Modelling temperature dynamics

This work presents a time series model for daily average temperatures. The data is modeled by flexible low-order autoregressive terms, seasonality components and a deterministic volatility capturing the heteroscedasticity of the residuals. The study attempts to find evidence in shifts in the variance part over time which is attributed to the global warming effect. The model is applied to the industrial Bleu Banana European area with the data covering the period from 1973 to 2008. After deseasonalizing and detrending the data, four standard approaches for modelling daily temperature dynamics are estimated and evaluated. We found out that the multiplicative model of Fourier and GARCH terms in volatility outperforms the others. Furthermore, expectile curves and quantile curves, applied on temperature residuals, are presented to detect fluctuations in temperature variance and evidence for global warming