The Promises of Machine Learning, Bayesian Methods & Big Data Architectures for Advancing Predictive Credit Risk Analytics in the Banking Sector - Université d'Angers Accéder directement au contenu
Communication Dans Un Congrès Année : 2015

The Promises of Machine Learning, Bayesian Methods & Big Data Architectures for Advancing Predictive Credit Risk Analytics in the Banking Sector

Résumé

Banks have successfully managed to grow large and diversified customer portfolios, as well as to offer complex financial products that match divergent customer needs while generating high profitability. However, such achievements have been unrivalled with matching advances in risk governance. Banks are now looking into possible solutions to proactively predict consumer credit risk to suggest pre-emptive risk mitigation actions.

The machine learning has marked its penetration in the field of financial engineering technology with the era of Big Data. IT services companies are facing more and more of an insatiable demand given the volume of heterogeneous and homogeneous data coming from multiple platforms and information systems associated with the difficulty of data representation and association that emanate from such cases.

Machine-learning techniques based on Bayesian rules are ideally suited to handle predictive risk analytics’ particular challenges because of the large dataset sizes and the complexity of the possible relationships among consumer transactions and characteristics. The objective of the machine-learning models is to identify statistically reliable relationships in constructing delinquency forecasting models in an unsupervised or semi-supervised learning setting. This would translate in establishing strong links between ex ante observable characteristics (customer related, market related and/or transaction related) and future delinquencies.

Big data is an emerging paradigm that is taking the forefront of many IT based discussions and is considered by many as the next technology revolutionary phase after the massive advancements witnessed in the computing and the networking phases. The sheer size and scope of the customer dataset is a significant challenge, reason why we also look into possible big data architectures, to suggest which are the most suitable for predictive risk analytics.

Our research theme on risk analytics embraces the theoretical framing of the Design Theory for Systems That Support Emergent Knowledge Processes (EKP) discussed by Markus et al. (2000, 2002). The continuing evolution of complex financial products, along with new regulations designed to contain risk and protect markets and investors, makes the requirements formulated for banking risk analytics a challenging task for emergent knowledge. Reason why this class of user requirements could be considered as EKP.

This paper discusses the potential of combining machine-learning Bayesian techniques with big data architectures for banks given their actual state of informational readiness to establish dynamic predictive risk analytics models. More specifically the paper addresses the following two-folded research question:

- Which learning methods and big data architectures are best suited for banking predictive risk analytics? 

Fichier non déposé

Dates et versions

hal-02528780 , version 1 (02-04-2020)

Identifiants

  • HAL Id : hal-02528780 , version 1
  • OKINA : ua13786

Citer

Myriam Raymond. The Promises of Machine Learning, Bayesian Methods & Big Data Architectures for Advancing Predictive Credit Risk Analytics in the Banking Sector. International Forecasting Financial Markets Conference, 2015, Rennes, France. ⟨hal-02528780⟩
19 Consultations
0 Téléchargements

Partager

Gmail Facebook X LinkedIn More