Targeted Estimation in Heterogeneous Panel Data Models
Predicting more accurately by learning from comparable units
Data scientist Georgia Banava investigated how statistical predictions for specific units, such as a single country or a specific region, can be improved. While standard econometric models often calculate an average effect across all available data, Banava developed three new methods that focus specifically on one particular unit. This makes it possible, for example, to predict the GDP of the Netherlands very precisely, without leaving valuable data from the rest of Europe unused.
Banava's research demonstrates that 'borrowing' information from comparable units leads to much more reliable, accurate, and stable estimates. Instead of analyzing each unit separately or simply averaging everything, her methods make optimal use of all information while carefully accounting for differences between them. This approach offers major practical benefits: for instance, hospitals can better evaluate the effectiveness of treatments per patient group, and smaller regions with limited data can still make reliable predictions regarding unemployment or economic growth by learning from comparable regions.
More information on the thesis
