The purpose of the project is to study methods of Data Mining, to support the identification of suspicious transactions, deemed to be at risk of money laundering. This was possible, first of all, through a theoretical analysis of the various most commonly used Data Mining techniques nowadays. The study will continue with an in-depth study of the literature on the detection and handling of suspicious transactions. Focusing on the four approaches currently considered most effective: • Rule-based approach; • Model based approach; • Classification-based approach; • Clustering-based approach. After this purely theoretical part, we will proceed to the elaboration and then to the implementation of a deterministic model of Data Mining which will have the objective of determining a risk score to be attributed to each branch of CNP Vita UniCredit, the insurance company that provided the real data for analysis. These data include both alert indicators recorded as a result of transactions carried out by customers, and a series of variables considered significant for work purposes. The methodology that will be used will consider predictive and classification statistical models, focusing also on parametric and non-parametric models for the rating study. This analysis will be placed in the center of today's national and international regulatory context. In fact, in the first chapter will be addressed a detailed study of the legislation, starting from the first recommendations of the FATF, passing from the European directives, to get into Italian legislation. A special note was also made for the rules dedicated to insurance institutions. An analysis of the history of money laundering legislation allows us to understand how this phenomenon has evolved over the years, becoming increasingly difficult to identify, thus forcing the various financial and non-financial institutions to use increasingly innovative methods.
L’elaborato ha lo scopo di studiare metodiche di Data Mining, a supporto dell’identificazione di operazioni sospette, ritenute a rischio riciclaggio di denaro. Questo è stato possibile, in primo luogo, attraverso un’analisi teorica delle varie tecniche di Data Mining più utilizzate al giorno d’oggi. Lo studio proseguirà poi con un approfondimento della letteratura in tema di rilevazione e trattazione di operazioni sospette. Soffermandosi sui quattro approcci al momento ritenuti più efficaci: • Rule-based approach; • Model based approach; • Classification-based approach; • Clustering-based approach. Dopo questa parte prettamente teorica, si passerà all’elaborazione e successivamente all’implementazione di un modello deterministico di Data Mining che avrà come obiettivo quello di determinare un punteggio di rischio da attribuire ad ogni filiale di CNP Vita UniCredit, compagnia assicurativa che ha fornito i dati reale per l’analisi. Tali dati comprendono sia indicatori di anomalia registrati in seguito ad operazioni eseguite dai clienti, che una serie di variabili ritenute significative ai fini del lavoro. La metodologia che si utilizzerà, prenderà in considerazione modelli statistici predittivi e di classificazione, concentrandosi anche sui modelli parametrici e non parametrici per lo studio del rating. Questa analisi sarà posta al centro del contesto normativo odierno nazionale e internazionale. Infatti, nel primo capitolo sarà affrontato uno studio dettagliato della normativa partendo dalle prime raccomandazioni del GAFI, passando dalle direttive europee, per arrivare alla legislazione italiana. Una nota di riguardo è stata fatta anche per le norme dedicate agli enti assicurativi. Un’analisi della storia della normativa sul riciclaggio di denaro permette di evidenziare come questo fenomeno si sia evoluto nel corso degli anni diventando sempre più difficile da individuare, costringendo quindi i vari istituti finanziari e non, a servirsi di metodologie sempre più innovative.
Modelli di Data Mining a supporto dell’antiriciclaggio: studio di un modello deterministico per la realizzazione di rating antiriciclaggio
BRAMBILLA, ANDREA
2016/2017
Abstract
The purpose of the project is to study methods of Data Mining, to support the identification of suspicious transactions, deemed to be at risk of money laundering. This was possible, first of all, through a theoretical analysis of the various most commonly used Data Mining techniques nowadays. The study will continue with an in-depth study of the literature on the detection and handling of suspicious transactions. Focusing on the four approaches currently considered most effective: • Rule-based approach; • Model based approach; • Classification-based approach; • Clustering-based approach. After this purely theoretical part, we will proceed to the elaboration and then to the implementation of a deterministic model of Data Mining which will have the objective of determining a risk score to be attributed to each branch of CNP Vita UniCredit, the insurance company that provided the real data for analysis. These data include both alert indicators recorded as a result of transactions carried out by customers, and a series of variables considered significant for work purposes. The methodology that will be used will consider predictive and classification statistical models, focusing also on parametric and non-parametric models for the rating study. This analysis will be placed in the center of today's national and international regulatory context. In fact, in the first chapter will be addressed a detailed study of the legislation, starting from the first recommendations of the FATF, passing from the European directives, to get into Italian legislation. A special note was also made for the rules dedicated to insurance institutions. An analysis of the history of money laundering legislation allows us to understand how this phenomenon has evolved over the years, becoming increasingly difficult to identify, thus forcing the various financial and non-financial institutions to use increasingly innovative methods.È consentito all'utente scaricare e condividere i documenti disponibili a testo pieno in UNITESI UNIPV nel rispetto della licenza Creative Commons del tipo CC BY NC ND.
Per maggiori informazioni e per verifiche sull'eventuale disponibilità del file scrivere a: unitesi@unipv.it.
https://hdl.handle.net/20.500.14239/5709