动量模型回测——基于米矿ricequant的jupyter notebook平台

动量模型回测——基于米矿ricequant的jupyter notebook平台

受@earletf 启发，做个动量模型验证验证：

思路来源文章：https://xueqiu.com/3559889031/166758703

标的：

biaodichi=['159996.XSHE','512400.XSHG','515210.XSHG','515220.XSHG','515030.XSHG','512800.XSHG','512880.XSHG','512290.XSHG','512170.XSHG','512200.XSHG','512790.XSHG','515050.XSHG','512660.XSHG','512690.XSHG','510650.XSHG']#标的

参数设置：

time1p=10  #时间参数1,前x天动量
time2p=10  #时间参数2，后x天动量
headc=3  #买入前3个
tailc=3  #买入后3个
startdate= '20191001' #数据开始时间
enddate='20201231' #数据开始时间
frequency='1d' #日线数据

代码：
提取某证券历史行情数据

def tiqusinglestock(stock,fretemp):
    aaa=get_price(stock,startdate,enddate, frequency=fretemp, fields=None, adjust_type='pre', skip_suspended=False)
    aaa.dropna(axis=0, how='any', inplace=True)
    if aaa is None:  
        print('-')   
    else: 
        aaa=aaa.reset_index(inplace=False)
        aaa['code']=stock
        aaa['name']=instruments(stock).symbol
    return aaa

计算某证券之前x天动量（涨跌）和之后x天动量（涨跌）：

def cal_dongliang(aaa,time1ptemp,time2ptemp):
    aaa['t1dongliang']=NaN    
    aaa['dleffect']=NaN
    length=len(aaa)
    co=aaa.columns.get_loc('close')
    c1=aaa.columns.get_loc('t1dongliang')
    c2=aaa.columns.get_loc('dleffect')
    aaa.iloc[time1ptemp:length,c1]=aaa.iloc[time1ptemp:length,co].values/aaa.iloc[0:(length-time1ptemp),co].values
    aaa.iloc[0:(length-time2ptemp),c2]=aaa.iloc[time2ptemp:length,co].values/aaa.iloc[0:(length-time2ptemp),co].values
    bbb=aaa[['date','code','name','close','t1dongliang','dleffect']]
    return bbb

计算动量效应

def cal_effect(dataalltemp):
    a=dataalltemp.date.drop_duplicates()
    guocheng = pd.DataFrame()#
    new_col = ['date','time1p','time2p','best1','best1r','worst1','worst1r','avrr','chaoeshouyi','chaoeshouyi2','chaoeshouyib','chaoeshouyiw','chaoeshouyiavrr']
               #统计信息的格式

    h=len(a.values)
    x=0 #计数器
    b=1 #最好值净值
    w=1 #最差值净值
    av=1 #平均值净值
    
    while x<h:
        date=a.values[x]
        best1=dataalltemp[dataalltemp.date==date].head(headc).name.values[0]   #当天动量最好证券名称
        best1r=dataalltemp[dataalltemp.date==date].head(headc).dleffect.values.mean()  #当天动量最好前x个证券之后x天持仓期动量平均值
        worst1=dataalltemp[dataalltemp.date==date].tail(tailc).name.values[-1] #当天动量最差前x个证券名称
        worst1r=dataalltemp[dataalltemp.date==date].tail(tailc).dleffect.values.mean() #当天动量最差前x个证券之后x天持仓期动量平均值
        avrr=dataalltemp[dataalltemp.date==date].dleffect.mean()   #全部证券之后x天持仓期动量平均值        
        b=b*best1r
        w=w*worst1r
        av=av*avrr
        x=x+time2p  #跳过之后x天持仓期
        chaoeshouyi=best1r-avrr  #收益是否跑赢平均值
        chaoeshouyi2=b-av  #相对收益累计绝对值
        chaoeshouyib=best1r-1  #最好动量是否大于1
        chaoeshouyiw=worst1r-1  #最差动量是否大于1
        chaoeshouyiavrr=avrr-1 #同期均值动量是否大于1
        
        sgl1=[[date,time1p,time2p,best1,best1r,worst1,worst1r,avrr,chaoeshouyi,chaoeshouyi2,chaoeshouyib,chaoeshouyiw,chaoeshouyiavrr]]
        guocheng=guocheng.append(sgl1,ignore_index=True)

    guocheng.columns = new_col        
    slav=len(guocheng[guocheng.chaoeshouyi>0])/len(guocheng)  
    slb=len(guocheng[guocheng.chaoeshouyib>0])/len(guocheng)      
    slw=len(guocheng[guocheng.chaoeshouyiw<0])/len(guocheng)   
    sla=len(guocheng[guocheng.chaoeshouyiavrr<0])/len(guocheng)          
    tongji=[[time1p,time2p,b,w,av,slav,slb,slw,sla]]
    
    return guocheng,tongji

主程序运行：

plguocheng = pd.DataFrame()
pltongji = pd.DataFrame()
count=1
for time1p in range(1,61):  #动量效应参数（前）
    for time2p in range(1,61):  #动量效应参数（后）
        print('进度：'+str(count/3600))   #显示下进度免得着急
        count=count+1
#-------------------------------------------------------------------
        dataall = pd.DataFrame()
        for x in biaodichi:
            a=tiqusinglestock(x,frequency)    #提取原始数据
            c=cal_dongliang(a,time1p,time2p)   #计算动量
            dataall=dataall.append(c,ignore_index=True)  #结果合并
        dataall.sort_values(by=['date','t1dongliang'],inplace=True,ascending=[True,False])  #数据排序，先按时间从低到高，再按照前期动量由高到低
        dataall= dataall.reset_index(drop=True)  #重置索引
        dataall.dropna(axis=0, how='any', inplace=True)  #丢掉Nan
        a1,a2=cal_effect(dataall)   #计算动量效应
        plguocheng=plguocheng.append(a1,ignore_index=True)   #计算动量效应结果添加
        pltongji=pltongji.append(a2,ignore_index=True)   #计算动量效应结果添加

#-------------------------------------------------------------------

输出结果

new_col = ['time1p','time2p','b','w','av','slav','slb','slw','sla']
pltongji.columns = new_col
pltongji.to_excel('统计-'+str(ticks)+'-.xlsx',encoding="utf_8")     
plguocheng.to_excel('过程-'+str(ticks)+'-.xlsx',encoding="utf_8")  

结果分析：
动量效应确实存在。