Statistically Robust Estimation of Unprotected Identity Theft in Individual Tax Returns: A Non-Parametric Simulation Based Approach

• Activity Number: 548 - Using Artificial Intelligence and Advanced Statistical Methods to Improve Official Statistics
• Type: Topic Contributed
• Date/Time: Thursday, August 6, 2020 : 1:00 PM to 2:50 PM
• Sponsor: Government Statistics Section
• Author(s): Sabyasachi Guharay*
• Companies: US Internal Revenue Service
• Keywords: Monte-Carlo; bootstrapping; non-parametric

Abstract

Identifying the true amount (counts and dollars in revenue lost) of identity theft (IDT) within the individual tax returns is a challenging analytical problem for the US Internal Revenue Service (IRS). IRS doesn’t have the resources to individually examine every return which was flagged as potential IDT. Therefore, analytical methods need to be developed to robustly estimate the total amount and counts of IDT unprotected based on properly developed random sampling. In light of this issue, we demonstrate a robust statistical approach to estimating the total counts and total dollar amounts in unprotected IDT in individual tax returns. Specifically, we will show how one can use non-parametric method of bootstrapping to estimate the average loss amounts in IDT based on a properly selected random sample of flagged IDT returns. After the bootstrapping implementation is done, we will show how a large-scale Monte-Carlo (MC) simulation can be used to robustly estimate a 95% confidence interval for the amount of IDT lost in revenue and the amount of cases that were truly IDT. This approach does not make any unreasonable distributional assumptions of the nature of the IDT losses. Therefore, i