将 Data.Tables (R) 与循环或映射组合

Question

I am new to data tables in R and have managed to get 80% of the way through my analysis. The background is that I want to get the returns of a stock 5 days (before and after), and then 25 and 45 days after they report. I have successfully managed to do it for one set of dates (effectively hardcoding) but when I try and automate the process it falls apart.

I will start with my current formulas and then explain the data.

This formula successfully looks at the data tables and returns the sum that I need. The issue is that datem5 and V1 need to go through a loop (or mapply) to automate the process.

CQR_Date[CQR_DF[CQR_Date, sum(CQR), on = .(unit, date >= date1, date <= datem5),
            by = .EACHI], newvar := V1, on = .(unit, date1=date)]

I tried this (along with many other variants). Please note the newvar needs to be addressed as well.

for (i in 1:4) {
              CQR_Date[CQR_DF[CQR_Date, sum(CQ), on = .(unit, date >= date1, date <= cols[,..i]),
              by = .EACHI], newvar := v, on = .(unit, date1=date)]

but get this error

Error: argument specifying columns specify non existing column(s): cols[3]='cols[, ..i]'

Interestingly, when I try

for (i in 1:2) {
 y <- cols[,..i]}

There is no issue.

Now in terms of data;

• col just contains the column headings that I need from CQR_Data

  cols <- data.table("datem5", "datep5", "datep20", "datep45")

CQ_Data has the reporting dates for the stock CQ such as the following

 CQ_Date <- data.frame("date1" = anydate(c("2016-02-17", "2016-06-12", "2016-08-17")))
 CQ_Date$datem5 <- CQ_Date$date1 - 5  # minus five days
 CQ_Date$datep5 <- CQ_Date$date1 + 5  # plus five days
 CQ_Date$datep20 <- CQ_Date$date1 + 20
 CQ_Date$datep45 <- CQ_Date$date1 + 45
 CQ_Date$unit <- 1    # I guess I need this for some sort of indexing

Then CQ_DF (it is the log returns for the stock) is formed by:

 CQ_DF <- data.frame("unit" = rep(1,300))
 CQ_DF$CQ <- rnorm(10)
 CQ_DF$date <- seq(as.Date("2015-12-25"), by = "day", length.out = 300)
 CQ_DF$unit <- 1

Before setting them as DT

setDT(CQ_DF)
setDT(CQ_Date)

Any help would be greatly appreciated. Note this uses

  library(data.table)
  library(anytime)     

A simplified version is:

  CQ_Date <- data.frame("date1" = c(10, 20))
  CQ_Date$datep5 <- CQ_Date$date1 + 5  # plus five days
  CQ_Date$datep20 <- CQ_Date$date1 + 10
  CQ_Date$unit <- 1 

  CQ_DF <- data.frame("unit" = rep(1,100))
  CQ_DF$CQ <- seq(1, by = 1, length.out = 100)
  CQ_DF$date <- seq(1, by = 1, length.out = 100)
  CQ_DF$unit <- 1

  setDT(CQ_DF)
  setDT(CQ_Date)

  cols <- c("datep5", "datep20" )

  tmp <- melt(CQ_Date, measure.vars = cols)
  setDT(tmp)

  tmp[CQ_DF[tmp, sum(CQ), on = .( unit, date >= date1,  date <= value), by = 
  .EACHI],newvar := V1, on = .(unit, date1=date  )]

The issue is now that the sum does not appear to work correctly. It may have something to do with "variable" variable.

Answer 1

Instead of using mapply or for loop, try reshaping the dataset in long format using melt , create sequence between the numbers, perform the join and calculate the sum .

library(data.table)
cols <- c("datep5", "datep20" )

tmp <- melt(CQ_Date, measure.vars = cols)
tmp <- melt(CQ_Date, measure.vars = cols)
tmp <- tmp[, list(date = seq(date1, value)), .(unit, variable, date1, value)]
tmp <- merge(tmp, CQ_DF, by = c('unit', 'date'))
tmp[, .(newvar = sum(CQ)), .(unit, variable, date1)]

#   unit variable date1 newvar
#1:    1   datep5    10     75
#2:    1  datep20    10    165
#3:    1   datep5    20    135
#4:    1  datep20    20    275

If you need the data back in wide format you can use dcast .

---

Equivalent tidyverse option is:

library(tidyverse)

CQ_Date %>%
  pivot_longer(cols = cols) %>%
  mutate(date = map2(date1, value, seq)) %>%
  unnest(date) %>%
  left_join(CQ_DF, by = c('unit', 'date')) %>%
  group_by(unit, name, date1) %>%
  summarise(newvar = sum(CQ))