Estimating Housing Costs in Switzerland Using Public Registers and Survey Data: A Machine Learning Approach

Housing costs are an important indicator of affordability and financial pressure on households. This project develops a housing cost estimation model for the canton of Aargau by combining survey data (Structural Survey 2019–2022) with public register data (e.g., population and buildings). To capture the relationship between predictor variables and monthly net rent we used random forest and an elastic net regression models. The random forest regression model achieved the better out-of-sample performance (RMSE = CHF 315, MAE = CHF 224, R2 = 0.55). The most important predictor was floor area, followed by building construction period and number of rooms. Model performance was stable across regions and dwelling sizes. The geographical variation of the monthly net rent corresponded to patterns known from surveys and market analyses. Robustness checks indicated that the results were stable across alternative model settings. These findings suggest that data from public registers and surveys can be effectively combined to estimate housing costs at scale. Since this approach relies on existing data, it avoids costly additional data collection and is therefore useful for public administration, research, monitoring, and planning. Since the necessary data are available for all Swiss cantons, the model can be applied nationwide.