Forecasting House Prices

This article identifies the factors that drove house prices in 13 advanced countries over the past 35 years. It does so based on Breiman s (2001) random forest model. Shapley values indicate that annual house price growth across countries is explained first and foremost by price momentum, initial valuations (proxied by price to rent ratios) and household credit growth. Partial effects of explanatory variables are also elicited and suggest important non-linearities, for instance as to what concerns the effects of CPI inflation on house price growth. The out-of-sample forecast test reveals that the random forest model delivers 44% lower house price variation root square mean errors (RMSEs) and 45% lower mean absolute errors (MAEs) when compared to an OLS model that uses the same set of 10 pre-determined explanatory variables. Notably, the same model works well for all countries, as the random forest attributes minimal values to country fixed effects.

💡 Research Summary

The paper investigates the determinants of house‑price dynamics across 13 advanced economies over a 35‑year period (1988‑2019) using a machine‑learning approach—specifically Breiman’s (2001) Random Forest algorithm. The authors compile annual nominal house‑price indices from the Bank for International Settlements, yielding 374 country‑year observations. Ten macro‑financial and demographic predictors are assembled: price momentum (last year’s growth), price‑to‑rent ratio, household credit growth, GDP growth, CPI inflation, short‑ and long‑term interest rates, stock‑market volatility (VXO), and population growth.

Three modeling strategies are compared: (1) a naïve AR(1) benchmark that regresses next‑year growth on its own lag, (2) a linear OLS model that includes all ten predictors, and (3) a Random Forest ensemble of 500 regression trees with a minimum terminal node size of 10. Model performance is evaluated both in‑sample and out‑of‑sample using root‑mean‑square error (RMSE) and mean absolute error (MAE).

Results show that the Random Forest dramatically outperforms the linear benchmarks. In‑sample adjusted R² rises from 35.3 % (AR) and 44.0 % (OLS) to over 60 % for the forest. Relative to OLS, the forest reduces RMSE by 48.6 % and MAE by 53.3 %, confirming that non‑linearities and complex interactions are crucial for explaining house‑price movements. Variable‑importance analysis based on Shapley values identifies price momentum, price‑to‑rent ratio, and household credit growth as the top three drivers. Partial‑dependence plots reveal a non‑linear inflation effect: CPI inflation between 0 % and 3 % maximizes price growth, peaking around 3 %; beyond 5 % inflation, house‑price growth falls below the inflation rate, indicating that housing ceases to be an effective hedge in high‑inflation environments.

A temporal analysis of importance scores shows that, in the last decade, population growth and the yield‑curve (long‑term rates) have become more influential, reflecting greater financial integration and the increasing view of housing as an investment asset. When country fixed effects are introduced, the Random Forest essentially ignores the country identifier, suggesting that the model captures common macro‑financial mechanisms rather than country‑specific institutional nuances. Robustness checks—varying the set of predictors, tree depth, and sampling fractions—confirm the stability of the findings.

The study contributes to the housing‑price literature by providing a data‑driven quantification of macro‑financial drivers, demonstrating that machine‑learning techniques can substantially improve forecasting accuracy, and showing that the “black‑box” perception of such models can be mitigated through Shapley‑based importance and partial‑effect visualizations. The authors argue that the Random Forest framework offers policymakers and market participants a powerful analytical tool for monitoring housing‑market risks and for designing more effective macro‑prudential policies.