arXiv subjects

Rasmus Ingemann Tuffveson Jensen

Publications and source records attributed to Rasmus Ingemann Tuffveson Jensen.

3 recordsLinked to original sources

arXiv2026

Are Higher Suspicious Transaction Report Volumes Associated with More Future Money-Laundering Convictions?

Almost all countries require banks to file suspicious transaction reports (STRs) to help authorities detect and prosecute money laundering. The number of STRs filed in a country is an often-cited metric in international anti-money laundering (AML) compliance evaluations. However, it remains unclear if the volume of STRs filed in a country is associated with the country's future number of money laundering convictions. Using data from Europol, the World Bank, the International Monetary Fund, and the European Sourcebook of Crime and Criminal Justice Statistics, we examine if STR volumes can be associated with future money-laundering conviction volumes in countries in the European Union. Our analysis employs pooled and fixed-effects regression models with one- to five-year lags, controlling for population sizes, shadow-economy estimates, and police force sizes. In our pooled and country-specific fixed-effects models, STR volumes appear to be positively associated with future convictions. However, the association weakens and becomes insignificant when we control for both country-specific and common time effects. Our results provide a caution against assuming that STR volumes, by themselves, are reliable indicators of future money-laundering convictions and, to the extent convictions reflect outcomes, AML effectiveness.

econ.GN↗

arXiv2025

Searching for Smurfs: Testing if Money Launderers Know Alert Thresholds

Objectives: To combat money laundering, banks raise and review alerts on transactions that exceed confidential thresholds. However, the thresholds may be leaked to criminals, allowing them to break up large transactions into amounts under the thresholds. This paper introduces a data-driven approach to detect the phenomenon, popularly known as smurfing. Methods: Our approach compares an observed transaction distribution to a counterfactual distribution estimated using a high-degree polynomial. We investigate the approach with simulation experiments and real transaction data from a systemically important Danish bank. Results: Our simulation experiments suggest that the approach can detect smurfing when as little as 0.1-0.5% of all transactions are subject to smurfing. On the real transaction data, we find no evidence of smurfing and, thus, no evidence of leaked thresholds. Conclusions: Our approach may be used to test if transaction thresholds have been leaked. This has practical implications for criminal justice and anti-money laundering (AML) systems. If criminals gain knowledge of AML alert thresholds, the effectiveness of the systems may be undermined. An implementation of our approach is available online, providing a free and easy-to-use tool for banks and financial supervisors. The null result obtained on our real data helps raise confidence in (though it cannot prove the effectiveness of) anti-money laundering efforts.

econ.GN↗

arXiv2024

Software Vulnerability and Functionality Assessment using LLMs

While code review is central to the software development process, it can be tedious and expensive to carry out. In this paper, we investigate whether and how Large Language Models (LLMs) can aid with code reviews. Our investigation focuses on two tasks that we argue are fundamental to good reviews: (i) flagging code with security vulnerabilities and (ii) performing software functionality validation, i.e., ensuring that code meets its intended functionality. To test performance on both tasks, we use zero-shot and chain-of-thought prompting to obtain final ``approve or reject'' recommendations. As data, we employ seminal code generation datasets (HumanEval and MBPP) along with expert-written code snippets with security vulnerabilities from the Common Weakness Enumeration (CWE). Our experiments consider a mixture of three proprietary models from OpenAI and smaller open-source LLMs. We find that the former outperforms the latter by a large margin. Motivated by promising results, we finally ask our models to provide detailed descriptions of security vulnerabilities. Results show that 36.7% of LLM-generated descriptions can be associated with true CWE vulnerabilities.

cs.SE↗