Tech for tackling tax evasion

Ofir Reich

25 Apr 2017, 3 p.m.
Room 9

The Delhi government collects Value Added Tax from companies registered in its districts. One way to commit VAT fraud is to set up a fictitious new company listed as buying from retailers and selling to suppliers, and then disappear when the tax authorities attempt to locate the company.

Using digitised tax return data from the Delhi government going back several years, and the results of 80,000 audits of companies performed by the tax authority, CEGA researchers have created an automatic Machine Learning model that classifies certain companies as more likely to be fictitious. The Delhi tax authority can then target such companies for audits or sanctioning of their owners, making it harder for people to evade tax payment and increasing government revenue.

Such a project could have an added benefit for honest company owners as well — currently there is a perception that tax audits are misguided, or are intentional red tape to shake down business owners for a bribe. An automatic and transparent system could lend justified credibility to the audit process and improve the business environment.