Title: A multi-parametric simulation study of neural networks' performance for nonlinear data against linear regression analysis in economics

Authors: Evangelos Sambracos; Marina Maniati; Sokratis Sklavos

Addresses: Department of Economics, University of Piraeus, 80, M. Karaoli and A. Dimitriou St., Piraeus, Attica, GR 18534, Greece ' Department of Economics, University of Piraeus, 80, M. Karaoli and A. Dimitriou St., Piraeus, Attica, GR 18534, Greece ' Research Synelixis Lab, CharilaouTrikoupi 45, Ahens, Attica GR 10681, Greece

Abstract: Different mathematical and dynamic methods have been developed addressing the problem of forecasting, with the regression analysis to be one of the most frequently used statistical procedures. Meanwhile, neural networks (NNs) are considered to be well suited in finding accurate solutions in an environment characterised by volatility, noisy, irrelevant or partial information. In this chapter, a simulation study compares the performance of NNs against linear regression analysis is based on multiple combinations (421 in total) of five different factors providing those cases that the NN performs better than the LRM and defining the output bias as the main contributor to the NN outcome.

Keywords: artificial neural networks; regression analysis; bias.

DOI: 10.1504/IJBFMI.2020.109256

International Journal of Business Forecasting and Marketing Intelligence, 2020 Vol.6 No.1, pp.17 - 31

Received: 05 Jul 2019
Accepted: 13 Oct 2019
Published online: 02 Sep 2020 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article