Application of linear and nonlinear time series analysis on some selected macroeconomic variables

Nowadays, the determination of the movement of economic variables and the direction of the change they will show in the future have gained importance for decision and policy makers. Accuracy and consistency are required in the prediction of the goals or plans to be created. In this respect, reliability gains importance in the forecasting of macroeconomic variables. Artificial Neural Networks and hybrid modeling techniques based on artificial neural networks, which have become widespread recently, appear as a successful analysis tool in this sense. Traditional time series models, autoregressive conditional hetereoscedasticity models, Markov regime switching models, threshold and transition models alone are insufficient in predicting variables compared to hybrid structures. The hypothesis of the thesis is that "hybrid techniques will produce the most reliable and consistent results in predicting macroeconomic variables". In this thesis, seven macroeconomic variables including different features of time series were analyzed for the Turkish economy. The working period was discussed for 01/01/1997-31/10/2020 and predictions were made. Forecasting values for the period 01/01/2020-31/10/2020 were compared with the actual serial values. The study has been handled under three main headings. In the first chapter, the properties of time series and artificial neural network theory are mentioned. In the second part, the theoretical structures of the above-mentioned models are discussed. In the last section, necessary estimations and comparisons are made for the variables of BIST 100 Index return, unemployment rate, inflation, real money supply, real effective exchange rate, exports and net external debt stock. Hybrid model results for all series discussed in the study; the model predictions based on the structure of the series and the model results based on the artificial neural network were more successful and obtained the closest values to the truth. In this sense, it has been determined that hybrid models based on artificial neural networks can be used to increase the forecasting performance of macroeconomic variables for decision makers.