Artificial neural network regression models in a panel setting: Predicting economic growth |
| |
| Affiliation: | 1. Department of Management, Università Politecnica delle Marche, Ancona, Italy;2. Department of Economics and Social Science, Università Politecnica delle Marche, Ancona, Italy;1. Department of Financial Engineering, Ajou University, Suwon, 16499, Republic of Korea;2. Department of Applied Mathematics & Institute of Natural Science, Kyung Hee University, Yongin, 17104, Republic of Korea;3. Department of Mathematical Sciences, Seoul National University, Seoul, 08826, Republic of Korea;1. Graduate School of Economics, Kobe University, 2-1 Rokko-dai, Nada, Kobe, 657-8501, Japan;2. Institute of Social and Economic Research, Osaka University, 6-1, Mihogaoka, Ibaraki, Osaka, 567-0047, Japan;1. Texas A&M University, Department of Finance, Mays Business School, College Station, TX, 77843, USA;2. University of Valladolid (Spain), NRU Higher School of Economics (Russia), School of Business and Economics, Avda. Valle Del Esgueva 6, 47011, Valladolid, Spain;3. University of Valladolid, School of Business and Economics, Avda. Valle Del Esgueva 6, 47011, Valladolid, Spain |
| |
| Abstract: | Economic time series often feature non-linear structures such as non-linear time trends, non-linear autoregressive effects, and non-linear interaction effects. In this paper, it is shown that artificial neural network regression models are suitable tools for the analysis of economic panel data because they allow for a compromise between the ability to model these features and the model size. As model specification is a concern in artificial neural network models, previous approaches are discussed critically. It is shown that the growth rates of the gross domestic product of 24 industrialized economies in the period 1992–2016 follow a non-linear time trend which cannot be explained by autoregressive features or polynomial time variables. The unrestricted functional form of the time trend in the artificial neural network model is also the main reason for the superior statistical performance compared to conventional panel models. This is confirmed by out-of-sample predictions for 2017. |
| |
| Keywords: | Neural networks Forecasting Panel data C45 C53 C61 O40 |
| 本文献已被 ScienceDirect 等数据库收录！ |
|
