Forecasting Real Estate Performance during the COVID-19 Pandemic Crisis: A Comparison of Statistical and Machine Learning Models

Abstract

Machine learning (ML) methods, such as long short-term memory (LSTM) models, are increasingly proposed as alternatives to traditional statistical approaches for time series forecasting. However, given the speed of the real estate industry in providing data that reflect economic climates, there are few comparisons of ML techniques with statistical methods in the context of real estate data during crisis periods. The study investigates the predictive accuracy of the autoregressive integrated moving average (ARIMA) and LSTM models by using daily data from the Financial Times Stock Exchange/Johannesburg Stock Exchange South Africa Listed Property Index. Through a comprehensive analysis of 1628 observations from January 2, 2015, to July 8, 2021, the study finds that the ARIMA models produce fewer forecasting errors compared to the LSTM models during the COVID-19 crisis. These findings suggest that traditional ARIMA models may be more efficient for forecasting volatile real estate data in crisis periods, although the results could vary with larger and more complex datasets. This research is crucial as it provides insights into the comparative performance of statistical and ML models, thus emphasizing the need for context-specific model selection in economic forecasting.

View PDF – https://doi.org/10.53383/100411