Fraud Detection Under Limited State Capacity: Experimental Evidence From Senegal

Tax administrations in low-income countries face widespread tax evasion and high enforcement costs. They thus need information to detect where tax evasion is most severe, and allocate scarce resources accordingly. This paper shows that leveraging large firms’ trading network to collect information about their suppliers is a cost-efficient way to detect tax evasion and increase future audit returns. We collaborate with the Senegalese tax administration on a vast data collection effort to digitise lists of payments submitted by the largest firms and show that 88.6% of these firms provide incomplete information about their suppliers. This prevents any cross-checking against income declared by the suppliers themselves. We then randomise a low-cost communication campaign across all 3,487 misreporting firms, to discourage future misreporting. The intervention increases the prevalence of suppliers’ identification information by 52%. In aggregate, this allows to uncover $145.5 million in unreported revenue (i.e. 0.5 % of GDP). Most of it accrues to a few tax-registered suppliers, as opposed to informal ones. A simulation exercise shows that exploiting the newly available information to target the largest under-reporting suppliers would increase audit returns by at least 100%.