Detecting Anomalies in Heterogeneous Population-Scale VAT Networks - A Network and Machine Learning Approach to Detect Value Added Tax Fraud

Aug 4, 2026

Upload by Jiven

Anomaly detection in network science identifies aberrant edges, nodes, subgraphs, or events. Heterogeneous networks include information beyond the observed structure, and population-scale Value Added Tax (VAT) networks—built from pairwise interactions among VAT-registered taxpayers—demand scalable algorithms. By quantitatively characterizing the nature of VAT anomalies, the proposed framework detects them using micro-, meso-, and global-scale pattern information, enabling automated, real-time implementation. This supports early identification of fraud and helps revenue authorities prevent major tax revenue losses.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages14

Reading Time35 min

AccessFree to download

Abstract

Introduction

Value Added Tax and invoice-credit mechanism

Vulnerabilities and VAT fraud types

Need for scalable network-science modeling

View metadata, citation and similar [papers at ](papers at core.ac.uk)[core.ac.uk](papers at core.ac.uk) brought to you by CORE [provided by](provided by arXiv.org)[ arXiv.org](provided by arXiv.org) e-Print Archive arXiv :2106 . 14005v1 [physics .soc-ph] 26 Jun 2021 Detecting anomalies in heterogeneous population{scale VAT networks A. Alexopoulos ∗ P. Dellaportas † Stanley Gyoshev ‡ Christos Kotsogiannis § So􀀌a C. Olhede ¶ Trifon Pavkov ‖ June 29, 2021 Abstract Anomaly detection in network science is the method to determine aberrant edges, nodes, subgraphs or other network events. Heterogeneous networks typically contain information going beyond the observed network itself. Value Added Tax (VAT, a tax on goods and services) networks, de􀀌ned from pairwise interactions of VAT registered taxpayers, are analysed at a population{scale requiring scalable algorithms. By adopting a quantitative understanding of the nature of VAT{anomalies, we de􀀌ne a method that identi􀀌es them utilising information from micro{scale, meso{scale and global{scale patterns that can be interpreted, and e􀀎ciently implemented, as population{scale network analysis. The proposed method is automatable, and implementable in real time, enabling revenue authorities to prevent large losses of tax revenues through performing early identi􀀌cation of fraud within the VAT system. Keywords| anomaly detection; clustering; heterogeneous data sources; fraud detection; value added tax; carousel fraud 1 Introduction Value Added Tax (VAT) is a major source of revenue for, remarkably, 1 over 160 countries. VAT is a consumption tax in the sense that the VAT collected through the supply chain is the VAT paid by the consumers in the place where the good is consumed. Underlying VAT therefore there is an \invoicecredit" mechanism where the net tax liability of a business2 is calculated by subtracting from the sales the aggregate value of VAT paid on invoices for the inputs used in production. The \invoice{credit"mechanism requires sellers along the production chain to provide invoices to their buyers showing the amount of VAT that was paid on a given transaction. The fractional revenue collection on the value added that is generated at every stage of the production chain is remitted to the appropriate revenue authority. The business-to-business (B2B) transactions and the VAT \invoice{credit" mechanism de-facto create a network through which businesses are interacting within and across economic sectors. Despite its remarkable rise as a tax innovation, it is universally recognised that the current VAT system has both weaknesses and vulnerabilities (Keen and Smith, 2006) making it not 􀀌t for purpose ∗ MRC Biostatistics Unit, University of Cambridge, University Forvie Site, Robinson Way, Cambridge CB2 0SR, UK. Email: [angelos@mrc-bsu.cam.ac.uk](angelos@mrc-bsu.cam.ac.uk). †Department of Statistical Science, University College London, UK, The Alan Turing Institute, London, UK and Department of Statistics, AUEB, Greece. Email: [p.dellaportas@ucl.ac.uk](p.dellaportas@ucl.ac.uk). ‡Department of Finance, University of Exeter Business School, Streatham Court, Rennes Drive, EX4 4PU, England, UK. Email: [S.Gyoshev@exeter.ac.uk](S.Gyoshev@exeter.ac.uk). § Department of Economics, University of Exeter Business School, Streatham Court, Rennes Drive, EX4 4PU, England, UK, Tax Administration Research Centre (TARC), University of Exeter, UK and CESIfo, Munich, Germany. Email: [C.Kotsogiannis@exeter.ac.uk](C.Kotsogiannis@exeter.ac.uk). ¶ Institute of Mathematics, Ecole Polytechnique Federale de Lausanne, Lausanne, Switzerland and Department of Statistical Science, University College London, UK. Email: [sofia.olhede@epfl.ch](sofia.olhede@epfl.ch). ‖Department of Finance, University of Exeter Business School, Streatham Court, Rennes Drive, EX4 4PU, England, UK and National Revenue Agency, So􀀌a, Bulgaria Email: tp335@exeter .ac .uk. 1 For the remarkable rise of VA

FAQs

What is the core goal of the paper’s anomaly detection method?

To identify anomalous edges, nodes, subgraphs, or other network events in VAT networks that indicate potential fraud.

Why are VAT frauds linked to heterogeneous, population-scale networks?

VAT fraud involves interactions across multiple VAT-registered traders, so the behavior can be expressed both at individual (node) level and group (community) interaction level, producing heterogeneous patterns.

How does the proposed approach support real-world fraud prevention?

The method is designed to be automatable and implementable in real time, enabling early identification of fraud to reduce large losses in tax revenues.

Aug 4, 2026

Upload by Jiven