Which tax audits should be increased? Evidence from Italy using a machine learning approach

Abstract In this paper, we combine a machine learning approach and causal inference to analyze the dynamic effects of tax audits on a large sample of self-employed individuals and sole proprietors from Italy. We identify a plausible subset of audit criteria, and we use them to match audited with unaudited taxpayers within a coarsened exact matching approach. We show that tax audits have a short-run positive impact on subsequent reports and that this effect is likely to justify an increase in the number of audits with respect to observed levels. However, we also show that this effect is almost exclusively due to a decrease in reported costs and that it is heterogeneous across different audit types, sectors and accounting regimes.