Current challenges in non-life pricing: Risks & Opportunities
Key speakers:
Hatim Maskawala | Managing Director Badri Management Consultancy
Annabelle LONGO | Senior Manager, Head of Profitability, Addactis
Juan DE OYARBIDE | Consultant, Pricing Expert, Addactis
Nabil RACHDI | Senior Manager, Head of Data Science, Addactis
Introduction
Presented by Badri Management Consultancy
Thanks to a fruitful partnership, Badri Management Consultancy, who was recently named Strategic Partner of the Industry in the «7th Middle East Insurance Industry Awards 2020» by the Middle East Insurance Review and major player in actuarial consulting in the MENA region, and addactis®, The RiskTech for Insurance, as software editor, have been supporting carriers for 3 years to face their insurance key challenges.
Part 1 | Implementing Generalized Linear Models in a changing environment: What are the benefits?
Presented by Addactis
The implementation of Generalized Linear Models (GLMs) facilitates the construction of statistically sound rating structures, bringing fairness and transparency to the insurance market. While in developed markets they have been applied in full generality for the past two decades, some emerging markets are still in transition to this multivariate technique.
The Saudi market, encouraged by its local regulator SAMA, is an example of such transformation. In this talk, we argue that the scope of GLMs can go much beyond ratemaking and, essentially, their application can be highly beneficial for insurers during the current changing environment.
Part 2 | Bad data, bad models, the dark side of big data
Presented by Addactis
With the increasing availability of data and modeling libraries, one could expect a systematic improvement in our understanding of certain risks and better control of the claims experience. However, in the absence of precaution and clear methodological approaches linked to operational issues, the outcomes could be misleading.
Data is usually available in a «raw» state, i.e. in a format that cannot be directly used, and with errors or inaccuracies that may be significant. The question of the selection of relevant data also remains unresolved and represents an important issue in the preliminary phases of modelling.
Data Science methods and Machine Learning algorithms can contribute to improving our understanding of claims experience – provided that such modelling has been built in line with technical operational and production objectives. For example, modelling a loss event with GLM models – i.e. with a central tendency – and deriving or quantifying extreme behavior from these models can produce an inconsistency between design and objective, resulting in a lack of performance in the overall modelling process. The same applies to the choice of the right granularity and quality of data according to the desired risk analysis objectives.
In this presentation, we will present what we call «Bad Data» and «Bad Models» by giving some oncrete examples, and we will conclude with some recommendations in order to take full advantage of new data and advanced risk modelling techniques based on Data Science and Machine Learning.
