Firm networks and tax compliance: experimental evidence from Uganda

How do policy interventions diffuse through firm transaction networks? We design a novel two-stage randomization strategy that assigns a tax enforcement treatment at the seller–buyer link level and ensures separation within the network to identify direct and spillover effects. Using Ugandan transaction-level VAT data, we find that treated links correct 23.8% of reporting discrepancies, fourteen times the control rate. Corrections are driven by sellers—even when only buyers receive letters—providing evidence of communication between firms. Spillovers extend to other transactions, with persistent improvements in post-treatment reporting. Sellers evade by reclassifying firm-to-firm transactions as unverifiable final sales, weakening the VAT’s self-enforcing property.