Replicates Chris Giles' analysis of IMF WEO

imf.R

library(gdata)
library(countrycode)

infile <- "IMFmultipliers.xls"
imf_rep <- read.xls(infile, sheet = 2,
	skip = 3,
	header=F)

names(imf_rep) <- c("Country",
	"gdp_4cast_2011","gdp_4cast_2012",
	"struct_bal_2010","struct_bal_2011","struct_bal_2012",
	"cyc_adj_prim_bal_2010","cyc_adj_prim_bal_2012",
	NA,"gdp_4cast_2011_new","gdp_4cast_2012_new",
	NA,NA,"cumgrowth_2010","cumgrowth_2012","Country.again",
	"delta_growth","delta_struct_bal","delta_cyc_adj_prim_bal","ca_def")

imf_rep <- imf_rep[which(imf_rep$Country!=""),]
imf_rep$Country.again<-NULL
imf_rep$iso3c <- countrycode(imf_rep$Country,"country.name","iso3c")
imf_rep$iso3c [ which(imf_rep$Country == "Kosovo") ] <- "KOS"
rownames(imf_rep) <- imf_rep$iso3c

## Exclude missing countries
imf_rep <- imf_rep[,!is.na(names(imf_rep))]
imf_rep <- imf_rep[!is.na(imf_rep$delta_struct_bal),]

par (mar=c(3,3,2,1), mgp=c(2,.7,0), tck=-.01)
with(imf_rep,
	plot(delta_struct_bal,delta_growth,type="n",
	xlab = "Fiscal consolidation",ylab="Growth forecast error"))
with(imf_rep,
	text(delta_struct_bal,delta_growth,iso3c,cex=.9))

summary(imf.mod <- lm(delta_growth~delta_struct_bal,data=imf_rep))
abline(imf.mod)
with(imf_rep,
	text(min(delta_struct_bal,na.rm=T),max(delta_growth,na.rm=T),
		paste("y=", 
			round(coef(imf.mod)["(Intercept)"],2), 
			"+  x * ",
			round(coef(imf.mod)["delta_struct_bal"],2),
			", p-value = ",
			round(summary(imf.mod)$coef["delta_struct_bal","Pr(>|t|)"],2),
		sep=""),
	pos=4))

png(file="imf_outliers.png",width=600,height=500)
par(mfrow=c(2,2),mar=c(3,3,2,1), mgp=c(2,.7,0), tck=-.01)
plot(imf.mod)
dev.off()

outliers <- names(which(cooks.distance(imf.mod)>1))

imf.mod2 <- update(imf.mod, 
	data = imf_rep[!is.element(imf_rep$iso3c,outliers),])
summary(imf.mod2)

png(file="imf_outliers2.png",width=600,height=500)
par(mfrow=c(2,2),mar=c(3,3,2,1), mgp=c(2,.7,0), tck=-.01)
plot(imf.mod2)
dev.off()

imf.mod3 <- update(imf.mod, 
	data = imf_rep[!is.element(imf_rep$iso3c,c("DEU","GRC")),])
summary(imf.mod3)

## Do mash-up of all
png(file="imf_final.png",width=600,height=500)
par (mar=c(3,3,2,1), mgp=c(2,.7,0), tck=-.01)
with(imf_rep,
	plot(delta_struct_bal,delta_growth,type="n",
	xlab = "Fiscal consolidation",ylab="Growth forecast error"))
with(imf_rep,
	text(delta_struct_bal,delta_growth,iso3c,cex=.9))

abline(imf.mod)
abline(imf.mod2,lty=2,col="red")
abline(imf.mod3,lty=3,col="blue")

with(imf_rep,
	text(min(delta_struct_bal,na.rm=T),max(delta_growth,na.rm=T),
		paste("y=", 
			round(coef(imf.mod)["(Intercept)"],2), 
			"+  x * ",
			round(coef(imf.mod)["delta_struct_bal"],2),
			", p-value = ",
			round(summary(imf.mod)$coef["delta_struct_bal","Pr(>|t|)"],2),
			", All",
		sep=""),
	pos=4))


with(imf_rep,
	text(min(delta_struct_bal,na.rm=T),max(delta_growth,na.rm=T)-1,
		paste("y=", 
			round(coef(imf.mod2)["(Intercept)"],2), 
			"+  x * ",
			round(coef(imf.mod2)["delta_struct_bal"],2),
			", p-value = ",
			round(summary(imf.mod2)$coef["delta_struct_bal","Pr(>|t|)"],2),
			", w/o ",paste(outliers,collapse=", "),
		sep=""),
	pos=4,col="red"))

with(imf_rep,
	text(min(delta_struct_bal,na.rm=T),max(delta_growth,na.rm=T)-2,
		paste("y=", 
			round(coef(imf.mod3)["(Intercept)"],2), 
			"+  x * ",
			round(coef(imf.mod3)["delta_struct_bal"],2),
			", p-value = ",
			round(summary(imf.mod3)$coef["delta_struct_bal","Pr(>|t|)"],2),
			", w/o GRC, DEU",			
		sep=""),
	pos=4,col="blue"))

dev.off()

## Choose to exclude countries taken two at a time
combinations <- combn(unique(imf_rep$iso3c),2,simplify=F)
my.pvals <- rep(NA,length(combinations))

for (i in 1:length(combinations)) {
	my.subset <- imf_rep[!is.element(imf_rep$iso3c,combinations[[i]]),]
	the.mod <- update(imf.mod,data=my.subset)
	my.pvals[i] <- summary(the.mod)$coef[2,4]
}


length(my.pvals)
length(which(my.pvals>0.05))
length(which(my.pvals>0.1))

This gist is pretty interesting, and I just forked it to remix the code for a class. However, I am confused about the naming of the columns. It seems that you import sheet #2 of this spreadsheet, but use the names from sheet #1, which are at T+1. As a result, your CAPB 2012 column, for instance, is actually CAPB 2011, based on the IMF WEO from the previous year. I hope to be mistaken.