Machine Learning and PRIIPs

PRIIPs regulation has entered in place on January this year. Fairmat has invested in R&D in order to be able to offer to our client base the best set of tools and methodologies.

In particular, Fairmat had an remarkable success in the life insurance market as we managed to offer to our clients a recognized solution aiming to simplify several technicalities in the redaction of KIDs. Below we will discuss the issues related to the generation of the KID figures for the life insurance PRIIPs linked to the performance of segregated funds (Gestioni Separate in Italian). This type of product belongs to the Category IV PRIIPs where some of the factors which influences the PRRIP performance are not observable and must hence be estimated.

Just to help the reader not familiar with the PRIIPs regulation, PRIIPs are divided into categories also depending on the relationships between the return provided to the investor and the underlying risk factors. In this respect:

• Category II treats products where the payoff is linear (namely an ETFs, funds)
• Category III treats products where the payoff is nonlinear (i.e. structured bonds, certificates)
• Category IV treats products where it not possible to describe the relationship between product performances and underlying risk factors in terms of a clearly identifiable payoff but the payoff depends on internal company assumptions. Such assumptions must be derived using well recognized market practices and the insurance market has identified the solvency measures based on ALM simulations such practices.

The issues with ALM simulations is that usually they are very time consuming and usually insurance companies have the ALM simulators programmed to work only with the assumptions used in the solvency II regulation (i.e. risk neutral projections). Having ALM simulators work with the PRIIPs regulation requires a lot of effort.

In Fairmat we developed a procedure based on Machine Learning techniques able to identify the relationships between risk factors and fund returns starting from the existing information that life insurance company already have because they have to produce estimates for other goals (i.e. Solvency, ORSA). These techniques allows to estimate a model for the segregated funds return which can also work with PRIIPs compliant scenarios (risk neutral, performance, stress).

A technical description of the procedure is available here, contact me if you need to know more.