This thesis offers a thorough review of time series forecasting models, starting with simple forecasting methods such as linear models and ending up combining this method with dynamic models to represent interdependencies of the time series dynamics. Decomposition methods such as those from X-11 method to Seasonal and Trend decomposition using Loess (STL) method are applied to increase the prediction accuracy. As a new contribution to the literature, Artificial Neural Networks are applied to increase the chance to capture different patterns in the data and improve the forecast performance.
The main parts of this thesis include the practical application of the time series forecasting models on the health insurance sector and electricity consumption, as well as on simulated data sets where different model adjustments are compared and practical recommendations regarding model choice and calibration derived.
