R: Plotting IRF manually - r

Question

Consider following script to plot an impulse response function:
library(vars)
Canada <- Canada * 999
var <- VAR(Canada, p = 2, type = "both")
plot(irf(var, impulse = "rw", response = "U", boot = T, cumulative = FALSE, n.ahead = 20))
plot(irf(var, impulse = "rw", response = "U", boot = T, cumulative = TRUE, n.ahead = 20))
I wonder how I could access the data of the plot (and 95% intervals)?
It would be great to print a plot with a color filled confidence band, a green impulse response line and different axis descriptions. A solution with R's inbuild plot features would be preferred over ggplot.
Thanks!

Answer 1

You can view the data returned by irf:
library("vars")
# generate some dummy data
df <- data.frame(n=rnorm(100), p=rpois(100, 2))
var <- VAR(df, p = 2, type = "both")
irf <- irf(var, impulse = "n", response = "p", boot = T,
cumulative = FALSE, n.ahead = 20)
# inspect coefficients object
str(irf)
All the data you need is accessible from here (e.g. check irf$Lower and irf$Upper).
One way to customise the default plot would be to look at the source of the function being called when you run plot(irf):
vars:::plot.varirf
In this case it's a bit involved but you can copy the body of this function and edit the code to change the colours, draw a filled polygon and edit the labels of the axes to get them exactly the way you want.
Updated:
Here's a starting point for the confidence bands:
# set up the base plot
plot(irf$irf$n, type="n", ylim = c(-.3, .5),
ylab = "Your label", xlab = "Another label")
abline(h=0)
# draw the filled polygon for confidence intervals
polygon(
c(1:length(irf$Upper$n), length(irf$Lower$n):1),
c(irf$Upper$n, rev(irf$Lower$n)),
col = "grey80", border = NA)
# add coefficient estimate line
lines(irf$irf$n, col = "darkgreen")

Answer 2

I had a similar problem, so I modeled it myself. I am not an advanced R user so maybe someone can put that into a function or so.
This method creates a plot of all IRFs, with a vertical at y=0, the names of the impulses on the x-axis and the responses on the y-axis. The IRF-plots are also size-adjusted.
"VAR_BS_9016_5VAR" is my "varest" object. I used 5 variables but this method can easily be shortened or expanded.
par(mfrow=c(5,5), oma = c(0,0,0,0) + 0.1, mar = c(5,5,0,0) + 0.1)
for (i in 1:5){
for (j in 1:5){
var_plot=irf(VAR_BS_9016_5VAR, impulse = paste(colnames(VAR_BS_9016_5VAR$y)[i]), response=paste(colnames(VAR_BS_9016_5VAR$y)[j]), n.ahead = 20, ortho=TRUE, boot=TRUE, runs=1000, ci=0.9)
plot(x=c(1:21), y=unlist(var_plot$Lower), type="l", lwd = 3, lty=2,col="red", ylab=paste(colnames(VAR_BS_9016_5VAR$y)[j]), xlab=paste(var_plot$impulse), ylim=range(c(unlist(var_plot$Lower),unlist(var_plot$Upper))) )
lines(x=c(1:21),y=unlist(var_plot$Upper),type="l",lwd = 3, lty=2,col="red")
lines(x=c(1:21),y=unlist(var_plot$irf),type="l", lwd = 3)
abline(a = NULL, h = 0)
}
}

Answer 3

Here is my solution for obtaining a data frame that can be used in ggplot when you have multiple impulses and multiple responses.
For the pipe operator please get library(dplyr). Be careful since dplyr and MASS (dependency of vars-package) have naming conflicts (e.g., for "select"):
getIRFPlotData <- function(impulse, response, list) {
cbind.data.frame(Week = 0:(nrow(list[[1]][[1]])-1),
Lower = list[[2]][names(list[[2]]) == impulse][[1]] %>% as.data.frame() %>% dplyr::select_(response) %>% pull(1),
irf = list[[1]][names(list[[1]]) == impulse][[1]] %>% as.data.frame() %>% dplyr::select_(response) %>% pull(1),
Upper = list[[3]][names(list[[3]]) == impulse][[1]] %>% as.data.frame() %>% dplyr::select_(response) %>% pull(1),
Impulse = impulse,
Response = response, stringsAsFactors = FALSE)
}
With this you can return a data.frame with columns = Lower, irf, Upper, Impulse, Response. When you use dplyr::bind_rows() on the data frames you can stack the different data.frames on top of each other and using ggplot2::facet_wrap() and facet_grid() you can produce charts similar to the ones outputted by vars:::plot.varirf(), but are fully flexible to append stuff and work with the data.
getIRFPlotData("Spendings", "Returns", irf4c) %>% ggplot(.) + geom_line(aes(Week, Lower), linetype="dashed") + geom_line(aes(Week, irf)) + geom_line(aes(Week, Upper),linetype="dashed") + geom_ribbon(aes(Week, ymin=Lower, ymax=Upper), alpha = 0.3) + theme_minimal()