Analysis of the Housing Market Dynamics Using NARX Neural Network

Authors

  • Daria Wotzka Faculty of Electrical Engineering, Opole University of Technology
  • Grażyna Suchacka Institute of Informatics, University of Opole
  • Faculty of Economics, University of Opole
  • Łukasz Mach Faculty of Economics, University of Opole
  • Marzena Stec Narodowy Bank Polski, Regional Branch in Opole
  • Joachim Foltys Humanitas University in Sosnowiec

Keywords:

ZHB Luzern

Abstract

This study employs a Nonlinear Autoregressive with eXogenous inputs (NARX) neural network to model the dynamics of the housing construction market in Poland, with a distinction made between segments of developers and individual investors. The dataset under analysis contains the 19-year data corresponding to the numbers of housing units approved for construction, under construction, and completed. The NARX model was calibrated thoroughly to suit unique characteristics of the data, with an emphasisputon the hidden layer size and delay parameters, to capture the estate market's nonlinear trends. Results show a very high efficiency of NARX models and highlight distinct patterns and dynamics in the housing completion, construction starts, and permit issuance between the two market segments. These variations are vital for understanding the distinct forces and trends shaping the developers’ and individual investors’ markets in the Polish housing sector. Findings of the analysis provide valuable insight into the nuanced functioning of these market segments.

Time and Location

This article covers a time period from to .  Temporal properties of the article's content specified with a day resolution.

Location(s) or Area(s)  The geometric shape(s) in blue represent the locations or areas of the article's content as provided by the authors. The administrative unit shown in a black rectangle represents the smallest administrative unit that encloses the spatial extent that the article is dealing with.

The administrative units enclosing the article's places are:   Here the administrative units are listed, which are superior to the location(s) the article is dealing with, with the highest level on the left and the lowest on the right.

Temporal and spatial metadata are published under the following license: CC-0

Downloads

Published

2025-07-14

Issue

Vol. 1 No. 1 (2024): geoMetadata Demo Issue Edition 1

Section

Articles

Download Geospatial Metadata

GeoJSON About GeoJSON

Geodata License: CC-0