Tell Me Who Your Neighbors Are and I Will Tell You Your Informal Economy Size: The Case of Sweden

Estimating the size and dynamics of the informal economy (IE) remains a persistent challenge, while consistent in its social and economic impact. Multiple studies tackled the temporal dynamics of IE in different timeframes, countries, and parameter spaces, providing ever-increasing accuracy. Nonetheless, these models neglect the spatial component of the IE dynamics. To this end, in this study, we explore the usage of a wide range of models, from linear regression to graph neural networks, in different spatio-temporal settings. Using the data between 2006 and 2023 about Sweden’s 21 regions, we evaluate the performance of these models in IE temporal prediction, given different resolutions of spatial information. Moreover, we evaluated the usefulness of such spatial data on both the regional and country levels. Our results show that machine learning based models consistently outperform both linear regressions and deep networks at the regional level, while deep learning becomes most powerful when predictions are aggregated to the national scale. We further demonstrate that the influence of one region on another declines with geographic distance, and that including data from neighboring regions improves predictive accuracy statistically significantly. However, adding all regions yields only small additional gains, with LR not improving beyond neighbors, indicating a highly non-linear spatial relationship. These findings suggest that informal economic activity in Sweden is best understood through spatially aware, data-driven models, which can capture both local and national dynamics more effectively than traditional approaches.