Data Science: research working parties

This working party will investigate tools and packages available to perform data visualisation and produce outputs that conform to best practice in the area.

This working party will conduct a systematic review of AI applications in actuarial science, leaning towards practitioner's perspective.

The aim of this study would be to investigate whether the pricing/reserving/capital (actuarially driven departments in insurance companies) could improve their output through the use of AI-based tools and why. Both cases, where AI does not bring satisfiable results that would justify disruption and the converse should be examined, yielding a clear view of the direction of the actuarial profession.

This working party will undertake research into the use of Bayesian Methods

The aim of this working party is to:

1. Broaden knowledge of Privacy Preserving ML methods and their application in insurance industry.
2. Popularize Federated Learning Machine Learning techniques.
3. Increase awareness of the industry on how the technology might change the way insurance works.
4. Understand potential difficulties and risks in adopting FL technology in insurance and attempt to address those issues.
5. Understand when it is worth collaborating and when companies should rely on their own data.

The aim of this working party is to focus on:

• Members’ engagement and education (events, content).
• DS-support of IFoA Education (exams, etc.).
• Collaboration with DS WPs.
• Content distribution channels.

The aim of this research is to construct highly flexible actuarial models such as:

• finite mixture regression models for the number and the costs of claims
• univariate and multivariate regression models with varying dispersion and shape for claim frequencies and severities
• copula-based models with regression structures on the mean, dispersion and dependence parameters for different claim types and their associated claim counts and costs
• dependence modelling in risk management and sensitivity analysis
• first-order integer-valued autoregressive INAR(1) regression models with varying dispersion for time series of claim counts
• neural network embeddings of the aforementioned models which are able to capture the stylized characteristics of structured, semi-structured and unstructured insurance data
• classification of green bonds using statistical learning methods and decarbonization
• Gaussian process spatial-temporal regression models; and
• heavy tails and extremes in spatial and temporal settings.

This working party will:

• Looking at applications of data science techniques to solve climate-related problems.
• Producing related content, including in collaboration with the Sustainability Board.

This research intends to focus on the following:

1. Climate change - Time series of severe weather events, clustering on geographic location.
2. Covid - Clustering and segmentation geographically, intersecting of multiple data sets.