Keywords

inherent risk, risk assessment, data quality, prevention of money laundering and terrorist financing, permutation method, least-squares method, method of minimising within-cluster variance, algorithmic scoring

In the context of increasingly demanding regulatory frameworks and the dynamic nature of the financial market, an accurate and objective assessment of inherent Money laundering and terrorist financing (ML/TF) risks has become a regulatory imperative. The paper proposes a methodological approach based on the permutation method and the method of minimising within-cluster variance, adapted for the classification of quantitative ML/TF risk indicators into categories using an interval scoring system. The model allows for the empirical evaluation of inherent risk categories – such as customer risk, product and service risk, geographical risk, and distribution channel risk – based on quantitative data provided by supervised entities. A key mathematical challenge of this methodology is the combinatorial complexity arising from the distribution of dana across multiple intervals, which requires algorithmic processing and cannot be conducted manually in a reliable way. By applying a permutation-based optimisation approach in combination with least-squares minimisation, the model ensures statistically consistent scoring with significantly reduced subjectivity. Although the method applied in this analysis is not a classical least-squares regression, it is based on the same mathematical principle of quadratic optimisation. Specifically, the method minimises the sum of squared deviations of observations from their group means, with the objective of achieving minimum within-cluster variance. The optimisation is performed over discrete group boundaries rather than over a continuous function, which makes the approach suitable for risk classification and score discretisation rather than regression modelling. The method applies a least-squares-type-quadratic minimisation principle to discretise risk indicators into homogeneous groups by minimising withincluster variance. This methodology is adaptable for implementation in dynamic Excel environments or Python-based systems, which makes it a practical and scalable solution for supervisory authorities. The model aligns with the evolving requirements of riskbased supervision frameworks.