autocorrelation. Moreover, we observe that the variable MA has a very high t-statistic so adding moving average terms into the model may also help to improve the estimation power and the robustness of the model. Typically, the order of autoregressive components can be determined by analyzing the autocorrelation function, however, in this case, the autocorrelation functions of the 48 models show different orders. To keep the models simple and improve their practical usability, we decided to include the same autoregressive and moving average components with all models. After some trial
and error, we decided to use AR(1), MA(1) and MA(7).