_opt = algorithm_opt.all_signals.get(algorithm_ic_opt[0]["signal_name"])
print(algorithm_data_opt.tail())
signal return quantile
trade_date symbol
20171222 603160.SH -0.482838 0.0 2
603799.SH -0.741520 0.0 1
603833.SH 0.262818 0.0 5
603858.SH -0.375411 0.0 3
603993.SH -0.490465 0.0 2
print(algorithm_ic_opt[0]["signal_name"])print(algorithm_ic_opt[0]["ret"])print(algorithm_ic_opt[0]["ic"])
Cor_Vwap_Volume{'LEN': 13}
long_ret short_ret long_short_ret top_quantile_ret \
t-stat 2.501894 0.664812 1.771286 6.961844
p-value 0.013720 0.507470 0.079090 0.000000
skewness -0.210919 0.413379 0.350763 -0.162317
kurtosis -0.322710 -0.189578 0.084759 7.356145
Ann. Ret 0.152959 0.050741 0.099284 0.148084
Ann. Vol 0.095461 0.119172 0.087521 0.250003
Ann. IR 1.602325 0.425775 1.134411 0.592330
occurance 119.000000 119.000000 119.000000 6687.000000
bottom_quantile_ret tmb_ret
t-stat -4.380417 4.989744
p-value 0.000010 0.000000
skewness 1.159265 0.422506
kurtosis 18.815342 1.235578
Ann. Ret -0.109962 0.259470
Ann. Vol 0.297067 0.081195
Ann. IR -0.370158 3.195655
occurance 6779.000000 119.000000
return_ic
IC Mean 6.290123e-02
IC Std. 9.922008e-02
t-stat(IC) 6.915651e+00
p-value(IC) 2.558899e-10
IC Skew -2.534709e-02
IC Kurtosis -4.108898e-01
Ann. IR 6.339567e-01
import matplotlib.pyplot as pltdef signal_obj(signal, name, period, quantile):price = dv.get_ts('close_adj')price_bench = dv.data_benchmarkobj = SignalDigger(output_folder="hs300/%s" % name,output_format='pdf')obj.process_signal_before_analysis(signal,price=price,n_quantiles=quantile, period=period,benchmark_price=price_bench,can_enter = can_enter,can_exit = can_exit,mask=mask)obj.create_full_report()return objdef signal_plot(signal, name, period=5, quantile=5):obj = signal_obj(signal, name, period, quantile)obj.fig_objsplt.show()def signal_data(signal, name, period=5, quantile=5):obj = signal_obj(signal, name, period, quantile)return obj.signal_data
dv.add_formula('Cor_Vwap_Volume','-Correlation(vwap_adj, volume, 8)', is_quarterly=False)
signal_plot(dv.get_ts('Cor_Vwap_Volume'), 'Cor_Vwap_Volume')
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Value of signals of Different Quantiles Statistics
min max mean std count count %
quantile
1 -14.862705 0.374615 -0.815487 0.145372 83265 20.144579
2 -0.948111 0.688406 -0.609548 0.176436 82670 20.000629
3 -0.910831 0.816059 -0.413336 0.227705 82649 19.995548
4 -0.857445 0.906101 -0.166788 0.266489 82670 20.000629
5 -0.762486 0.995462 0.276271 0.315837 82083 19.858614
Figure saved: C:\Users\small\OneDrive\notebook\Internet_Course\4_Selection\JAQS_Plus\hs300\Cor_Vwap_Volume\returns_report.pdf
Information Analysis
ic
IC Mean 0.040
IC Std. 0.132
t-stat(IC) 11.396
p-value(IC) 0.000
IC Skew -0.018
IC Kurtosis 0.502
Ann. IR 0.300
Figure saved: C:\Users\small\OneDrive\notebook\Internet_Course\4_Selection\JAQS_Plus\hs300\Cor_Vwap_Volume\information_report.pdf
2_优化Ann. Ret
外部自定义算法
- 设置Optimizer()
- 生成字典并输入自定义算法计算出来的signal
- 将all_signals赋值为signal字典
- 优化Ann. Ret启动枚举enumerate_optimizer()
- 将绩效图形化
1. 设置Optimizer()
2. 生成字典并输入自定义算法计算出来的signal
import pandas as pdimport talib as taTA_Signal = Optimizer(dataview=dv,period=5,n_quantiles=5,mask=mask,can_enter=can_enter,can_exit=can_exit,commission=0.0008,#手续费 默认0.0008is_event=False,#是否是事件(0/1因子)is_quarterly=False)#是否是季度因子 默认为Falseclose = dv.get_ts("close_adj")signals_dict = {}for param in range(2,9,1):slope_df = pd.DataFrame({name: -ta.LINEARREG_SLOPE(value.values, param) for name, value in close.iteritems()}, index=close.index)slope_df.columns.name='symbol'dv.append_df(slope_df,'SLOPE_PARAM='+str(param))signals_dict['SLOPE_PARAM='+str(param)] = TA_Signal.cal_signal(dv.get_ts('SLOPE_PARAM='+str(param)))
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
**3. 将all_signals赋值为signal字典 **
TA_Signal.all_signals = signals_dict
4. 优化Ann. Ret启动枚举enumerate_optimizer()
ret_best = TA_Signal.enumerate_optimizer(target_type="long_ret",#优化目标类型target = "Ann. Ret",in_sample_range=None,ascending=False)
print(ret_best[0]["signal_name"],',',ret_best[0]["ret"].loc['Ann. Ret', 'long_ret'])print(ret_best[1]["signal_name"],',',ret_best[1]["ret"].loc['Ann. Ret', 'long_ret'])
SLOPE_PARAM=6 , 0.0592788850637
SLOPE_PARAM=7 , 0.0526284491294
signal_plot(dv.get_ts(ret_best[0]["signal_name"]), 'slope_opt')
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Value of signals of Different Quantiles Statistics
min max mean std count count %
quantile
1 -27.756178 0.636693 -0.531595 0.945832 82996 20.143487
2 -1.979614 1.055930 -0.074165 0.162311 82402 19.999320
3 -0.906883 1.688512 0.006881 0.137509 82395 19.997621
4 -0.527076 3.048853 0.083919 0.205871 82402 19.999320
5 -0.304564 31.619752 0.469493 0.961335 81829 19.860251
Figure saved: C:\Users\small\OneDrive\notebook\Internet_Course\4_Selection\JAQS_Plus\hs300\slope_opt\returns_report.pdf
Information Analysis
ic
IC Mean 0.052
IC Std. 0.168
t-stat(IC) 11.706
p-value(IC) 0.000
IC Skew -0.047
IC Kurtosis 0.088
Ann. IR 0.308
Figure saved: C:\Users\small\OneDrive\notebook\Internet_Course\4_Selection\JAQS_Plus\hs300\slope_opt\information_report.pdf
3_优化Ann. IR
均线事件驱动
- 设置事件驱动的Optimizer
- 优化Ann. IR启动枚举enumerate_optimizer()
- 将绩效图形化
1. 设置事件驱动的Optimizer
event_opt = Optimizer(dataview=dv,formula='(Ts_Mean(close_adj, SHORT)>=Ts_Mean(close_adj, LONG))&&(Delay(Ts_Mean(close_adj, SHORT)<Ts_Mean(close_adj, LONG), DELAY))',params={'SHORT':range(2,6,1),'LONG':range(30,50,5),'DELAY':range(1,3,1)},name='Cross',price=price,benchmark_price=None,#=None求绝对收益 #=price_bench求相对收益period=5,n_quantiles=5,mask=mask,can_enter=can_enter,can_exit=can_exit,commission=0.0008,#手续费 默认0.0008is_event=True,#是否是事件(0/1因子)is_quarterly=False)#是否是季度因子 默认为False
2. 优化IR启动枚举enumerate_optimizer()
ir_best = event_opt.enumerate_optimizer(target_type="long_ret",target="Ann. IR",in_sample_range=[20160105,20170701],ascending=False)
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 52%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 52%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 52%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 52%
3. 将绩效图形化
print(ir_best[0]["signal_name"])print(ir_best[0]["ret"])
Cross{'SHORT': 5, 'LONG': 45, 'DELAY': 2}
long_ret long_short_ret
t-stat 4.320175 0.001870
p-value 0.000020 0.998510
skewness 1.066033 -0.697733
kurtosis 8.095972 6.272898
Ann. Ret 0.146586 0.000174
Ann. Vol 0.295865 0.243576
Ann. IR 0.495450 0.000714
occurance 3681.000000 333.000000
import matplotlib.pyplot as pltdef event_obj(signal, name, period):price = dv.get_ts('close_adj').loc[20160105:]# price_bench = dv.data_benchmark.loc[20160105:]# Step.4 analyze!obj = SignalDigger()df_all, df_events, df_stats = obj.create_binary_event_report(signal.loc[20160105:], price, mask.loc[20160105:], can_enter.loc[20160105:], can_exit.loc[20160105:],periods=[period,period*2, period*3])res = obj.create_full_report()print(df_stats)return objdef event_plot(signal, name, period=5):obj = event_obj(signal, name, period)obj.fig_objsplt.show()def event_data(signal, name, period=5):obj = event_obj(signal, name, period)return obj.signal_data
event_data_opt = event_opt.all_signals.get(ir_best[0]["signal_name"])
dv.add_formula('Cross','(Ts_Mean(close_adj,5)>=Ts_Mean(close_adj,45))&&(Delay(Ts_Mean(close_adj, 5)<Ts_Mean(close_adj,45),2))', is_quarterly=False)
event_plot(dv.get_ts('Cross'),'Cross')
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Nan Data Count (should be zero) : 0; Percentage of effective data: 53%
Calendar Distribution (8329 occurance from 2016-01-05 to 2017-12-22):
Figure saved: C:\Users\small\OneDrive\notebook\Internet_Course\4_Selection\JAQS_Plus\event_report.pdf
Value of signals of Different Quantiles Statistics
min max mean std count count %
quantile
1 0.0 1.0 0.034396 0.182245 136469 100.0
Figure saved: C:\Users\small\OneDrive\notebook\Internet_Course\4_Selection\JAQS_Plus\returns_report.pdf
Information Analysis
ic
IC Mean 0.002
IC Std. 0.073
t-stat(IC) 0.447
p-value(IC) 0.655
IC Skew -0.261
IC Kurtosis 0.030
Ann. IR 0.021
Figure saved: C:\Users\small\OneDrive\notebook\Internet_Course\4_Selection\JAQS_Plus\information_report.pdf
Annu. Ret. Annu. Vol. t-stat p-value skewness \
trade_date Period
all_sample 5 0.143463 0.290879 4.857081 0.0 1.008173
10 0.212185 0.270920 10.907838 0.0 0.863981
15 0.168983 0.269880 10.680238 0.0 0.625013
kurtosis occurance Annu. Ret. (all samp) \
trade_date Period
all_sample 5 7.176076 4694 0.019249
10 5.198268 4694 0.052127
15 2.938994 4694 0.066563
Annu. Vol. (all samp)
trade_date Period
all_sample 5 0.310950
10 0.293830
15 0.288476
