A Risk-Based AML Framework: Finding Associates Through Ultimate Beneficial Owners

A Risk-Based AML Framework: Enhancing Detection of Associates Through Ultimate Beneficial Owners

Introduction: Addressing AML Compliance Challenges

Money laundering represents a significant portion of the global economy, estimated at 2-5% of the world’s GDP, and poses a critical risk by funding criminal networks and terrorism. Despite the extensive efforts to curb these illicit activities, only a small fraction of laundered funds are captured, particularly in Europe, where it is estimated that about 1.1% of such funds are detected. The European Union’s Fifth Anti-Money Laundering Directive (5AMLD), enacted in 2018, expands regulatory requirements by broadening transparency measures, including mandates for publicly accessible Ultimate Beneficial Owner (UBO) registries and extending due diligence obligations to various entities, including small businesses. Financial institutions (FIs) and small and medium businesses (SMBs) face increasing complexity in complying with these requirements, especially in identifying Politically Exposed Persons (PEPs) and their associates. Traditional rule-based systems often fall short in detecting complex money laundering schemes, resulting in inefficiencies and high false-positive rates. This article presents a novel framework that leverages graph analysis and machine learning techniques to enhance the Know Your Customer (KYC) process through improved identification and risk assessment of beneficial owners and their associates.

Dataset Insights: Ultimate Beneficial Owner Registry in Luxembourg

The framework is developed using a rich dataset extracted from Luxembourg’s UBO registry, which comprises nearly 94,000 companies and over 71,000 UBOs. Due to privacy concerns and GDPR rulings, the dataset is anonymized by replacing names, company identifiers, and addresses with unique random IDs, ensuring individuals cannot be re-identified.

Two key risk factors emerge from data analysis: companies with underage shareholders and the concentration of companies registered at the same address. The presence of minors as UBOs, though not illegal, is recognized by the Financial Action Task Force (FATF) as a potential indicator of ownership concealment. Furthermore, some addresses host hundreds of companies, which may indicate risks associated with “letterbox” companies or other suspicious clustering. These insights form the basis for novel risk indicators integrated into the AML framework.

Methodology: Using SimRank to Detect Known Associates

To bridge gaps in current KYC practices, especially regarding the identification of close associates of PEPs, the authors apply the SimRank algorithm on a bipartite graph constructed from investors (UBOs) and companies. SimRank calculates structural similarity between nodes based on their relationships; two investors are considered similar if they invest in similar companies, and two companies are similar if they share investors. This recursive measure allows detection of indirect associations not identifiable through traditional methods.

A GPU-accelerated implementation of SimRank addresses computational challenges given the size of the dataset. The graph is partitioned into disconnected components to optimize performance further. This approach enables uncovering known associates beyond direct matches on sanctions or PEP lists by considering structural context within ownership networks.

Bastian Schwind-Wagner_Follow "This research introduces a novel AML risk assessment framework that utilizes graph-based similarity algorithms to identify associates through Ultimate Beneficial Owner networks, enhancing detection capabilities beyond traditional methods while maintaining operational scalability."

Risk Metric Construction: Integrating Multiple Indicators

A comprehensive risk score is formulated combining several indices reflecting different dimensions of AML risk:

• Country Risk Index (CRI): Based on country of birth and citizenship, weighted by FATF black/grey lists and Corruption Perceptions Index.
• Address Overlap Index (AOI): Measures risk associated with the number of companies registered at the same address.
• PEP and Sanctions List Index (PSLI): Flags direct matches with PEP or sanctions lists.
• Minor Involvement Index (MII): Indicates involvement with companies having underage shareholders.
• Adverse Media Index (AMI): Represents presence in negative media reports (excluded in this study).
• Associates Risk: Aggregates risk contributions from associates identified via SimRank.

Weights are assigned to each indicator based on regulatory guidance and expert input. The final metric supports a risk-based approach by allowing thresholds to trigger Enhanced Due Diligence (EDD) procedures when exceeded.

Evaluation and Results: Scalability and Effectiveness

Experiments simulate different scenarios by randomly flagging subsets of UBOs as PEPs to test framework scalability and robustness. Both weighted and unweighted variants of SimRank are examined. Results show that incorporating weights based on ownership shares reduces false positives and limits excessive flagging.

The framework efficiently identifies known associates for individuals, with the number of flagged entities increasing linearly with PEP count rather than exponentially. This scalability is crucial for practical deployment in financial institutions handling large customer volumes. Furthermore, the modular design allows integration of additional risk factors as needed.

Discussion: Limitations and Future Directions

While comprehensive, the study acknowledges several limitations. The exclusion of adverse media analysis due to data complexity limits the scope of risk detection. Ongoing monitoring and behavioral profiling remain areas for further development. Moreover, identifying nominees and relatives used to obfuscate ownership requires advanced techniques potentially incorporating social media or media reports with acceptable false-positive controls.

Future research could benefit from multimodal data processing tools like large language models combined with retrieval-augmented generation to summarize diverse information sources effectively. Enhanced visualization tools would also aid compliance officers in navigating complex data relationships.

Conclusion: Towards Robust AML Compliance Using Advanced Analytics

This work presents a risk-based KYC framework that enhances AML compliance by leveraging graph-based similarity measures to identify associates through UBO networks. The use of SimRank enables uncovering indirect relationships critical for detecting hidden risks linked to PEPs and their networks. By integrating multiple risk indicators into a single metric, the framework supports flexible regulatory scrutiny levels and operational efficiency.

The anonymized Luxembourg UBO dataset released alongside this research encourages collaboration and validation within the research community. Ultimately, combining advanced analytical techniques with evolving regulatory standards promises stronger defenses against financial crime and greater financial system integrity.

Dive deeper

• Research ¦ Sasan JAFARNEJAD, François ROBINET, Raphaël FRANK (2024) A Risk-Based AML Framework: Finding Associates Through Ultimate Beneficial Owners, Paper published in a book (Scientific congresses, symposiums and conference proceedings), DOI: 10.1109/CIFEr62890.2024.107728168 ¦ Link ¦ licensed under the following terms, with no changes made: CC BY 4.0