A seminar by Professor Bonsoo Koo from Monash University
Title: Sequential Decision-Making for Life-Cycle Retirement Pension Optimisation via Deep Reinforcement Learning
Abstract: Defined-contribution pension systems require households to manage retirement wealth under uncertainty in financial markets, longevity, and policy settings. This paper develops a sequential life-cycle optimisation framework that jointly models pension withdrawals for consumption, portfolio allocation, reverse-mortgage borrowing, and bequest. A key feature is that household risk preferences evolve endogenously over time in response to realised consumption behaviour and macroeconomic conditions, with the classical fixed-preference specification recovered as the special case in which the preference state remains constant. To solve the resulting high-dimensional problem, we use a Deep Deterministic Policy Gradient (DDPG) algorithm that operates directly on continuous action variables and high-dimensional state representations without discretising the state and action spaces. The model is calibrated to the Australian retirement system, incorporating Age Pension means-testing and the Home Equity Access Scheme (HEAS). Across heterogeneous retiree profiles and macroeconomic scenarios, the framework produces stable and economically interpretable retirement strategies. Within the calibrated sequential specification, the estimated preference dynamics map into differentiated consumption, investment, and borrowing paths across retiree profiles, while housing wealth remains central to retirement consumption stability and the role of HEAS borrowing within the Australian institutional setting.
For further information, please contact RSFAS Seminars.
All information collected by the University is governed by the ANU Privacy Policy.

