Research

WORKING PAPERS

Who Should the Taxman Visit? Evidence from Door-to-Door Tax Enforcement in Indonesia

Job Market Paper - Latest Version
(with Muhammad Khudadad Chattha)

Abstract: We investigate the effects and optimal design of a large-scale tax-nudging campaign in Indonesia, where officials conducted door-to-door visits to more than 30,000 property owners.To estimate the program the program’s average and heterogeneous impact on property-tax compliance we propose a modified causal forest estimator. We find that visits increased payment rates by an average of 7.8 p.p., with effects persisting for at least four subsequent years and being particularly pronounced among lower-value properties. We then use individualized treatment-effect estimates—rather than conventional compliance risk scores—to design targeting rules that answer “Who should the taxman visit?” The optimal targeting depends on a prioritization rule induced by policy objective:A participation-maximizing rule, which seeks to induce the largest number of taxpayers to comply, prioritizes lower-value properties and delivers meaningful gains in both participation and revenue. A revenue-maximizing rule directs visits toward higher-value properties, generating substantial fiscal returns with little change in the number of payers. By contrast, a standard risk-based rule that targets likely noncompliers is effectively neutral—allocating effort broadly across the value distribution. However it yields no discernible improvements in either revenue or participation. These results suggest that risk-based prioritization is ill-suited to nudging contexts and make explicit the equity–efficiency trade-offs inherent in algorithmic targeting for tax enforcement.

Key Words: Tax Compliance, Behavioral Nudges, Causal Forest, Policy Learning

Low-Cost Reminders and Property Tax Compliance: Evidence from a Field Experiment

(with Muhammad Khudadad Chattha, Naranggi Pramudya Soko and Prabaning Tyas)
Latest Version

Abstract: This study examines the impact of low-cost digital nudges sent via WhatsApp on property tax compliance in Gorontalo, Indonesia. In a randomized controlled trial involving 801 delinquent taxpayers, individuals received either (i) a soft-tone message emphasizing civic duty, (ii) a hard-tone message, or (iii) no message (control). Four key findings emerge. First, despite referencing overdue taxes, the positive message substantially boosted current-year compliance by 9.9 percentage points. Second, the soft-tone message completely eliminated the compliance gap between high- and low-compliance individuals. Third, message framing mattered: the soft-tone message consistently outperformed the penalty message in the long-term. While the hard-tone message led to short-term increases in compliance, the effect dissipated in the long-term due to intertemporal substitution. Finally, the intervention has potential for scalability, we calculate that the marginal cost per additional tax bill paid is USD 0.126. These findings underscore the potential for such behaviorally informed messaging to improve tax collection in low-capacity settings.

Key Words: Taxation in Developing Countries; Field Experiment; Microeconomic Policy

The Mechanics of Deforestation Electoral Cycles in Brazil

(with Marcelo Goncalves)
Draft available upon request

Abstract: Empirical evidence shows that deforestation rates fluctuate with political cycles: There is positive correlation between election periods and increased deforestation. However the mechanism the mechanism remains uncertain. We propose an “administrative hypothesis” in Brazil, suggesting that political leaders may strategically manipulate the assignments of environmental agency staff—IBAMA—to modulate enforcement intensity and gain voter support. Brazil’s alternating federal and local elections every two years provide a unique setting to examine how political alignment with the federal government impacts deforestation. Our analysis of close electoral races indicates that alignment with the federal government is associated with increased deforestation. In the current phase, we are investigating the characteristics of local environmental agency heads to determine if the selection mechanisms in public service relate to more intense deforestation cycles.

Key Words: Political Cycle; Deforestation; Personnel Economics of the State; Selection into Public Sector

WORK IN PROGRESS

Retirement Security and Reproductive Choices: Causal Evidence from Brazilian Pension Reforms

Slides available upon request

Abstract: The “old-age security hypothesis” suggests that, in environments with limited formal social insurance, children partly serve as a form of old-age security: higher fertility helps diversify income risk and secure support in old age. This mechanism should be especially relevant in economies where social protection systems are weaker. Exploiting two quasi-experimental pension reforms that altered expected retirement benefits while leaving current income unchanged, the project seeks to identify whether these exogenous shifts in future income led to reductions in fertility. The results may have important policy implications: as developing countries expand contributory and pay-as-you-go pension systems, benefit formulas often rely on assumed fertility rates. If fertility declines in response to pension expansion, these systems may become financially fragile, with contribution bases shrinking relative to promised benefits. (The project is currently in the final stages of establishing the data-sharing agreement with the Brazilian goverment.)

Key Words: Pensions; Fertility; Social Insurance; Brazil

Algorithmic Design of Foster-Care Placements: Evidence from Brazilian Child-Protection Data

(with Yuri Costa)
Slides available upon request

Abstract: This leverages on a novel rich administrative dataset from Brazil’s foster-care and child-protection systems to study how institutional design and decision-making rules shape children’s trajectories. Brazil has a large and heterogeneous system of institutional and family-based care, with substantial regional variation in capacity, implementation quality, and the interaction between child-protection services, courts, and social assistance. Despite a sizable pool of prospective adoptive parents, there is a striking mismatch: adults wait on average more than two years to adopt a child, while children wait roughly one and a half years to be adopted. These delays are partly driven by institutional rules requiring repeated reassessments of family risk and complex, multi-stage case processing.

We are currently investigating whether initial triage and risk-assessment data are informative enough to predict key policy decisions early in the process—such as reunification, long-term foster care, or adoption—and how alternative decision rules would affect child welfare and time to stable placement. Using tools from causal inference and market design, we simulate counterfactual scenarios in which (i) human decisions are augmented by algorithmic decision-support tools and (ii) certain stages of the process are partially automated by algorithms. The goal is to quantify the trade-offs between speed, stability, and fairness in matching children to families, and to inform the design of ethically grounded, data-driven reforms to foster-care and adoption systems.

Key Words: Foster Care; Child Protection; Market Design; Algorithmic Decision-Making; Administrative Data

Optimal Stop in Auditing: Estimating a Dynamic Tax Auditing Strategy

Slides available upon request

Abstract: Building on the foundational work of Rust (1987), we propose a dynamic discrete-choice (DDC) model to determine the optimal frequency of tax audits. We frame the auditing decision as an optimal stopping problem, balancing two key considerations: (i) the high fixed costs of conducting audits—while auditing incurs substantial expenses, successful detection and recovery of tax evasion has time persistent effects, and (ii) forgoing audits, tax authorities avoid immediate costs but may face increased evasion over time.

Key Words: Dynamic Discrete Choice; Taxation; Auditing

Tax Rebate Policy Evaluation in Pakistan

(with Muhammad Khudadad Chattha and Cem Ozdemir)
Early-stage work in progress

Abstract: This project evaluates the impact of tax rebate policies on payment behavior and revenue in Pakistan, using administrative tax records and staged policy roll-outs.

Key Words: Tax Incentives; Compliance; Developing Countries; Public Finance