In AML-land weten we allemaal: het schatten van de omvang van witwassen is een ambitieuze onderneming. Nu heeft de Italiaanse FIU een poging gedaan om de grootte van het probleem inzichtelijk te maken. Afgelopen week verscheen het rapport.
Hoe hebben ze het aangepakt? In het kort:
“1. First, we employ a rigorous selection of relevant STRs to build a database of “observed ML” that closely aligns with the phenomenon under analysis. In this regard, we only consider STRs that, based on financial analyses conducted by UIF experts, are either classified as high-risk or have received significant feedback from enforcement agencies.
2. Secondly, we use a machine learning algorithm to estimate the value of ML financial transactions that each bank in each of the Italian provinces is expected to report to the UIF on the basis of a set of explanatory variables related to ML. By comparing these estimates with observed STRs, we identify two sets of bank-province observations: those featuring a value of ML roughly in line with the expected value, thus considered as “reliable” observations, and those that are inconsistent with respect to it (either less or in excess), and may thus be regarded as “unreliable” observations.
3. Finally, we apply a version of the Predictive Mean Matching method for imputation (…). This approach is used to replace the values of ML financial transactions reported by “unreliable” bank-province observations (i.e., the output of phase 2) with those from the most statistically similar “reliable” bank-provinces, based on an appropriate distance function.”
“The results provide an estimate of money laundering in Italy for the period 2018-2022 of around 1.5-2.0% of GDP, corresponding to approximately 25-35 billion euros per year.”
Meer weten? Het hele rapport is via een link in de comments te vinden. Zie in relatie tot dit onderwerp ook de link naar een publicatie uit 2021 van Edwin Kruisbergen over vragen die je eigenlijk (nog) niet moet stellen.