FORECASTING HOUSING PRICE INDICES WITH MACROECONOMIC VARIABLES: EVIDENCE FROM SEOUL'S 25 DISTRICTS¹

We investigate whether pure macroeconomic variables can forecast district-level housing price indices in Seoul without using local information, building-specific, or historical price data. Comparing six forecasting models across 25 districts using a rolling window framework, we find that Lasso regression outperforms complex machine learning algorithms, including XGBoost and Random Forest. Lasso selects only five to nine macroeconomic variables from sixteen candidates to forecast each district’s housing index, with M2 money supply and household credit emerging as the dominant predictors across all districts. Our findings reveal that the local housing market indices follow simple linear relationships with macroeconomic fundamentals. The heterogeneity in district-level sensitivities implies that uniform monetary policies create highly uneven spatial effects. These results suggest that accurate housing market forecasting may not require sophisticated nonlinear models but proper identification of district-specific sensitivities to core macroeconomic variables.