This file is part of IDEAS, which uses RePEc data


[ Papers | Articles | Software | Books | Chapters | Authors | Institutions | JEL Classification | NEP reports | Search | New papers by email | Author registration | Rankings | Volunteers | FAQ | Blog | Help! ]

Neural Networks for Macroeconomic Forecasting: A Complementary Approach to Linear Regression Models

Author info | Abstract | Publisher info | Download info | Related research | Statistics
Author Info
Steven Gonzalez

Additional information is available for the following registered author(s):

Abstract

In recent years, neural networks have received an increasing amount of attention among macroeconomic forecasters because of their potential to detect and reproduce linear and nonlinear relationships among a set of variables. This paper provides a highly accessible introduction to neural networks and establishes several parallels with standard econometric techniques. To facilitate the presentation, an empirical example is developed to forecast Canada's real GDP growth. For both the in-sample and out-of-sample periods, the forecasting accuracy of the neural network is found to be superior to a well-established linear regression model developed in the Department, with the error reduction ranging from 13 to 40 per cent. However, various tests indicate that there is little evidence that the improvement in forecasting accuracy is statistically significant. A thorough review of the literature suggests that neural networks are generally more accurate than linear models for out-of-sample forecasting of economic output and various financial variables such as stock prices. However, the literature should still be considered inconclusive due to the relatively small number of reliable studies on the topic. Despite these encouraging results, neural networks should not be viewed as a panacea, as this method also presents various weaknesses. Contrary to many researchers in the field, who tend to adopt an all-or-nothing approach to this issue, we argue that neural networks should be considered as a powerful complement to standard econometric methods, rather than a substitute. The full potential of neural networks can probably be exploited by using them in conjunction with linear regression models. Hence, neural networks should be viewed as an additional tool to be included in the toolbox of macroeconomic forecasters.

Depuis quelques années, les prévisionnistes macro-économiques ont de plus en plus recours aux réseaux neuronaux en raison de leur capacité de repérer et de reproduire des relations linéaires et non linéaires dans un ensemble de variables. Le présent document se veut une introduction conviviale aux réseaux neuronaux, qui dresse des parallèles avec les techniques économétriques les plus répandues. Pour mieux illustrer nos propos, nous avons procédé à l’élaboration d’un exemple pratique sur les prévisions de croissance réelle du PIB au Canada. Qu’il s’agisse d’une période échantillonée ou non, on constatera que l’exactitude des prévisions du réseau neuronal est supérieur à celle du modèle de régression linéaire que le Ministère a élaboré et institué, avec diminution de la marge d’erreur, qui peut fluctuer entre 13 et 40 pour cent. Divers essais ont cependant permis d’établir que des prévisions plus exactes n’ont pas grande incidence sur le plan statistique. Une révision approfondie de la littérature nous a permis de conclure que les réseaux neuronaux sont dans l’ensemble plus exacts que les modèles linéraires pour la prévision des extrants économiques non échantillonés et pour certaines variables financières, tel le cours du marché. Cela dit, on ne saurait accorder une importance décisive à ce qui a été écrit en la matière, étant donné que les études fiables sont peu nombreuses pour le moment. Malgré des résultats encourageants, les réseaux neuronaux ne doivent pas être pris pour la panacée, car cette méthode présente également des faiblesses. Contrairement à nombre de chercheurs dans le domaine, qui ont tendance à adopter une approche absolutiste (soit tout, soit rien) à ce sujet, nous avançons que les réseaux neuronaux doivent être considérés comme un puissant complément aux méthodes économétriques en usage, plutôt que comme un substitut. Pour tirer le plus grand parti des réseaux neuronaux, il faudrait s’en servir en conjonction avec des modèles de régression linéaire. En somme, les réseaux neuronaux doivent être perçus comme un outil d’appoint à placer dans la boîte à outils des prévisionnistes macro-économiques.

Download Info
To download:

If you experience problems downloading a file, check if you have the proper application to view it first. Information about this may be contained in the File-Format links below. In case of further problems read the IDEAS help page. Note that these files are not on the IDEAS site. Please be patient as the files may be large.

File URL: http://www.fin.gc.ca/scripts/Publication_Request/request2_e.asp?doc=wp2000-07e.pdf
Our checks indicate that this address may not be valid because: 500 Internal Server Error. If this is indeed the case, please notify (Gustavo Durango)
File Format:
File Function:
Download Restriction: no

Publisher Info
Paper provided by Department of Finance Canada in its series Working Papers-Department of Finance Canada with number 2000-07.

Download reference. The following formats are available: HTML (with abstract), plain text (with abstract), BibTeX, RIS (EndNote, RefMan, ProCite), ReDIF
Length:
Date of creation:
Date of revision:
Handle: RePEc:fca:wpfnca:2000-07

Contact details of provider:
Postal: 140 O'Connor St., Ottawa, K1A 0G5
Phone: 613-992-1573
Web page: http://www.fin.gc.ca/
More information through EDIRC

For technical questions regarding this item, or to correct its listing, contact: (Gustavo Durango) The email address of this maintainer does not seem to be valid anymore. Please ask Gustavo Durango to update the entry or send us the correct address..

Related research
Keywords:

This paper has been announced in the following NEP Reports:

Cited by:
(explanations, Please report citation or reference errors to , or , if you are the registered author of the cited work, log in to your RePEc Author Service profile, click on "citations" and make appropriate adjustments.)
  1. Aristizábal, María Clara, 2006. "Evaluación asimétrica de una red neuronal: aplicación al caso de la inflación en Colombia," Lecturas de Economia, UNIVERSIDAD DE ANTIOQUIA - CIE. [Downloadable!]
  2. María Clara Aristizábal Restrepo, . "Evaluación asimétrica de una red neuronal artificial:Aplicación al caso de la inflación en Colombia," Borradores de Economia 377, Banco de la Republica de Colombia. [Downloadable!]
  3. Haider, Adnan & Hanif, Muhammad Nadeem, 2007. "Inflation Forecasting in Pakistan using Artificial Neural Networks," MPRA Paper 8898, University Library of Munich, Germany. [Downloadable!]
  4. María Clara Aristizábal Restrepo, 2006. "Evaluación asimétrica de una red neuronal: aplicación al caso de la inflación en Colombia," Lecturas de Economía, Universidad de Antioquia, Departamento de Economía, issue 65, pages 73-116, Julio-Dic. [Downloadable!]
Statistics
Access and download statistics

Did you know? All the bibliographic data shown here has been contributed by volunteers, thereby helping to keep this service free.

This page was last updated on 2009-11-6.


This information is provided to you by IDEAS at the Department of Economics, College of Liberal Arts and Sciences, University of Connecticut using RePEc data on a server sponsored by the Society for Economic Dynamics.