Spillover Effects of Tax Enforcement on Production Networks: Evidence from Paraguay
How to cite
Best, Michael; Grosset-Touba, Florian; Pierri, Gastón; Sadler, Evan; Toulis, Panos. 2025. "Spillover Effects of Tax Enforcement on Production Networks: Evidence from Paraguay." AEA RCT Registry. AEARCTR-0016101. https://doi.org/10.1257/rct.16101-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
Our project focuses on understanding how optimal targeting rules may change when enforcement activities have spillovers through production networks. The theoretical part of our project develops a tractable network model in which to study tax enforcement. The model-predicted optimal targeting rules will depend on a number of features of the production network and on a number of assumptions in the model. The experimental part of our project studies these questions empirically.
In partnership with Paraguay's tax authority, the Dirección Nacional de Ingresos Tributarios (DNIT), we are doing a series of three interlinked experiments. We developed a scalable tax enforcement intervention, building on the work of Carrillo et al (2017) in Ecuador. The first experiment studied the direct effects of the intervention on the targeted taxpayers. The second experiment is described in this pre-registration. The third experiment, planned for July/August 2025 will use the findings from the first two experiments and our theoretical model to derive and then experimentally test alternative tax enforcement targeting rules.
Intervention
Our intervention is a modified version of the intervention studied by Carrillo et al (2017) in Ecuador. Among the eligible taxpayers, taxpayers were randomly selected to be sent a notice. The notice presents taxpayers with a summary by month of their reported sales, the purchases reported by their clients, and the discrepancies between them. The invoices underlying the purchases reported by clients are attached as an annex to back up the discrepancies. The notice then requests that the taxpayer file amendments of the relevant tax returns to address the inconsistencies.
We made two modifications to the notifications studied by Carrillo et al (2017). First, we strengthened the language used. Specifically, the notices emphasized that the request was for the taxpayer to file an amendment to their tax return in which they increase their reported sales to account for the discrepancy, but not make any other amendments to their return. This was intended to avoid taxpayers simultaneously increasing their reported purchases. Second, the notices were accompanied by greater follow-up and sanctions for non-response. In particular, auditors were assigned to follow up on each case. If the taxpayer responded but did not amend or amended only partially, the auditors reviewed the case and made a determination about whether to accept the partial response. If the taxpayer did not respond by the 10-day deadline, or responded unsatisfactorily, the auditor blocked the taxpayer's ID ("RUC"). Blocked IDs are not able to request more tax invoices, and are unable to perform a range of other processes with the tax authority (though they are still able to file and amend returns), potentially disrupting business activities for non-compliant taxpayers, especially larger ones that rely more on being able to issue tax invoices and have more complicated interactions with the tax authority.
Experimental design
The randomization protocol is described in detail in the pre-analysis plan. In brief, the randomization occurs in two stages amongst the eligible pool of 62,344 taxpayers.
In stage 1 we select focal units and then randomly assign their clients and/or suppliers to treatment.
In stage 2 we focus on estimating the direct effect of the notices by focusing on anticliques: taxpayers who are not connected to other eligible taxpayers and who are not connected to any of the focal taxpayers from stage 1. In the anticliques we perform stratified random assignment to treatment, stratifying by size, degree, and baseline amendment propensity.
Full details of the procedure are available in the (public) pre-analysis plan included with this registration.
Design details
Cited by
No citation data available for this registration yet.