失眠网 > python量化策略——混合择时策略（动量效应+pe_ttm pb估值+美林时钟）——股债轮动

python量化策略——混合择时策略（动量效应+pe_ttm pb估值+美林时钟）——股债轮动

时间：2021-03-19 13:00:08

将下面三个策略结合判断，

动量策略

估值策略

改进美林时钟

三个策略都判断股票上涨（做多股票，则股：债=0.5:0.5

三个中有两个策略判断做多股票信号，则股：债=0.4:0.6

三个中有一个策略判断做多股票信号，则股：债=0.3:0.7

否则，股：债=0.1:0.9

""" .09.25 15:27@zp数据端，既用到了153数据库，也用到了Tushare Pro数据库显然，加入的条件过多，必定导致过拟合问题，(由于未留出测试集，因此，可以肯定的是过拟合问题存在，但却无法量化验证)"""# coding=utf-8import mathimport tushare as tsimport pandas as pdimport matplotlibimport matplotlib.pyplot as pltimport numpy as npimport talibimport pandas as pdfrom datetime import datetime, dateimport pymysqlimport threadingfrom queue import Queueimport warningswarnings.filterwarnings("ignore")import seaborn as sns sns.set(style="darkgrid", palette="muted", color_codes=True) from scipy import stats,integrate%matplotlib inline sns.set(color_codes=True)matplotlib.rcParams['axes.unicode_minus']=Falseplt.rcParams['font.sans-serif']=['SimHei']ts.set_token(' ')pro = ts.pro_api()#移动均值策略部分def mean_fun():code1,code2,test_start,test_end,t1,t2,t3,t4,t5,n=["000300.SH","000012.SH","0101","20550101",50,70,90,110,130,1]def dataread():df_stock=pro.index_daily(ts_code=code1, start_date=test_start, end_date=test_end, fields='close,trade_date')df_bond=df=pro.index_daily(ts_code=code2,start_date=test_start, end_date=test_end , fields='trade_date,close')return df_stock,df_bonddf_stock,df_bond=dataread() #计算均值，时间为t1 t2 t3 t4def mean(t):df_stock.index=pd.to_datetime(df_stock.trade_date)return df_stock.close.sort_index().rolling(window=t).mean()def ret_base():df_stock.index=pd.to_datetime(df_stock.trade_date)df_bond.index =pd.to_datetime(df_bond.trade_date)ret_stock=(df_stock.close-df_stock.close.shift(-1))/df_stock.close.shift(-1)ret_bond= (df_bond.close- df_bond.close.shift(-1))/df_bond.close.shift(-1)return ret_stock,ret_bond.sort_index()def ret_same_time(x):return x[x.index>=mean(max(t1,t2,t3,t4,t5)).dropna().index[0] ]ret_stock=ret_same_time(ret_base()[0]).sort_index()#ret_base()[0][ret_base()[0].index>=mean(max(t1,t2,t3,t4,t5)).dropna().index[0] ]ret_bond= ret_same_time(ret_base()[1] )#ret_base()[1][ret_base()[1].index>=mean(max(t1,t2,t3,t4,t5)).dropna().index[0] ]DF=ret_same_time(df_stock.close).sort_index()return ret_same_time(mean(t1)),ret_same_time(mean(t2)),ret_same_time(mean(t3)),ret_same_time(mean(t4)),ret_same_time(mean(t5)),DFmean1,mean2,mean3,mean4,mean5,DFF=mean_fun()#估值pe_ttm部分def fun():start,end,code1,code2,t,T,start1,end1=["0101","20550101","000300.SH","000012.SH",10,1400,"0901","20550101"]#读取数据def dataread():df_base=pro.index_dailybasic(ts_code=code1,start_date=start1,end_date=end1, fields='trade_date,pe_ttm,pb')df_stock=pro.index_daily(ts_code=code1, start_date=start, end_date=end, fields='close,trade_date')df_bond=df=pro.index_daily(ts_code=code2,start_date=start, end_date=end , fields='trade_date,close')return df_base,df_stock,df_bonddf,df_stock,df_bond=dataread() class prepare:#计算单日收益率def ret_base(self):df_stock.index=pd.to_datetime(df_stock.trade_date)ret_stock=(df_stock.close-df_stock.close.shift(-1))/df_stock.close.shift(-1)df_bond.index=pd.to_datetime( df_bond.trade_date )ret_bond=(df_bond.close-df_bond.close.shift(-1))/df_bond.close.shift(-1)return ret_stock,ret_bond#计算pe均值，标准差及滚动均值def data_fun(self,freq,df):df.index=pd.to_datetime(df.trade_date )df=df.sort_index()df_std=df.sort_index().rolling(window=T).std()[df.index>=start]#[df.index<=end]df_std=df_std[df_std.index<=end]df_pe_std=df_std.pe_ttmdf_pb_std=df_std.pbdf_mean=df.sort_index().rolling(window=T).mean()[df.index>=start]#[df.index<=end]df_mean=df_mean[df_mean.index<=end]df_pe_mean=df_mean.pe_ttmdf_pb_mean=df_mean.pbdf_roll=df.sort_index().rolling(window=t).mean()[df.index>=start]#[df.index<=end]df_roll=df_roll[df_roll.index<=end]df_pb_roll_mean=df_roll.pbdf_pe_ttm_roll_mean=df_roll.pe_ttmreturn df_pe_mean,df_pb_mean,df_pe_std,df_pb_std,df_pe_ttm_roll_mean,df_pb_roll_meanmean1,mean2,std1,std2,mean_roll1,mean_roll2=prepare().data_fun(t,df)return mean1,mean2,std1,std2,mean_roll1,mean_roll2mean_pe,mean_pb,std1,std2,mean_roll1,mean_roll2=fun()#美林时钟部分class read_data:def fetchData(self, query, q, name, db, host='192.168.0.153', user='jcyj',password='jcyjQwer'):'get data from mysql database'conn = pymysql.connect(host, user, password, database=db, charset='utf8', cursorclass=pymysql.cursors.DictCursor)cursor = conn.cursor()res = cursor.execute(query)fetch = cursor.fetchall()data = pd.DataFrame(fetch)q.put((name, data))def multithread(self, dataList, db):threads = []q = Queue()for key, value in dataList.items():t = threading.Thread(target=self.fetchData, args=(value, q, key, db)) # initiate a threadt.start() # activate a threadthreads.append(t) # collect threads listfor thread in threads:thread.join()result = []for _ in range(len(threads)):result.append(q.get())result = {i[0]: i[1] for i in result}return resultdef readhgyz(self, db='py_daziguan_2_1'):baseQuery = "select report_date , shibor_1w "\" from py_hgyz_pbc_shanghai_shibor_statistics_month where DATE(report_date) > '-12-01' "data = self.multithread({'base': baseQuery},db)return datadef readetl1(self, db='py_etl'):baseQuery = "select trading_date , bond_1y "\" from py_etl_risk_free_rate_daily_2_1 where DATE(trading_date) > '-04-30'"data = self.multithread({'base2': baseQuery},db)return dataBO1=read_data()BO1=BO1.readetl1()["base2"]BO1.set_index('trading_date',inplace=True)BO1.index = pd.to_datetime(BO1.index)BO1.bond_1yR_B=BO1.bond_1y############################读取数据类###################################class readData:def read_index_daily(self,code,star,end):#指数数据dsb = pro.index_daily(ts_code=code, start_date=star, end_date=end,fields='ts_code,trade_date,close,change')#默认读取三个数据return dsbdef read_daily(self,code,star,end):dsc1 = pro.daily(ts_code=code, start_date=star, end_date=end,fields='ts_code,trade_date,close,change')return dsc1def read_CPI(self,star,end):#时间格式start_cpi='09'dc=_ppi(start_m=star, end_m=end,fields='month,ppi_yoy')return dcdef read_GDP(self,star,end):#时间格式star='Q4'df1 = _gdp(start_q=star, end_q=end,fields='quarter,gdp_yoy')return df1def read_bond(self,code,star,end):df=pro.cb_daily(ts_code=code,start_date=star,end_date=end)return dfdef readshibor(self,star,end):return pro.shibor(start_date=star, end_date=end)##################################################################### start_time='20000430'#发布GDP需要时间，我们延迟1个月，即第一季度的GDP4月份才发布。end_time="20550731"star_GDP='Q1'#延后一年，因为我们找的是差分 end_GDP='2055Q1'#不是以3 4季度结尾star_CPI='01'end_CPI='205503'df1=readData()#读取dc=readData()dsc1=readData()dsp=readData()dsb=readData()df1=df1.read_GDP(star_GDP,end_GDP)dc=dc.read_CPI(star_CPI,end_CPI)dsc1=dsc1.read_index_daily('000300.SH',start_time,end_time) dsb=dsb.read_index_daily('000012.SH',start_time,end_time)dsp=dsp.read_index_daily('NHCI.NH',start_time,end_time) ##########################GDP信号处理################################def GDP_fun(df1):df1.set_index('quarter',inplace=True)#设置日期索引df2=(df1.shift(4)-df1).shift(-4)df2=df2.dropna()G=pd.Series(0,index=df2.gdp_yoy.index)for i in range(len(df2.gdp_yoy)):if df2.gdp_yoy[i]>0:G[i]=1elif df2.gdp_yoy[i]<0:G[i]=0else:G[i]=G[i-1]return GG=GDP_fun(df1)###############################################################################################CPI信号函数##############################def CPI_fun(dc):dc=dc.sort_index()dc.set_index('month',inplace=True)dc2=(dc.shift(12)-dc.shift()).shift(-12).dropna()C=pd.Series(0,index=G.index)for j in range( 3*math.floor( len(dc2.ppi_yoy)/3)):if (3+j)%3==0:for i in range(int(j/3),int((3+j)/3)):if dc2.ppi_yoy[j]+dc2.ppi_yoy[j+1]+dc2.ppi_yoy[j+2] >0:C[i]=1elif dc2.ppi_yoy[j]+dc2.ppi_yoy[j+1]+dc2.ppi_yoy[j+2]<0:C[i]=0else:C[i]=C[i-1] else:pass return CC=CPI_fun(dc)######################################################################################################利率处理##################################DF=read_data()DF2=DF.readhgyz()DF=DF2['base']def LV_fun(dc):dc=dc.sort_index()dc.set_index('report_date',inplace=True)dc2=(dc.shift(12)-dc.shift()).shift(-12).dropna()C=pd.Series(0,index=G.index)for j in range( 3*math.floor( len(dc2.shibor_1w)/3)):if (3+j)%3==0:for i in range(int(j/3),int((3+j)/3)):if dc2.shibor_1w[j]+dc2.shibor_1w[j+1]+dc2.shibor_1w[j+2] >0:C[i]=1elif dc2.shibor_1w[j]+dc2.shibor_1w[j+1]+dc2.shibor_1w[j+2]<0:C[i]=0else:C[i]=C[i-1] else:pass return CLV=LV_fun(DF)##########################################################################将季度时间转化成对应的可读的年月日时间。def time_fun(i):if C.index[i][5:6]=='1':time=C.index[i][0:4]+"0505"for j in range(2,10):if len(R_B[R_B.index==time])==0:time=C.index[i][0:4]+"050"+str(j)else:breakif C.index[i][5:6]=='2':time=C.index[i][0:4]+"0801"if len(R_B[R_B.index==time])==0:time=C.index[i][0:4]+"0802"if len(R_B[R_B.index==time])==0:time=C.index[i][0:4]+"0803"if C.index[i][5:6]=='3':time=C.index[i][0:4]+"1101"if len(R_B[R_B.index==time])==0:time=C.index[i][0:4]+"1102"if len(R_B[R_B.index==time])==0:time=C.index[i][0:4]+"1103"if C.index[i][5:6]=='4':time=C.index[i-1][0:4]+"0201"for j in range(2,10):if len(R_B[R_B.index==time])==0:time=C.index[i-1][0:4]+"020"+str(j)else:breakfor j in range(10,29):if len(R_B[R_B.index==time])==0:time=C.index[i-1][0:4]+"02"+str(j)else:breakreturn time############################计算收益率+索引函数##########################class ZC_fun:def ret_function(self,ds):#standard_base=pro.index_daily( ts_code='000300.SH', start_date='1031', end_date="2030")standard_base=ds.sort_index()standard_base.index=pd.to_datetime(standard_base.trade_date,format='%Y-%m-%d')#设置日期索引close_base= standard_base.closestandard_ret=standard_base.change/standard_base.close.shift(-1)ret=standard_ret.sort_index(axis=0,ascending=True)ret=ret.dropna()return ret#,close_base/close_base[start_time_jz].values#=str(RETT.index[0])[:4]+str(RETT.index[0])[5:7]+str(RETT.index[0])[8:10]################有点问题，上面一行def bsc_fun(self,ds):dsc=ds.drop(['ts_code'],axis=1)dsc=dsc.sort_index()dsc.index=pd.to_datetime(dsc.trade_date,format='%Y-%m-%d')#设置日期索引dsc=dsc.drop(['trade_date'],axis=1)return dsc######################################################################### gc=ZC_fun()R_c=gc.ret_function(dsc1)#股票的利gb=ZC_fun()R_b=gb.ret_function(dsb)#债券的利gp=ZC_fun()R_p=gp.ret_function(dsp)#商品的利R_B=pd.DataFrame(R_B,dtype=np.float).bond_1y/100#无风险利率 #######################################对应季度的收益####################################def RET_sig(sig_c,i):#输入ret，输出i对应的时间段收益#####截取收益段if G.index[i][5:6] =='1':a=G.index[i][0:4]+'0505'b=G.index[i][0:4]+'0731'at1=pd.to_datetime(a,format='%Y-%m-%d')bt1=pd.to_datetime(b,format='%Y-%m-%d')sig_c=sig_c[ sig_c.index<=bt1] sig_c=sig_c[ sig_c.index>=at1]elif G.index[i][5:6] =='2':a=G.index[i][0:4]+'0801'b=G.index[i][0:4]+'1031'at1=pd.to_datetime(a,format='%Y-%m-%d')bt1=pd.to_datetime(b,format='%Y-%m-%d')sig_c=sig_c[ sig_c.index<=bt1] sig_c=sig_c[ sig_c.index>=at1]elif G.index[i][5:6] =='3':a= G.index[i][0:4]+'1101'a1=G.index[i][0:4]+'1231'at1=pd.to_datetime(a,format='%Y-%m-%d')bt1=pd.to_datetime(a1,format='%Y-%m-%d')sig_c1=sig_csig_c1=sig_c1[ sig_c1.index<=bt1] sig_c1=sig_c1[ sig_c1.index>=at1]sig_c1=sig_c1b= G.index[i-2][0:4]+'0101'b1=G.index[i-2][0:4]+'0131'at11=pd.to_datetime(b,format='%Y-%m-%d')bt11=pd.to_datetime(b1,format='%Y-%m-%d')sig_c=sig_c[ sig_c.index<=bt11] sig_c=sig_c[ sig_c.index>=at11]sig_c=sig_c1.append(sig_c)else:a=G.index[i-1][0:4]+'0201'b=G.index[i-1][0:4]+'0430'at1=pd.to_datetime(a,format='%Y-%m-%d')bt1=pd.to_datetime(b,format='%Y-%m-%d')sig_c=sig_c[ sig_c.index<=bt1] sig_c=sig_c[ sig_c.index>=at1]return sig_c#######################################################################################################################################################由三个部分构建不同策略#动量策略的信号，sig=1 判断高配股票（判断条件换成bool值形式）def SIG1(i):n=1if DFF[DFF.index== time_fun(i)].values>n*mean1[mean1.index== time_fun(i)].values \and DFF[DFF.index== time_fun(i)].values>n*mean2[mean2.index== time_fun(i)].values \and DFF[DFF.index== time_fun(i)].values>n*mean3[mean3.index== time_fun(i)].values \and DFF[DFF.index== time_fun(i)].values>n*mean4[mean4.index== time_fun(i)].values \and DFF[DFF.index== time_fun(i)].values>n*mean5[mean5.index== time_fun(i)].values :sig=1else:sig=0return sig#基本面—估值（pe_ttm）策略信号， sig=1高配股票def SIG2(i):if mean_roll1[mean_roll1.index== time_fun(i)].values<mean_pe[mean_pe.index== \time_fun(i)].values-0.5*std1[std1.index== time_fun(i)].values or \mean_roll2[mean_roll2.index== time_fun(i)].values<mean_pb[mean_pb.index== \time_fun(i)].values-0.5*std2[std2.index== time_fun(i)].values:sig=1else:sig=0return sig #美林时钟高配股票信号def SIG3(i):if C[i]==0 and LV[i]==0:sig=1else:sig=0return sig #################################策略判断条件##################################def quanzhong_fun(R_c,R_p,R_b,R_B):RETT=None#初始化n=1for i in range(len(G)):if SIG1(i)+SIG2(i)+SIG3(i)==3:#三个策略信号约束RETT=( 0.5*RET_sig(R_c,i)+0*RET_sig(R_p,i) +0.5*RET_sig(R_b,i)+0*RET_sig(R_B,i) ).append(RETT)elif SIG1(i)+SIG2(i)+SIG3(i)==2:#三个策略信号约束RETT=( 0.4*RET_sig(R_c,i)+0*RET_sig(R_p,i) +0.6*RET_sig(R_b,i)+0*RET_sig(R_B,i) ).append(RETT)elif SIG1(i)+SIG2(i)+SIG3(i)==1:#三个策略信号约束RETT=( 0.3*RET_sig(R_c,i)+0*RET_sig(R_p,i) +0.7*RET_sig(R_b,i)+0*RET_sig(R_B,i) ).append(RETT)else:RETT=( 0.1*RET_sig(R_c,i)+0*RET_sig(R_p,i)+0.9*RET_sig(R_b,i) +0*RET_sig(R_B,i) ).append(RETT) return RETTRETT=quanzhong_fun(R_c,R_p,R_b,R_B)RETT=RETT.sort_index(axis=0)cum=np.cumprod(1+RETT)cum=cum.dropna()############################指数基准图#########################################start_time_jz=str(RETT.index[0])[:4]+str(RETT.index[0])[5:7]+str(RETT.index[0])[8:10]end_time_jz=str(RETT.index[-1])[:4]+str(RETT.index[-1])[5:7]+str(RETT.index[-1])[8:10]def bj_standard(code,lab='000300.SH',col='k'):#针对沪深股票，直接画出比较基准（收益情况）standard_base = pro.index_daily( ts_code=code, start_date=start_time_jz, end_date=end_time_jz)standard_base=standard_base.sort_index()standard_base.index=pd.to_datetime(standard_base.trade_date,format='%Y-%m-%d')#设置日期索引close_base= standard_base.closestandard_ret=standard_base.change/standard_base.close.shift(-1)standard_sig=pd.Series(0,index=close_base.index) standard_trade=standard_sig.shift(1).dropna()/100#shift(1)整体下移一行standard_SmaRet=standard_ret*standard_trade.dropna()standard_cum=np.cumprod(1+standard_ret[standard_SmaRet.index[0:]])-1plt.plot(close_base/close_base[-1],label=lab,color=col)return close_base/close_base[-1] #standard_cum########################################################################### def performance(port_ret):first_date = port_ret.index[0]final_date = port_ret.index[-1]time_interval = (final_date - first_date).days * 250 / 365# calculate portfolio's indicatornv = (1 + port_ret).cumprod()arith_mean = port_ret.mean() * 250geom_mean = (1 + port_ret).prod() ** (250 / time_interval) - 1sd = port_ret.std() * np.sqrt(250)mdd = ((nv.cummax() - nv) / nv.cummax()).max()sharpe = (geom_mean - 0) / sdcalmar = geom_mean / mddresult = pd.DataFrame({'算术平均收益': [arith_mean], '几何平均收益': [geom_mean], '波动率': [sd],'最大回撤率': [mdd], '夏普比率': [sharpe], '卡尔曼比率': [calmar]})print(result)return resultif __name__=="__main__":bj_standard('000300.SH')#比较基准函数（图）bj_standard('000012.SH',lab='上证国债',col='b')#bj_standard('NHCI.NH',lab='南华商品指数',col='y')performance(RETT)plt.plot(cum,label="组合策略",color='r',linestyle='-')plt.title("组合策略净值走势对比图")plt.legend()

看下权重变化

sig1=pd.Series( 0,RETT.index )sig2=pd.Series( 0,RETT.index )for i in range(len(G)):if SIG1(i)+SIG2(i)+SIG3(i)==3:#三个策略信号约束sig1[RET_sig(R_c,i).index]=0.5sig2[RET_sig(R_c,i).index]=0.5elif SIG1(i)+SIG2(i)+SIG3(i)==2:#三个策略信号约束sig1[RET_sig(R_c,i).index]=0.4sig2[RET_sig(R_c,i).index]=0.6elif SIG1(i)+SIG2(i)+SIG3(i)==1:#三个策略信号约束sig1[RET_sig(R_c,i).index]=0.3sig2[RET_sig(R_c,i).index]=0.7else:sig1[RET_sig(R_c,i).index]=0.1sig2[RET_sig(R_c,i).index]=0.9

import matplotlib.pyplot as pltimport numpy as npplt.stackplot(sig1.index, sig1, sig2, baseline='zero', labels=['沪深300权重', '上证国债权重'], colors=['r', 'b'])plt.legend(loc='upper right')plt.ylim(0, 1.2)plt.grid(axis='y', color='gray', linestyle=':', linewidth=2)plt.title("组合股债权重变动图")plt.show()

从上图可以看除，只有在股票配置存在一次权重为0.5情况。配置股票权重为0.4和0.3的次数相差不大，配置0.1的情况最多。

算术平均收益几何平均收益波动率最大回撤夏普比率卡尔曼比率组合策略9.17% 9.02% 7.34%12.47%1.230.72估值策略11.70% 11.51% 9.54%16.08%1.210.72动量策略15.33% 14.72% 15.51%32.11%0.950.46美林时钟17.28% 16.36% 18.09%34.13%0.900.48沪深30013.27% 9.62% 27.63%72.30%0.350.13上证国债3.63% 3.60% 0.96%2.25%3.751.60

通过上述结果分析，我们可以得到通过由三个策略构建的组合策略，其平均几何收益接近沪深300指数，单在风险方面沪深300的波动率高达27.63%，而我们的组合策略的波动率仅为9.17%显著优于沪深300。此外，组合策略也在波动率方面优于单一的估值、动量及美林时钟策略。组合的夏普比率也是优于任何单一策略。同时我们应该看到收益之所以会低于单一策略，是因为在我们的组合策略中，即使是三个判断信号SIG都判断高配股票，我们也仅仅配置0.5。现实大多数情况都是要充分重时风险的，做到稳中求胜。因此，综合比较之后，我们的组合策略能在较好的控制风险的情况下，取得不错的稳定收益。