While there a many measures of liquidity, the metric in Pastor and Stambaugh (2003) is the standard used in empirical asset pricing. The first thing we need is daily data. Academics can use the CRSP data set but the idea can be replicated with free data. You will need Date, unique stock identifier (I use CUSIP but ticker works just as well), Volume, Return (make sure to include dividends), and the market return (Here I use the CRSP value weighted index but any broad value weighted index should work). Assuming you’ve already formatted the date properly and the data so everything is numeric and not factor or string, we start with a panel I’ve named ‘crsp’. I also recommend setting up a parallel backend to make use of all your CPU cores. Here I use the parallel package but if you have multiple machines be sure to check out doRedis for a relatively simple way to make the poor man’s computer cluster.
require(data.table); require(doParallel);
require(parallel); require(foreach);
registerDoParallel(detectCores()-1) # Keep 1 CPU core for the OS
crsp=fread(paste(path,'Data_Sets/WRDS_CRSP_LiquidityCalc_1993-2013.csv',sep=''),header=TRUE)
setnames(crsp,c('PERMNO','Date','Ticker','CUSIP','Price','Volume','Return','Shares','MktReturn'))
crsp=crsp[Price>5]; crsp[,MktCap:=1000*Shares*Price];
crsp=crsp[MktCap>1000000]
crsp[,Date:=as.Date(as.character(Date),format='%Y%m%d')];
crsp[,Month:=as.yearmon(Date)]; crsp[,Return:=as.numeric(Return)]
#### Create Metrics ##############################################
crsp[,Ret.e:=Return-MktReturn];
crsp[,SVolume:=Volume*(Ret.e/abs(Ret.e))]
crsp[,L.Ret.e:=c(NA,head(Ret.e,-1)),by='CUSIP']
crsp[,L.SVolume:=c(NA,head(SVolume,-1)),by='CUSIP']
crsp=crsp[,list(Date,Month,CUSIP,Ret.e,L.Ret.e,L.SVolume)]
#### Eliminate months with insufficient observations #############
crsp[,Obs:=length(na.omit(L.Ret.e)),by=c("CUSIP","Month")]
crsp=crsp[Obs>10]
#### Run model on daily data for each month ######################
model=Ret.e~L.Ret.e+L.SVolume
gammas=foreach(cusip=unique(crsp[,CUSIP]),.combine='rbind') %dopar% {
stock=crsp[CUSIP==cusip];
stock[,Gamma:=coef(lm(model,data=.SD))[3],by=c('Month')]
stock=unique(stock[,list(Month,CUSIP,Gamma)]); return(stock)
}
save(gammas,file=paste(path,'Panels/Liquidity_Panel.RData',sep=''))
Stephen Rush 