Economist @ Customer Behavior Analytics, Amazon
I have recently started my career as an Economist at Amazon.
In my academic life, I was a development economist with interests in public economics and political economy related issues in India. I aimed to use large-scale government data sets (that have only recently begun to be collected) to better understand government capacity and to combine such data sets with field interventions to address questions of first-order causal interest.
I was also a postdoc at the UC Berkeley School of Information under Joshua Blumenstock. Here I was using cellular trace data from Afghanistan to study the impact of conflict.
Fields: Development, Public Economics
“I not only use all the brains that I have, but all I can borrow.” - Woodrow Wilson
(My most generous lender!)
- Phone+1 734 780 1120
(with Aprajit Mahajan)
A key stated advantage of the value-added tax (VAT) is that it allows the tax authority to verify transactions by comparing seller and buyer transaction reports. However, there is little evidence on how these paper trails actually affect VAT collections particularly in low compliance environments. We use a unique data set (the universe of VAT returns for the Indian state of Delhi over five years) and the timing of a policy that improved the tax authority's information about buyer-seller interactions to shed light on this issue. Using a difference-in-difference strategy we find that the policy had a large and significant effect on wholesalers relative to retailers. We also document significant heterogeneity with almost the entire increase being driven by changes in the behavior of the largest firms. We also find suggestive evidence that information and enforcement are complementary. Finally, we discuss the details of the policy implementation and argue that this policy which seems simple in principle, faces substantial hurdles in execution, particularly in a system with limited resources.
(with Jan Luksic)
Value-added tax systems across the world are afflicted with size-dependent regulations. The benefit of such regulations to the tax authority is unclear. In this paper, we use an administrative dataset from the state of Delhi in India to first show that a policy which mandated different frequencies of filing based on self-reported turnover resulted in bunching of firms below the thresholds at all levels. Using the subsequent change in these reporting policies, we provide evidence that such sharp bunching indeed occurs due to the VAT reporting frequency thresholds. We document that such bunching partly occurs due to turnover shifting and underreporting, provide evidence that the observed bunching has no growth consequences for the bunching firms - and find that bunching occurs to similar degree across firm types. Second, we calculate the VAT revenue losses due to such bunching. Third, the subsequent withdrawal of the policy allows us to show that in a regime with size-dependent reporting requirements, more frequent reporting is not associated with greater VAT collection. Finally, according to our back of the envelope welfare analysis, the sized-based filing policy is welfare improving if a welfare-maximizing government’s objective function assigns important weights to small- and medium-sized enterprises.
Who's Bogus? Learning to Identify Fraudulent Firms from Unbalanced and One-side Labelled Tax Returns Data
(with Aprajit Mahajan, Ofir Reich)In Proceedings of the 1st Annual ACM SIGCAS Conference on Computing and Sustainable Societies, COMPASS 2018
We apply a Random Forest classifier on Value Added Tax (VAT) returns from Delhi, India to increase tax compliance by identifying shell firms which can be further targeted for physical inspections. We face a nonstandard applied ML scenario. First, one-sided labels: firms that are not caught as shell are of unknown class: fake or legitimate, and we need to train as well as make predictions on them. Second, multiple time-periods: each firm files several periodic VAT returns but its class is timeless so prediction needs to made at the firm, not firm-period, level. Third, point in time simulation: we estimate the revenue saving potential by simulating the implementation of our system in the past. We do this by rolling back the data to the state of knowledge at a specific time and calculating the revenue impact of catching the fake firms.
- With Joshua BlumenstockMonitoring Conflict with Cellular Trace Data: Evidence from Afghanistan
- With Aprajit Mahajan, Ofir ReichBulding State Capacity to Better Use Administrative Data in India
March 2020 (ongoing)
Postdoctoral ScholarSchool of Information, UC Berkeley
Ph.D. in EconomicsUCLA Anderson School of Management
Awards: Dissertation Year Fellowship (2017-18), UCLA Graduate Division
Master of Public PolicyUniversity of Chicago
Awards: Dean’s Scholarship (2011-2013), J.N. Tata Scholar (2011), K.C. Mahindra Scholar (2011)
Bachelor of EngineeringUniversity of Delhi
Specialization: Computer Engineering
Ideas for India, October 18 2017
Value Added Tax 2.0.
- IGC"Building State Capacity to Better Use Administrative Data in India" (£25,907)
- JPAL-GI"Improving State Response to Public Grievances" ($50,000)
- JPAL-GI"Improving the Efficacy of Public Procurement and Public Grievance Monitoring" ($7500)
- EDI"Who is Bogus? Catching fraudulent firms in Delhi" (£22,000)
- JPAL-GI"Information Provision and Participatory Budgeting:Mohalla Sabhas in Delhi" ($49,050)
- JPAL-GI"Improving Public Service via the Ballot Box: Evidence from Delhi" ($5,000)
- IGC "Where’s Value? Using VAT data to Improve Compliance" (£50,830)
- Managerial Economics (MBA)Prof. Romain Wacziarg (2016-17); Prof. Paola Giuliano (2014-15, 2015-16)
- International Studies: India (MBA)Prof. Romain Wacziarg (2015-16, 2017-18)
- Impact Creation and Evaluation (MBA)Prof. Bhagwan Chowdhry (2016-17)
Prof. Romain Wacziarg
Prof. Aprajit Mahajan
Dept. of ARE, UC Berkeleyaprajit[at]berkeley[dot]edu
Prof. Joshua Blumenstock
School of Information, UC Berkeleyjblumenstock[at]berkeley[dot]edu
Get in Touch
Get in Touch
Sed eleifend sed nibh nec fringilla. Donec eu cursus sem, vitae tristique ante. Cras pretium rutrum egestas. Integer ultrices libero sed justo vehicula, eget tincidunt tortor tempus.