Detecting money laundering using hidden Markov model

Abstract

Recent money laundering scandals, like the Danske Bank and Swedbank’s failure to mitigate money laundering risks (Kim, 2019), have made “anti money laundering” (AML) a much discussed topic. Governments are making AML regulations tougher and financial institutions are struggling to comply, one of the requirements is to actively monitor financial transactions to detect suspicious ones. Most of the financial industry applies simple rule-based methods for monitoring. This thesis provides a practical model to detect suspicious transactions using the hidden Markov model (HMM). The use of HMM is justified, because the criminal nature of a transaction is hidden to the financial institution, only transaction parameters can be observed. By using past data, a model is built to detect if current transaction is suspicious or not. The model is assessed with artificial and real transactions data. It was concluded that this model performs better than a classical k-means clustering algorithm.