We develop a programming algorithm that predicts a balanced-panel mix-adjusted house price index for arbitrary spatial units from repeated cross-sections of geocoded micro data. The algorithm combines parametric and non-parametric estimation techniques to provide a tight local fit where the underlying micro data are abundant and reliable extrapolations where data are sparse. To illustrate the functionality, we generate a panel of German property prices and rents that is unprecedented in its spatial coverage and detail. This novel data set uncovers a battery of stylized facts that motivate further research, e.g. on the density bias of price-to-rent ratios in levels and trends, within and between cities. Our method lends itself to the creation of comparable neighborhood-level qualified price and rent indices for residential and commercial property.