We know that one of the main barriers to development of ML skills by reserving actuaries is time and resources. This case study from NFU Mutual shows how it is possible if a company puts the right support in place, as discussed by Sophie Lawless and Matthew Byrne.
Sophie
Machine learning in reserving is not as widely used as machine learning in pricing, but the significant potential in this area is undeniable.
Coming straight from university into the world of work at a time when AI and machine learning were rapidly taking off, I knew that I wanted to incorporate useful machine learning into my work, using skills I had built up during my postgraduate degree in Mathematical Sciences. I’m fortunate that my exploration into machine learning has been completely encouraged since starting my role at NFU Mutual, having received substantial support from my manager and wider team in this endeavour.
Needing a starting point, my initial research led me to the work of the IFoA General Insurance Machine Learning in Reserving Working Party (MLRWP). As a new actuarial trainee working in reserving, their blogs and online book provided exactly what I had been looking for - a direct application of machine learning, specifically XGBoost to reserving triangles.
Section 9 of the MLRWP Online Book: ML Modelling on Triangles - A Worked Example particularly captured my attention. Learning about classical methods for reserving triangles simultaneously with a machine learning approach gave a unique perspective on how new methods for reserving can complement classical methods. This article provides a comprehensive application of machine learning in reserving and an XGBoost method that can be experimented with on real reserving triangles. There is significant rising growth in the area of XGBoost in general and other machine learning techniques in reserving. We will continue to stay up to date with these advancements, to explore the usefulness of machine learning in our reserving.
It is fantastic that the MLRWP share their insights to help actuaries around the world. The availability of their applications means that actuaries who are motivated and inspired to learn machine learning can more readily acquire the necessary knowledge and allocate time and resources for integrating machine learning into their work. This leads to more advancements in this space and enables actuaries to enhance their data science skills, ultimately leading to improved overall reserving.
Matt
NFU Mutual has always had a focus on continuous learning and development for all employees. This support has enabled me to be involved in a wide range of actuarial profession working parties over the years, most recently chairing the Professionalism, Regulation & Ethics workstream for the Data Science Community.
To support colleagues developing their data skills at NFU Mutual, as well as proactive training opportunities like Sophie has discussed, there are a range of data science apprenticeships now on offer. These apprenticeships provide multi-year degree-level training in data science and AI, with colleagues spending one day a week attending online lectures. Currently two members of my own team are taking these courses, and I have been providing technical coaching support to several other colleagues from the wider business, giving them opportunities to build simple machine learning models to put their learning into practice.
Doing data science well requires a blend of technical modelling skills and business domain knowledge. With all the new tools and techniques available, it’s a great time for actuaries to be upskilling and learning new techniques to solve old problems.