Webinar: RESERVING – MACHINE LEARNING AND IFRS 17
Fagkomité skade inviterer til «Reserving – machine learning and IFRS 17»
SEMINARET VIL BLI HOLDT SOM ET TODELT WEBINAR
ONSDAG 26. Mai 2021 kl 08:45 – 11:00
TORSDAG 27. Mai 2021 kl 13:15 – 15:30
Program 26. Mai:
8:45 – 8:50 Welcome and introduction
v/ Mats Sollie, leder fagkomité skade
Reserving and Machine Learning
8:50 – 9:50 ‘Neural Networks in Reserving: how and why are they worth considering?’
v/ Muhammed Al-Mudafer, University of New South Wales
The application of Machine Learning and AI techniques in insurance continues to accelerate. These techniques have shown potential as valuable tools in a number of areas of actuarial work, including Loss Reserving, where in recent years there have been a number of proposals regarding how techniques such as Neural Networks can be used to improve the prediction accuracy of Outstanding Claims. However, with the complexity of Machine Learning and AI techniques comes a myriad of challenges which need addressing. For example, while these models are typically very accurate in terms of finding central estimates, there is in general very little work focusing on the distribution of such estimates, which is vital in Actuarial practice including in Loss Reserving. Furthermore, there is often reluctance towards using Machine Learning models in practice, partly due to their lack of interpretability and difficulty in incorporating actuarial judgement.
In this talk we use a framework which allows us to address the above problems. Specifically, we introduce to Loss Reserving a Neural Network design called the Mixture Density Network (MDN). Importantly, this not only inherits the accuracy of neural networks but is also able to provide a (statistical) distribution of outcomes. We will show that the MDN outperforms the benchmark GLM approach in the prediction of the mean, volatility, and tail quantiles of outstanding claims in a wide range of scenarios of varying complexity and specifications.
9:50 – 10:00 Pause
10:00 – 11:00 ‘Practical Examples of Machine Learning in Reserving’
v/ Gráinne McGuire, Director Taylor Fry Consulting Actuaries and member of IFoA’s Machine Learning in Reserving Working Party
Gráinne will cover some practical examples for getting started with using machine learning for reserving, based on her own work and the work of the IFoA’s Machine Learning in Reserving Working Party (https://institute-and-faculty-of-actuaries.github.io/mlr-blog/). Topics include:
• Using the LASSO to automatically assemble a GLM-like model
• Machine learning modelling frameworks for common tasks such as model fitting, hyper-parameter tuning, validation and benchmarking.
• Reserving model diagnostics
Program 27. Mai:
13:15 – 13:20 Welcome and introduction
v/ Mats Sollie, leder fagkomité skade
Reserving and IFRS 17
13:20 – 14:20 ‘Use of the NP-approximation to determine a risk adjustment under IFRS 17 in a non-life portfolio’
v/ Tatiana Santos Marques, PwC Portugal
We present a model that allows us to determine a risk adjustment under IFRS 17 for non-life business, using the NP-approximation. The model is an extension and generalisation of the model used by Kristiansen (1991). Risk adjustments can be calculated separately for the liability for incurred claims (LIC) and the liability for remaining coverage (LRC), as required by IFRS 17. The risk adjustment for the LIC can be further split into a risk adjustment for claims reported but not settled, and a risk adjustment for claims incurred but not reported. All formulas turn out to be quite manageable.
14:20 - 14:30 Pause
14:30 – 15:30 ‘Consistent Development Patterns’
v/ Walther Neuhaus, Zabler-Neuhaus AS, Alambra Consulting Lda and ISEG
Traditional claim estimation in general insurance works with accident year cohorts. For the accounting standard IFRS17 and often for reinsurance purposes, claims are required to be split by contract year in addition to accident year. The traditional approach falls short of that requirement. Rather than just replacing accident year by contract year in the definition of a claim cohort and carrying on as before, one should add the contract year as a cohort classifier. Following ideas of Norberg and Hesselager, we use a continuous time model to supply an internally consistent set of (different) development patterns needed for projecting the evolution of the different claim cohorts.
Muhammed Al-Mudafer is an Actuarial Honours Graduate at the University of New South Wales, Sydney, Australia. Muhammed will be presenting a paper co-authored by Benjamin Avanzi (Professor of the University of Melbourne, who has held academic positions in Europe, Canada and Australia), Greg Taylor (Adjunct Professor at the University of New South Wales with more than 45 years' experience as a qualified actuary, both as consultant and academic) and Bernard Wong (Professor and Head of School, Risk and Actuarial Studies, at the University of New South Wales).
Gráinne is a Fellow of the Institute of Actuaries of Australia and holds a PhD in Statistics from Heriot-Watt University. A director at Taylor Fry Consulting Actuaries, Gráinne develops statistical models of future liabilities, capital needs and further insights for accident compensation schemes, government social welfare and other insurances. She has written several actuarial papers and presentations on stochastic modelling and reserving.
Tatiana Santos Marques
Tatiana holds a Masters in Actuarial Science from the University of Lisbon. She is a Senior Associate at PwC Portugal and has over 4 years of experience working with Portuguese insurance companies in the valuation of life and non-life technical provisions. In the last two years, Tatiana has been involved in IFRS 17 projects, including gap analysis, implementation and training to clients.
Walther qualified as an actuary at the University of Oslo in 1982 and obtained his Ph.D. in Actuarial Science in 1988. His main areas of actuarial interest are loss reserving and credibility theory. Today he works as an independent consultant for non-life insurers in Norway. He teaches on the Master of Actuarial Science program at the University of Lisbon. Walther is a founder and partner of Alambra Consulting, lda. in Lisbon.