Forecasting Single Family House Prices in the US using GMDH

Abstract

The recent volatility in the housing market due to economic downtrend and supply chain disruption in large extent necessitates a forecasting model that can give an accurate outlook of the market trend for the benefit of accurate cost estimation and decision making. To solve the problem, the use of a machine learning technique called group method of data handling (GMDH) algorithms were explored to forecast house price index (HPI); an indicator that measures changes in single-family home values based on data from all 50 states and over 400 American cities, for the period of March 2006–January 2023. The goal of this study is to construct an accurate mathematical model which will not only contribute to the forecasting of house prices trend, but also add to the body of knowledge by rigorous investigation of macroeconomic indicators with significant relationships for more accurate prediction. Results indicate that the GMDH neural network algorithm recorded the lesser MAPE of 1.34% and 6.52% and from the last 20% testing data among both univariate and multivariate model, respectively. Group method of data handling neural network algorithm also responded sensitively and swiftly to quick, large changes in economic conditions. This study presents models and mathematical formulas that can accurately forecast HPI in the US.

Keywords: HPI, Forecasting, Macroeconomic Indicators, GMDH, Neural Network, Combinatorial Algorithm

JEL Classification: R,C