Step-1: ###R Packages Installed
library(fpp2)
library(readxl)
library(SPlit)
library(fpp2)
###Declare time-series data
###In this study I have file naming as HPI_AUS. All the date excel format.
HPI_AUS <-ts(AUSHPI[,2], start=c(2002,3), frequency=4)
Step-3: ###While forecasting, first that should be done within the sample, and if it is found valid, then only we forrecast out of sample.
###Split the given data into the training and test sets
split_HPI_AUS <- ts_split(ts.obj = HPI_AUS, sample.out= 8)
training <- split_AUSHPI$train
testing <- split_AUSHPI$test
length(training)
length(testing)
Step-4:###Select the appropriate forecasting tool. Here I have used ARIMA (easy and quick model)
###ARIMA In-sample testing
arima_diag(training)
###Model forecasting In-sample
arima1 <- auto.arima(training, seasonal= TRUE)
autoplot(arima1)
check_res(arima1)
###Forecasting In-sample
fcast1<- forecast(arima1, h=8)
test_forecast(actual=HPI_AUS, forecast.obj = fcast1, test = testing)
accuracy(fcast1, testing)
###Auto ARIMA model Out of Sample
arima_diag(HPI_AUS)
fit_arima <- auto.arima(HPI_AUS, seasonal = TRUE )
autoplot(fit_arima)
check_res(fit_arima)
###Forecasting Out of Sample
fcast1 <- forecast(HPI_AUS, model = fit_arima, h=8)
plot_forecast(fcast1)
summary(fcast1)
###Do we have any option to validate the Forecasting Result?
Box.test(resid(fcast1), lag=1, type = "Ljung-Box")
Box.test(resid(fcast1), lag=5, type = "Ljung-Box")
Box.test(resid(fcast1), lag=10, type = "Ljung-Box")
Box.test(resid(fcast1), lag=15, type = "Ljung-Box")
