Decoding the spatial dynamics of sales and rental prices in a high-pressure Portuguese housing market: A random forest approach for the Lisbon Metropolitan Area

Sales and rental prices were analysed at parish level using random forest regression for the Lisbon Metropolitan Area. Three dependent variables (new sales, new rents, and all rents) and a set of independent variables/associated factors were used, including location, building/dwelling characteristics, socioeconomic features, and tourism. This geographically-based approach aims not to predict housing prices, but to identify relevant factors associated with sales/rents, ranking their importance. The temporal dimension is also explored by comparing new and all existing rents. The results revealed similarities and differences between housing submarkets. New sales and new rents had similar spatial patterns and dynamics but were different from that of all rents, with different regulations over time. Strong associations were found between the dependent variables and the population's social status and urban quality. However, while location was more strongly related to new sales and new rents, revealing a greater dependence on the current dynamics of the housing market, socioeconomic features were more closely related to all rents, expressing the urban and demographic dynamics of recent decades. Different associated factors prevail inside and outside the Lisbon municipality. The results contribute to a better understanding of housing submarkets and the relationships between sales/rents and associated factors. • Random forest models identify factors relevant to housing prices at parish level. • Socioeconomic features help explain the spatial patterns of housing prices. • Relevant factors change depending on housing submarkets and study areas. • The prices of new sales/new rents are more associated with location and land value. • The prices of all rents are more related to socioeconomic features.