John Guyton (IRS) presents Integrating Reward Maximization and Inhabitants Estimation: Sequential Choice-Making for Inside Income Service Audit Choice at Georgetown right this moment as a part of its Tax Legislation and Public Finance Workshop hosted by Emily Satterthwaite and Dayanand Manoli:

We introduce a brand new setting, optimize-and-estimate structured bandits. Right here, a coverage should choose a batch of arms, every characterised by its personal context, that will permit it to each maximize reward and preserve an correct (ideally unbiased) inhabitants estimate of the reward. This setting is inherent to many private and non-private sector purposes and infrequently requires dealing with delayed suggestions, small information, and distribution shifts. We exhibit its significance on actual information from america Inside Income Service (IRS). The IRS performs yearly audits of the tax base. Two of its most essential aims are to determine suspected misreporting and to estimate the “tax hole” — the worldwide distinction between the quantity paid and true quantity owed. 

Primarily based on a novel collaboration with the IRS, we forged these two processes as a unified optimize-and-estimate structured bandit. We analyze optimize-and-estimate approaches to the IRS drawback and suggest a novel mechanism for unbiased inhabitants estimation that achieves rewards corresponding to baseline approaches. This method has the potential to enhance audit efficacy, whereas sustaining policy-relevant estimates of the tax hole. This has essential social penalties on condition that the present tax hole is estimated at almost half a trillion {dollars}. We propose that this drawback setting is fertile floor for additional analysis and we spotlight its fascinating challenges. The outcomes of this and associated analysis are at the moment being included into the continuous enchancment of the IRS audit choice strategies.

https://taxprof.typepad.com/taxprof_blog/2023/02/guyton-presents-sequential-decision-making-for-internal-revenue-service-audit-selection-today-at-geo.html