Aplicaciones actuariales mediante Gaussian Process Regression: Vida y no vida

David Rius Carretero; Salvador Torra Porras

doi:10.26360/2022_3

Authors

David Rius Carretero Universidad de Barcelona (España) https://orcid.org/0000-0002-8584-2703
Salvador Torra Porras Universidad de Barcelona (España) https://orcid.org/0000-0002-8786-8800

DOI:

https://doi.org/10.26360/2022_3

Keywords:

gaussian process, multivariate normal, covariance, Actuarial Science, distributions

Abstract

In this work, a brief introduction has been made on the Gaussian Process Regression (GPR) methodology and two applications in the Actuarial field. On the one hand, an interpolation exercise has been carried out on the PASEM Unisex 2020 mortality tables, concluding that the GPR is an excellent interpolation tool, and that it allows us a more adjusted pricing in the Life branch. On the other hand, the GPR has been integrated as a predictive measure for provisions in Non-Life branches, obtaining promising results. Finally, it is concluded that a GPR can be a useful instrument, as long as a good Kernel selection and a correct model training period are carried out.

Downloads

Download data is not yet available.

References

Ayuso. M., Corrales. H, Guillén. M, Pérez-Marín. A. M, Rojo. J. L. (2007). Estadística actuarial vida. Universitat de Barcelona. Boettiger, C. (2012). Basic regression in Gaussian processes. https://www.carlboettiger.info/2012/10/17/basic-regression-in-gaussian-processes.html

Flaxman, S., Gelman, A., Neill, D. B., Smola, A., & Vehtari, A. (2015). Fast hierarchical Gaussian processes. (Unpublished manuscript).

Lally, N. y Hartman, B. (2018). Estimating loss reserves using hierarchical Bayesian Gaussian process regression with input warping. Insurance: Mathematics and Economics, 82, 124-140.

Rasmussen, C. E., Williams, C. K. I. y Bach, F. (2005). Gaussian Processes for Machine Learning. MIT Press.

Ruitenberg, P. (2019). Adapting a Hierarchical Gaussian Process model to predict the loss reserve of a non-life insurer. University of Twente.

Snoek, J., Swersky, K., Zemel, R. S., & Adams, R. P. (2014). Input Warping for Bayesian Optimization of Non-Stationary Functions. ICML.