# I follow the R Programming guide, pages 54-59


library(vars) # Use the vars package
library(readxl)

currencies <- read_excel("D:/Programming guide/data/currencies.xls")
View(currencies)

head(currencies)
tail(currencies)

# Log returns
currencies$reur = c(NA ,100*diff(log(currencies$EUR)))
currencies$rgbp = c(NA ,100*diff(log(currencies$GBP)))
currencies$rjpy = c(NA ,100*diff(log(currencies$JPY)))
currencies = currencies [-1,] # Remove the 1st missing value

# OR #
rEUR <- ts(diff(log(currencies$EUR))*100,start=c(1998,12,15),frequency=252)
rGBP <- ts(diff(log(currencies$GBP))*100,start=c(1998,12,15),frequency=252)
rJPY <- ts(diff(log(currencies$JPY))*100,start=c(1998,12,15),frequency=252)
FX <- cbind(rEUR,rGBP,rJPY)
ts.plot(FX,col=1:3, main="FX returns")
###

# VAR estimation

VAR(currencies[c("reur","rgbp","rjpy")],p = 2) # VAR(2)

VARselect(currencies[c("reur","rgbp","rjpy")], lag.max = 10)

var = VAR(currencies[c("reur","rgbp","rjpy")],p = 1) # VAR(1)

# Granger causality

causality(var, cause=c("reur","rgbp"))$Granger

causality(var, cause=c("reur","rjpy"))$Granger

causality(var, cause=c("rgbp","rjpy"))$Granger

causality(var, cause=c("reur"))$Granger

causality(var, cause=c("rgbp"))$Granger

causality(var, cause=c("rjpy"))$Granger

# Impulse-response functions

ir = irf(var, n.ahead = 20)
plot(ir)

# Forecast error variance decomposition

vd = fevd(var,n.ahead = 20)
plot(vd)
par(mar = c(1,1,1,1))# Changing area margins
plot(vd)
vd

# Reversing the ordering in the VAR model

var_reverse = VAR(currencies[c("rjpy","rgbp","reur")],p = 1)
vd_reverse = fevd(var_reverse,n.ahead = 20)
plot(vd_reverse)
vd_reverse

# Forecasting

predict(var, n.ahead = 10)

# Normality test

normality.test(var)

# Test for arch

arch.test(var)

# Test for serial correlation, autocorrelation

serial.test(var)

# Model summary

summary(var)