Methods for the reconstruction of economic networks from aggregate information are important for modeling, analysis and policymaking. The traditional Fitness Model (FM) achieves good performance by using node-specific variables that are easily accessible (e.g., GDP for countries or total assets for banks or firms) and the overall link density as the only sufficient statistics. However, it often ignores additional contextual or mesoscopic features which may be more difficult to observe. Here we extend the framework by incorporating block structure as in the Fitness-Corrected Block Model (FCBM), which allows for heterogeneous densities within and across blocks, but in the deliberately more challenging setting where such block-specific densities are not empirically available. Our method uses a Jeffreys prior to provide an unbiased distribution of all compatible solutions and associated entropies, that are otherwise left unidentified, and selects a single reconstruction at the median of this entropy distribution to compensate for the missing sufficient statistics. We evaluate the procedure on nine international-trade networks using GDP as node fitness and World Bank regions as an externally specified partition. The method systematically outperforms the baseline Block-Agnostic FM (which uses the same observed network input statistic) and sometimes even the FCBM (which always uses more information). Controlled synthetic networks confirm that the median is the best of four entropy-based representatives for a range of densities and regional effects similar to the empirical ones, while other representatives can perform better when parameters are pushed far from their realistic range. Overall, our method compensates for missing sufficient statistics via a median-entropy estimate that is empirically very robust within a realistic range of parameters.
翻译:暂无翻译