5.1 Training set/test set

One approach is to ‘hold out’ some of your data as the test data and did not use it at all in your fitting. To measure the forecast performance, you fit to your training data and test the forecast against the data in the test set. This is the approach that Stergiou and Christou used.

Stergiou and Christou used 1964-1987 as their training data and tested their forecasts against 1988 and 1989.

5.1.1 Forecast versus actual

We will fit to the training data and make a forecast for the test data. We can then compare the forecast to the actual values in the test data.

fit1 <- forecast::auto.arima(traindat)
fr <- forecast::forecast(fit1, h=2)
fr
##      Point Forecast    Lo 80    Hi 80    Lo 95    Hi 95
## 1988       10.03216 9.789577 10.27475 9.661160 10.40317
## 1989       10.09625 9.832489 10.36001 9.692861 10.49964

Plot the forecast and compare to the actual values in 1988 and 1989.

plot(fr)
points(testdat, pch=2, col="red")
legend("topleft", c("forecast","actual"), pch=c(20,2), col=c("blue","red"))