Ceny bytů v Česku : stratifikovaná prostorová analýza

This thesis develops and estimates a two-level Spatial Durbin Model (SDM) for apartment prices across the Czech Republic, extending the Hierarchical Spatial Autoregressive Model of Dong and Harris (2014)in two principal directions: the inclusion of the Durbin spatial lag on exogenous covariate matrix and the adoption of MLE estimator instead of MCMC. The model uses the H3 hexagonal lattice methodology to aggregate covariates on the regional level. Extensive sensitivity analysis is conducted to establish the most fitting spatial weights configurations before fitting the final model. Sensitivity to alterations of spatial weight matrices on both levels of the model is evaluated. The best performing combinations of specifications all prefer more locally concentrated weight matrix and a coarser resolution for the hexagonal lattice, with an exponential decay specification at resolution 5 emerging as the overall best-fitting model. The estimates are interpreted using the spatial impacts decomposition and show both their own-location effects and the indirect effect affecting their neighborhoods. Among strong price determinants based on these impacts are the apartment size, disposition type, renovation status, ownership structure or energy performance certificates. Monte Carlo simulation study is run to confirm reliable parameter recovery under realistic conditions. During the simulation, a practical lower bound of active hex cells needed for the identification of the spatial autoregressive parameter at the neighborhood-level was found.