Faculty
Department of Statistical and Actuarial Sciences
Supervisor Name
Dr. Hyukjun Gweon & Dr. Shu Li
Keywords
Variable annuity, active learning, bias-corrected bagging
Description
The variable annuity (VA) is a modern insurance product that offers certain guaranteed protection and tax-deferred treatment. Because of the inherent complexity of guarantees’ payoff, the closed-form solution of fair market values (FMVs) is often not available. Most insurance companies depend on Monte Carlo (MC) simulation to price the FMVs of these products, which is an extremely computational intensive and time-consuming approach. The metamodeling approach can be used to circumvent the heavy computation.
In the modeling stage, the bagged tree method has proved to outperform other parametric approaches. Also, a bias-corrected (BC) bagging model was tried and showed significant improvement for prediction performance.
The active learning framework is considered when the number of unlabeled data is large, the budget for labeling is limited and labels are expensive obtain, like in our scenario. Two active learning approaches, the weighted random sampling (WRS) and repeated random sampling (RRS) have been tested on selecting representative contracts and showed good performance.
For the summer research program, we investigate whether the BC bagging model is also effective in the active learning framework.
Acknowledgements
I would like to express my deepest appreciation to my supervisors Dr. Hyukjun Gweon and Dr. Shu Li for their guidance and support. Thank you Western USRI program! We also appreciate the funding support from the Multi-hazard Risk and Resilience Interdisciplinary Development Initiative for this USRI project.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License
Document Type
Poster
Bias-corrected Bagging in Active Learning with an Actuarial Application
The variable annuity (VA) is a modern insurance product that offers certain guaranteed protection and tax-deferred treatment. Because of the inherent complexity of guarantees’ payoff, the closed-form solution of fair market values (FMVs) is often not available. Most insurance companies depend on Monte Carlo (MC) simulation to price the FMVs of these products, which is an extremely computational intensive and time-consuming approach. The metamodeling approach can be used to circumvent the heavy computation.
In the modeling stage, the bagged tree method has proved to outperform other parametric approaches. Also, a bias-corrected (BC) bagging model was tried and showed significant improvement for prediction performance.
The active learning framework is considered when the number of unlabeled data is large, the budget for labeling is limited and labels are expensive obtain, like in our scenario. Two active learning approaches, the weighted random sampling (WRS) and repeated random sampling (RRS) have been tested on selecting representative contracts and showed good performance.
For the summer research program, we investigate whether the BC bagging model is also effective in the active learning framework.