Forecasting Real Estate Prices with US Market Data Using Machine Learning Methods

The increased frequency of financial crises underscores the necessity for refined forecasting models to empower risk management teams in preempting and addressing imminent shocks. Leveraging more frequent data offers a promising avenue for improved forecasting accuracy, even within less liquid assets like real estate. This study explores the feasibility of employing weekly residential real estate price indices in the USA to predict future values using machine learning techniques, alongside assessing the impact of macroeconomic variables. Three key objectives were pursued: identifying suitable data sources, determining applicable machine learning methods, and evaluating model performance. Findings reveal the viability of utilizing weekly data for predicting real estate markets, particularly beneficial for lower liquidity markets. The study suggests that weekly regional indices can greatly aid real estate analytics and assist businesses in refining risk management strategies and decision-making processes. However, the efficacy of the presented methodologies relies on datasets with substantial time series length and frequency, cautioning against their application in datasets with limited data volume where traditional time series analysis may be more appropriate.