Hedonic regression models aim at decomposing the price of a dwelling in dependence of building characteristics, common amenities and households’ income distribution, usually by the use of the ordinary least squares (OLS). However, this method assumes spatial-invariant and linear relationships between rents and the covariates, which might be inappropriate when modelling real estate prices over space. Several methods have been introduced over the last decades to account for spatial and non-linear effects and throughout provided evidence for a significant decrease in the error forecasting variance. This paper tests thus the prediction accuracy and asymptotic properties of two innovative methods proposed along the hedonic debate: The geographically weighted regression (GWR) and the generalized additive model (GAM). Based on a sample with more than 500’000 observations across German residential markets, the results show that OLS estimates fail at explaining rents due to skewed errors, whereas the proposed methods lead to a significant increase in the forecasting performance. The paper provides furthermore evidence that rents across Germany do respond to spatial and non-linear effects, remarking potential value increases in the valuation of institutional residential portfolios.