I regard myself an engineer with extensive experience in management, however I wanted to explore the world of research. What could be better than taking the plunge into Economics?. My humble contribution has been to bring together two distant worlds: computation and econometrics—particularly machine learning and classical statistics.
You can find the original work here, or read brief summary of it below:
The major real estate crisis at the end of the 2000s brought about profound changes in the Spanish housing market, including a significant increase in demand for rental housing. A near-total halt in construction, combined with population growth, led to an imbalance between limited supply and growing demand—particularly noticeable in densely populated areas—resulting in a sustained rise in prices in large cities. Housing markets display unique, idiosyncratic behavior in each geographic area, which requires a high level of detail for proper study. Additionally, the increasing dynamism of these markets demands continuous and up-to-date monitoring. Current research tools, especially rental price indexes, are insufficient for detailed analysis. This is often due to a lack of granular data sources or because available information is published with significant delays.
This research addresses the challenge of constructing a rental price index with broad geographic and functional granularity, enabling updates in near real-time. To achieve this, the thesis combines machine learning techniques with traditional econometric models, utilizing both official sources and open data.
The thesis is structured into three cohesive and progressive sections. The first section lays the foundation by introducing the historical and contemporary context of the rental housing market, and reviewing the methodologies and data sources used in the analysis. Special attention is given to processes for cleaning and integrating data, eliminating errors and biases, and linking different sources of information. The second section gets into the technical aspects, focusing on the design and calibration of hedonic price models. Here, factors such as location, building characteristics, and indicators reflecting the real estate market’s dynamics play a central role. Building on this theoretical and methodological foundation, the third section presents two chapters: the first dedicated to temporal disaggregation of time series, and the second to empirical analysis in the case of the Community of Madrid._
Our experimental analysis confirms that the proposed methodology enables the construction of a price index based on a combination of national accounts and open data, ensuring consistency with both. The index also achieves the desired levels of granularity and update frequency, and demonstrates its ability to anticipate short-term market trends.