GARCH models for particulate matter PM10 air pollutant in the city of Ruse, Bulgaria

This paper presents empirical study of air pollution of Bulgarian city caused by PM10 (particulate matter 10 micrometers or less in diameter). Univariate ARIMA, based on the natural logarithm transformed values, hybrid ARIMA-GJR-GARCH and hybrid ARIMA-EGARCH models are constructed and statistically evaluated. The comparison between the three models is made by widely used for this purpose Root Mean Squared Error, Mean Absolute Error and Theil Inequality Coefficient, calculated separately for the fitted and forecasted values for the next 1, 2, 3, 4, 5, 6, 8 and 10 days. The hybrid ARIMA-GJR-GARCH model has better values for all RMSE, MAE and TIC compared to the ARIMA model. The characteristics for the fitted values by the two hybrid models are almost the same, without the apparent superiority of either. The hybrid ARIMA-EGARCH model has smaller values of RMSE, MAE and TIC based on the prediction for the next 3 to 10 days, compared to ARIMA-GJR-GARCH model. The main advantage of the hybrid models is that they interpret directly the original values of PM10 and at the same time model the conditional variance of the process. The predictions for the conditional variance by the two hybrid models are agreed andcapture well the increases in volatility of the values of PM10.