Smurf-Based Anti-money Laundering in Time-Evolving Transaction Networks

Money laundering refers to the criminal attempt of concealing the origins of illegally obtained money, usually by passing it through a complex sequence of seemingly legitimate financial transactions through several financial institutions. Given a large time-evolving graph of financial transactions, how can we spot money laundering activities? In this work, we focus on detecting smurfing, a money-laundering technique that involves breaking up large amounts of money into multiple small transactions. Our key contribution is a method that efficiently finds suspicious smurf-like subgraphs. Specifically, we find that the velocity characteristics of smurfing allow us to find smurfs by using a standard database join, thus bypassing the computational complexity of the subgraph isomorphism problem. We apply our method on a real-world transaction graph spanning a period of six months, with more than 180M transactions involving more than 31M bank accounts, and we verify its efficiency. Finally, by a careful analysis of the suspicious motifs found, we provide a classification of smurf-like motifs into categories that shed light on how money launderers exploit geography, among other things, in their illicit transactions.