rstatisticsarimaautoregressive-modelssarimax

ARIMA Model in R - datacamp exercise -

in a datacamp exercise it seems to me not clear at all the answer

  1. We have the data: globaltemp , provide by asta package, also I use astsa package functions
  2. The exercise says that evaluate this data using sarima and decide which of the two models provide on the bottom have the best fit:

sarima(globtemp,1,1,1) AIC and BIC values: -1.716773 and -1.630691, respectively

(graphic result with asta package for sarima(globtemp,1,1,1) is here) enter image description here

sarima(globtemp,0,1,2) AIC and BIC values: -1.723268 and -1.637185, respectively (graphic result with asta package for sarima(globtemp,0,1,2) is here) enter image description here

Datacamp says the model with the best fit is the second. However, the model with the less AIC and BIC value is sarima(globtemp,1,1,1). Why the correct answer is the second model? Is it a mistake there?

Thanks for your time!

I look at the AIC and BIC values, and I expect that the first model: sarima(globaltemp,1,1,1) is the best model in comparison of sarima(globtemp,0,1,2); however, in datacamp, it says that the best is sarima(globtemp,0,1,2)

Solution

  • Both models are pretty similar but I guess the simpler the better due to the negligible difference in AIC and BIC values.

    In this case ARIMA(0,1,2) its simpler because it lacks the autoregresive component p in the formulation ARIMA(p,d,q).

    See fpp2 for more info.