Optimal statistical model for forecasting ozone

The objective of this paper is to apply time series analysis to ozone data in order to obtain the optimal forecasting model. Different ARMA models are fitted to the ozone data and the best fitted model, ARMA(20,2), is found to produce the best predictions with MAPE = 42%. Applying simple exponential smoothing to the time series, however, results in even higher accuracy for predictions. This leads us to believe that in certain cases depending on the characteristics of the time series, naive methods of forecasting may produce more accurate results.