EAA Web Session ‚Time Series for Actuarial Modelling with Machine Learning‘

Actuaries have long relied on time-tested statistical models to forecast risk. Methods such as ARIMA, GLMs, and the Lee–Carter model remain valuable tools, and in many settings they still perform well. However, the environment in which actuaries will work is changing. This web session will explore why we are moving beyond these traditional boundaries and how „Actuarial Learning“ is redefining forecasting.

Steps towards machine learning are driven by the need to handle high-dimensional data and nonlinear patterns that standard regression techniques cannot capture. To bridge this gap, we first consider ensemble methods, such as LightGBM, which outperform traditional actuarial models on complex tasks, such as predicting flood injuries.

Beyond ensembling, deep neural networks offer even stronger representational capacity, enabling us to model complex interactions directly from raw data. For instance, while the Lee-Carter model has been the gold standard for mortality forecasting, it often fails to capture cohort effects and cross-population heterogeneity. By adopting deep learning architectures, such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Long Short-Term Memory (LSTM) networks, we can achieve significantly higher predictive accuracy. We will also briefly discuss emerging developments, such as foundation models, which enable the use of pre-trained models in actuarial contexts where data may be limited.
Anmeldeschluss: 2026-11-30
Link: https://actuarial-academy.com/en/continuing-education/upcoming-trainings/detail/time-series-for-actuarial-modelling-with-machine-learning-e0564/