Özhamaratlı, Fatih; (2024) Innovative Approaches to Pension Planning: Generative Models to Multi-Agent Systems in a Heterogeneous Environment. Doctoral thesis (Ph.D), UCL (University College London).

Preview

Text Ozhamaratli_10185920_thesis.pdf Download (9MB) | Preview

Abstract

The shift from Defined Benefit (DB) to Defined Contribution (DC) pension schemes, along with changing demographics and work patterns, has made the task of planning and modelling pension ecosystems more complex. This research focuses on modelling the pension ecosystem using novel methodologies to plan for retirement finances. Initially, an exploratory analysis of the factors shaping the pension ecosystem was conducted, focusing on age and income dynamics; subsequently, a generative model utilising a joint distribution of age and income was devised. Reinforcement Learning (RL) and Deep Neural Networks were applied to train the lifetime portfolio optimisation and saving strategy selection policies for contributors with varying age and income trajectories. Calibrated Agent-based models (ABMs) of the pension environment were used to accommodate increasing heterogeneity. In the next phase, pension modelling was explored using an RL-optimised Multi-Agent System of the Pension Ecosystem, involving different actors. This approach enables our model to exceed the limits of hard-coded environment dynamics by allowing endogenous market dynamics and interactions between agents as part of the trained model. This research demonstrates how a set of increasingly complex methodologies can effectively address the challenges of a complex and heterogeneous pension environment, providing valuable insights for financial planning and policy-making. This makes the transition from a one-size-fits-all approach to personalised solutions possible.

Download activity - last month

Export as XMLJSONCSV

Download activity - last 12 months

Export as XMLJSONCSV

Downloads by country - last 12 months

Export as XMLJSONCSV

Archive Staff Only

View Item