Constrained defined-contribution pension management: A reparameterized neural network approach
Yueman Feng, Wenyuan Li
Abstract
This paper studies constrained defined contribution (DC) pension management under a two-factor financial market model. The pension participant is allowed to invest in cash, two nominal bonds, an inflation-linked bond, and a stock index. Moreover, they can also purchase term life insurance to manage pre-retirement death risk. Unlike traditional literature, we consider a pension member restricted to a general non-empty, closed, and convex trading constraint set that includes no short-selling, no borrowing, and portfolio-mix constraints as special cases. Such restrictions reflect common limits on pension-account trading. To solve the problem, we construct a family of unconstrained artificial markets, in which the original trading constraints are satisfied by adjusting the drift terms of the risky assets. Using a duality transformation, we derive the Lagrangian dual problem and propose a reparameterized neural network method to compute lower and upper bounds on the primal value function. We calibrate our model with U.S. data using a Kalman filter method, and find that our method produces a smaller duality gap than the benchmark methods. We also compare the constrained results with the unconstrained explicit solution and find that trading constraints substantially change wealth accumulation, portfolio strategies, and insurance demand. This highlights that unconstrained solutions may provide misleading guidance for pension management in practice.
Read the AI summary and key takeaways for traders on WOBR Quant Research.