Causal Risk Models: A proposal based on associative classes 1 Modelli di Rischio Causali: Una proposta basata su classi associative Paola Cerchiello Department Of Statistics ‘L. Lenti’, University of Pavia, paola.cerchiello@unipv.it Paolo Giudici Department Of Statistics ‘L. Lenti’, University of Pavia, giudici@unipv.it Riassunto: nell’ambito della gestione dei rischi connessi all’attività di una società di telecomunicazioni, ruolo rilevante è assunto dalle problematiche rientranti nella categoria di rischio operativo. Con tale espressione ci si riferisce a tutte le perdite o gli eventi dannosi dovuti a cause rientranti nelle quattro macrocategorie individuate da Basilea: persone, processi, sistemi, eventi esterni. La stima di modelli causali in grado di mettere in relazione le possibili cause con le perdite (su scala ordinale) risulta essenziale nell’ambito di un piano di gestione, mitigazione e prevenzione di rischi operativi. In questo contesto proponiamo l’impiego di regole associative, comunemente utilizzate in ambiti differenti (tipicamente MBA), valutandone la validità su un database reale fornito da una società di telecomunicazioni. Keywords: IT-Operational Risk, Association Rules, HyperLift. Introduction According to Basel Committee and common industry, Operational Risk can be defined as ‘the risk of direct or indirect loss resulting from inadequate or failed internal processes, people and systems or from external events”. In particular we focus on IT- intensive organizations, in other words great relevance is given to losses coming from processes and systems intrinsically tied up to the core business of Telco enterprises. The main objective is to associate the noticed causes of a specific occurred problems (interruptions) to either the resulting loss or the customer type, in order to reveal regularity and associative patterns between those kind of data. In this context we propose to employ a typical unsupervised model, the association rules, developed in the field of computer science and typically used in other fields like Market Basket Analysis and Web Mining. To apply the association rules, the data can present either a transactional or binary format. The employed R package ‘ARules’ allows to choose between both of them and to transform one structure into the alternative one. The available dataset is provided by an israeli company providing Value Added Services (VAS) to SMEs, basically, a set of communications services, both voice and data. Those data deal with the IT problems risible from a communications service 1 This Paper is financially supported by MUSING project (FP6 /027097) and Italian National grant FIRB – 543 –