Research output: Working paper › Research

**Forecasting Macroeconomic Variables using Neural Network Models and Three Automated Model Selection Techniques.** / Kock, Anders Bredahl; Teräsvirta, Timo.

Research output: Working paper › Research

Kock, AB & Teräsvirta, T 2011 'Forecasting Macroeconomic Variables using Neural Network Models and Three Automated Model Selection Techniques' Institut for Økonomi, Aarhus Universitet, Aarhus.

Kock, A. B., & Teräsvirta, T. (2011). *Forecasting Macroeconomic Variables using Neural Network Models and Three Automated Model Selection Techniques*. Aarhus: Institut for Økonomi, Aarhus Universitet.

Kock AB, Teräsvirta T. 2011. Forecasting Macroeconomic Variables using Neural Network Models and Three Automated Model Selection Techniques. Aarhus: Institut for Økonomi, Aarhus Universitet.

Kock, Anders Bredahl and Timo Teräsvirta *Forecasting Macroeconomic Variables using Neural Network Models and Three Automated Model Selection Techniques*. Aarhus: Institut for Økonomi, Aarhus Universitet. 2011., 33 p.

Kock AB, Teräsvirta T. Forecasting Macroeconomic Variables using Neural Network Models and Three Automated Model Selection Techniques. Aarhus: Institut for Økonomi, Aarhus Universitet. 2011.

Kock, Anders Bredahl ; Teräsvirta, Timo. / **Forecasting Macroeconomic Variables using Neural Network Models and Three Automated Model Selection Techniques**. Aarhus : Institut for Økonomi, Aarhus Universitet, 2011.

@techreport{9af05103c64f41d8984adae7732d1dfe,

title = "Forecasting Macroeconomic Variables using Neural Network Models and Three Automated Model Selection Techniques",

abstract = "In this paper we consider the forecasting performance of a well-defined class of flexible models, the so-called single hidden-layer feedforward neural network models. A major aim of our study is to find out whether they, due to their flexibility, are as useful tools in economic forecasting as some previous studies have indicated. When forecasting with neural network models one faces several problems, all of which influence the accuracy of the forecasts. First, neural networks are often hard to estimate due to their highly nonlinear structure. In fact, their parameters are not even globally identified. Recently, White (2006) presented a solution that amounts to converting the specification and nonlinear estimation problem into a linear model selection and estimation problem. He called this procedure the QuickNet and we shall compare its performance to two other procedures which are built on the linearisation idea: the Marginal Bridge Estimator and Autometrics. Second, one must decide whether forecasting should be carried out recursively or directly. Comparisons of these two methodss exist for linear models and here these comparisons are extended to neural networks. Finally, a nonlinear model such as the neural network model is not appropriate if the data is generated by a linear mechanism. Hence, it might be appropriate to test the null of linearity prior to building a nonlinear model. We investigate whether this kind of pretesting improves the forecast accuracy compared to the case where this is not done.",

keywords = "artificial neural network, forecast comparison, model selection, nonlinear autoregressive model, nonlinear time series, root mean square forecast error, Wilcoxon’s signed-rank test",

author = "Kock, {Anders Bredahl} and Timo Ter{\"a}svirta",

year = "2011",

language = "English",

publisher = "Institut for {\O}konomi, Aarhus Universitet",

type = "WorkingPaper",

institution = "Institut for {\O}konomi, Aarhus Universitet",

}

TY - UNPB

T1 - Forecasting Macroeconomic Variables using Neural Network Models and Three Automated Model Selection Techniques

AU - Kock, Anders Bredahl

AU - Teräsvirta, Timo

PY - 2011

Y1 - 2011

N2 - In this paper we consider the forecasting performance of a well-defined class of flexible models, the so-called single hidden-layer feedforward neural network models. A major aim of our study is to find out whether they, due to their flexibility, are as useful tools in economic forecasting as some previous studies have indicated. When forecasting with neural network models one faces several problems, all of which influence the accuracy of the forecasts. First, neural networks are often hard to estimate due to their highly nonlinear structure. In fact, their parameters are not even globally identified. Recently, White (2006) presented a solution that amounts to converting the specification and nonlinear estimation problem into a linear model selection and estimation problem. He called this procedure the QuickNet and we shall compare its performance to two other procedures which are built on the linearisation idea: the Marginal Bridge Estimator and Autometrics. Second, one must decide whether forecasting should be carried out recursively or directly. Comparisons of these two methodss exist for linear models and here these comparisons are extended to neural networks. Finally, a nonlinear model such as the neural network model is not appropriate if the data is generated by a linear mechanism. Hence, it might be appropriate to test the null of linearity prior to building a nonlinear model. We investigate whether this kind of pretesting improves the forecast accuracy compared to the case where this is not done.

AB - In this paper we consider the forecasting performance of a well-defined class of flexible models, the so-called single hidden-layer feedforward neural network models. A major aim of our study is to find out whether they, due to their flexibility, are as useful tools in economic forecasting as some previous studies have indicated. When forecasting with neural network models one faces several problems, all of which influence the accuracy of the forecasts. First, neural networks are often hard to estimate due to their highly nonlinear structure. In fact, their parameters are not even globally identified. Recently, White (2006) presented a solution that amounts to converting the specification and nonlinear estimation problem into a linear model selection and estimation problem. He called this procedure the QuickNet and we shall compare its performance to two other procedures which are built on the linearisation idea: the Marginal Bridge Estimator and Autometrics. Second, one must decide whether forecasting should be carried out recursively or directly. Comparisons of these two methodss exist for linear models and here these comparisons are extended to neural networks. Finally, a nonlinear model such as the neural network model is not appropriate if the data is generated by a linear mechanism. Hence, it might be appropriate to test the null of linearity prior to building a nonlinear model. We investigate whether this kind of pretesting improves the forecast accuracy compared to the case where this is not done.

KW - artificial neural network, forecast comparison, model selection, nonlinear autoregressive model, nonlinear time series, root mean square forecast error, Wilcoxon’s signed-rank test

M3 - Working paper

BT - Forecasting Macroeconomic Variables using Neural Network Models and Three Automated Model Selection Techniques

PB - Institut for Økonomi, Aarhus Universitet

CY - Aarhus

ER -