Prediction of macro-economic time series. A comparison of linear models and neural nets (Q2865994)

This book is based on the dissertation of the author which was defended in 2012 in Vienna and is written in German. It summarizes and extends a variety of methods for neural network-based autoregressive time series models, where (part of) the regression function is modelled by a neural network as a way to include nonlinearities. In fact, one of the main aims of the book is the discussion of non-linearities in comparison to linear models, that are frequently used in the econometric literature. Throughout the book the methods are applied to two time series examples: the Austrian unemployment rate as well as the industry production index (each in monthly values from January 1960 to December 1997).NEWLINENEWLINEThe book contains the following six chapters: Chapter 1: Introduction; Chapter 2: Linear modelling of time series: in this chapter a short introduction into ARMA models and extensions (SARMA, ARDS) is given; Chapter 3: Test for nonlinearity: after a short introduction into nonlinear time series, tests for nonlinear structures are discussed; Chapter 4: Neural networks and time series analysis: neural network-based autoregressive time series models are introduced and methods for estimation and specification given; Chapter 5: Evaluation of the prediction: the prediction quality of neural network-based autoregressive time series is compared with linear methods; Chapter 6: Conclusions.NEWLINENEWLINEIn conclusion, the book is certainly well suited for a practitioner or someone interested in getting an overview of the topic in particular with emphasize on economic time series. However, it is not suitable for readers interested in the mathematical details of the proposed methods.