反转因子与流动性因子的构建，参照Fama-French五因子方法

流动性因子：

流动性因子构建过程为：首先根据t年6月底上市公司的流通市值进行排序，分为大公司（B）和小公司（S）两组，然后再根据t-1年年底上市公司的流动性比率进行排序，非流动性（Amihud指标,前面有计算方法）最高的前30%记为I组，流动性比率中间40%记为M组，非流动性最低的前30%记为L组，每一年分一次组。最后计算出非流动性因子IML。

setwd("e:/R/tail risk/day/sp")
setwd("e:/R/tail risk/month")

r1<-read_excel("r1.xlsx",sheet = 1);r2<-read_excel("r2.xlsx",sheet = 1)
r3<-read_excel("r3.xlsx",sheet = 1);r4<-read_excel("r4.xlsx",sheet = 1)
rr1<-read_excel("rr1.xlsx",sheet = 1);rr2<-read_excel("rr2.xlsx",sheet = 1);rr3<-read_excel("rr3.xlsx",sheet = 1)
rr4<-read_excel("rr4.xlsx",sheet = 1);rr5<-read_excel("rr5.xlsx",sheet = 1);rr6<-read_excel("rr6.xlsx",sheet = 1)
rr7<-read_excel("rr7.xlsx",sheet = 1);rr8<-read_excel("rr8.xlsx",sheet = 1);rr9<-read_excel("rr9.xlsx",sheet = 1)
r1<-slice(r1,-(1L:2L));r2<-slice(r2,-(1L:2L));r3<-slice(r3,-(1L:2L));r4<-slice(r4,-(1L:2L))
rr1<-slice(rr1,-(1L:2L));rr2<-slice(rr2,-(1L:2L));rr3<-slice(rr3,-(1L:2L));rr4<-slice(rr4,-(1L:2L));rr5<-slice(rr5,-(1L:2L))
rr6<-slice(rr6,-(1L:2L));rr7<-slice(rr7,-(1L:2L));rr8<-slice(rr8,-(1L:2L));rr9<-slice(rr9,-(1L:2L))
r5<-bind_rows(rr9,r1,r2,r3,r4,rr1,rr2,rr3,rr4,rr5,rr6,rr7,rr8)   
## 把时间改成数字并挑选其中的字符。
sp<-(data.frame(r5,Trdmnt=as.numeric(str_sub(str_replace_all(r5$Trddt,"-",""),start=1L,end=8L)))%>%select(-2))[,c(1,6,3,4,5)]
write.csv(sp,"sp.csv",row.names = FALSE)
sp<-read.csv("sp.csv")
a<-group_by(sp,Stkcd,Trdmnt)%>%summarise(hi=max(Hiprc),lo=min(Loprc),mean=mean(Clsprc))
yl<-read_excel("yl.xlsx",sheet = 1)%>%slice(-(1L:2L))
l5<-(data.frame(select(yl,-2),Trdmnt=as.numeric(str_replace_all(yl$Trdmnt,"-",""))))[,c(1,6,2,3,4,5)]
write.csv(l5,"l5.csv",row.names = FALSE)
l5<-read.csv("l5.csv")
l6<-mutate(l5,ltg=Msmvosd*1000/Mclsprc)%>%select(-3)%>%rename(tvn=Mnvaltrd)
l<-left_join(a,l6)
write.csv(l,"l.csv",row.names = FALSE)
l<-read.csv("l.csv")
# 计算出HHR指标(为另一种计算流动性指标的方法)
ll1<-na.omit(l) # 删除所有空值的行
ll2<-mutate(ll1,hhr=(hi-lo)/lo*ltg*mean/tvn)%>%filter(hhr!=0,hhr!="NA")
# 利用Amihud流动性指标
ll1<-select(l,1,2,7,8)%>%na.omit()
illiq<-read.csv("illiq.csv")%>%tbl_df() 
ll2<-inner_join(ll1,illiq,by=c("Trdmnt","Stkcd"))%>%rename(hhr=illiq)
# 计算出2×3分组的组合流动性
sh<-list();bh<-list();sr<-list();br<-list()
for(i in 0:24){
  ll3<-filter(left_join(left_join(select(filter(ll2,Trdmnt==199406+100*i)%>%arrange(Msmvosd)%>%slice(1:(n()/2)),1),
              filter(ll2,Trdmnt==199312+100*i))%>%arrange(desc(hhr))%>%slice(1:(n()*0.3))%>%select(1),ll2),
              Trdmnt>(199406+100*i)&Trdmnt<199507+100*i)
  sh[[i+1]]<-group_by(ll3,Trdmnt)%>%summarise(r=weighted.mean(Mretwd, Msmvosd))
  ll4<-filter(left_join(left_join(select(filter(ll2,Trdmnt==199406+100*i)%>%arrange(desc(Msmvosd))%>%slice(1:(n()/2)),1),
              filter(ll2,Trdmnt==199312+100*i))%>%arrange(desc(hhr))%>%slice(1:(n()*0.3))%>%select(1),ll2),
              Trdmnt>(199406+100*i)&Trdmnt<199507+100*i)
  bh[[i+1]]<-group_by(ll4,Trdmnt)%>%summarise(r1=weighted.mean(Mretwd, Msmvosd))
  ll5<-filter(left_join(left_join(select(filter(ll2,Trdmnt==199406+100*i)%>%arrange(Msmvosd)%>%slice(1:(n()/2)),1),
              filter(ll2,Trdmnt==199312+100*i))%>%arrange(hhr)%>%slice(1:(n()*0.3))%>%select(1),ll2),
              Trdmnt>(199406+100*i)&Trdmnt<199507+100*i)
  sr[[i+1]]<-group_by(ll5,Trdmnt)%>%summarise(r2=weighted.mean(Mretwd, Msmvosd))
  ll6<-filter(left_join(left_join(select(filter(ll2,Trdmnt==199406+100*i)%>%arrange(desc(Msmvosd))%>%slice(1:(n()/2)),1),
              filter(ll2,Trdmnt==199312+100*i))%>%arrange(hhr)%>%slice(1:(n()*0.3))%>%select(1),ll2),
              Trdmnt>(199406+100*i)&Trdmnt<199507+100*i)
  br[[i+1]]<-group_by(ll6,Trdmnt)%>%summarise(r3=weighted.mean(Mretwd, Msmvosd))}
sh<-do.call("rbind",sh);bh<-do.call("rbind",bh);sr<-do.call("rbind",sr);br<-do.call("rbind",br)
hmr<-data.frame(sh,bh,sr,br)%>%mutate(hmr=(r+r1)/2-(r2+r3)/2)%>%select(1,9)
write.csv(hmr,"hmr.csv",row.names = FALSE)

反转因子：

根据Jegadeesh（1990）和田利辉（2014）对上市公司反转效应的研究，本文将个股按照上一个月的收益率排序，把个股收益率最大的30%记为赢者组合，最小的30%记为输者组合，最后将赢者组合收益率与输者组合在t时期差记为反转因子。

y6<-read.csv("l5.csv")%>%select(-3,-4)%>%na.omit()
fz1<-numeric()
# 两表格匹配最好用inner_join 函数，不会出现缺失值
for(i in 0:23){
  for(h in 0:11){
    f1<-filter(y6,Trdmnt==max(select(filter(y6,Trdmnt<199501+h+100*i),Trdmnt)))%>%arrange(desc(Mretwd))%>%slice(1:(n()*0.3))
    f2<-filter(y6,Trdmnt==max(select(filter(y6,Trdmnt<199501+h+100*i),Trdmnt)))%>%arrange(Mretwd)%>%slice(1:(n()*0.3))
    fz1[i*12+(h+1)]<-weighted.mean(inner_join(select(f1,1),filter(y6,Trdmnt==199501+h+100*i),by = "Stkcd")$Mretwd,
                                   inner_join(select(f1,1),filter(y6,Trdmnt==199501+h+100*i),by = "Stkcd")$Msmvosd)-
      weighted.mean(inner_join(select(f2,1),filter(y6,Trdmnt==199501+h+100*i),by = "Stkcd")$Mretwd,
                    inner_join(select(f2,1),filter(y6,Trdmnt==199501+h+100*i),by = "Stkcd")$Msmvosd)}}
fz2<-data.frame(fz=fz1)
fz<-bind_cols(arrange(filter(distinct(y6,Trdmnt),Trdmnt>=199501&Trdmnt<=201812),Trdmnt),fz2)
write.csv(fz,"fz.csv",row.names = FALSE)