I leverage the staggered rollout of VAT in Pakistan to document the role of withholding mechanism in the self-enforcement of a VAT. Focusing on firms already in the tax net, I see how their outcomes respond when the tax is extended upward to intermediates used by them. I find that the upward extension of VAT, which triggers the withholding mechanism, causes an immediate and large (more than 40 log points) surge in sales reported by firms. The evolution of bunching above the zero-liability point and of input costs reported by firms suggests that this large response is indeed driven by the withholding mechanism. I also explore the role of withholding in the extensive margin compliance choices of firms.

We examine two Pakistani programs to see if the public disclosure of tax information and social recognition of top taxpayers promotes tax compliance. Pakistan began revealing income tax paid by all taxpayers in 2012. Simultaneously, another program began recognizing and rewarding the top 100 tax- paying corporations, partnerships, self-employed individuals, and wage earners. We find that the public disclosure caused an increase of 9 log points and the social recognition program 17 log points in the tax payments of agents exposed to the program. Our results suggest that such programs can be important policy levers to mobilize additional resources.

Using a series of Pakistani tax reforms and administrative records, I document that taxable income responses induced by to-zero tax cuts are orders of magnitude larger than ones induced by similar-sized other cuts. This finding is remarkably robust to alternative specifications and holds for both the self-employed and wage earners. I explore salience, selective enforcement, and discontinuous evasion costs as explanations of the observed behavior. I find that the data favor the last explanation. The difference between the two sets of responses is primarily driven by a large, discrete tax evasion response, which is included in the former but not in the latter behavior. I estimate the difference as a lower bound on tax evasion, showing that at least 70% of the income of low- and middle-income self-employed and 1% of low-income wage earners goes unreported.