Machine Learning-Based Elastic Cloud Resource Provisioning in the Solvency II Framework

The Solvency II Directive (Directive 2009/138/EC) is a European Directive issued in November 2009 and effective from January 2016, which has been enacted by the European Union to regulate the insurance and reinsurance sector through the discipline of risk management. Solvency II requires European insurance companies to conduct consistent evaluation and continuous monitoring of risks-a process which is computationally complex and extremely resource-intensive. To this end, companies are required to equip themselves with adequate IT infrastructures, facing a significant outlay. In this paper we present the design and the development of a Machine Learning-based approach to transparently deploy on a cloud environment the most resource-intensive portion of the Solvency II-related computation. Our proposal targets DISAR(R), a Solvency II-oriented system initially designed to work on a grid of conventional computers. We show how our solution allows to reduce the overall expenses associated with the computation, without hampering the privacy of the companies' data (making it suitable for conventional public cloud environments), and allowing to meet the strict temporal requirements required by the Directive. Additionally, the system is organized as a self-optimizing loop, which allows to use information gathered from actual (useful) computations, thus requiring a shorter training phase. We present an experimental study conducted on Amazon EC2 to assess the validity and the efficiency of our proposal.