A Tax-Efficient Model Predictive Control Policy for Retirement Funding

TL;DR

This paper introduces a tax-efficient, model predictive control approach for retirement funding that optimizes withdrawals and transfers, adapting annually to uncertainties and changing rules to sustain inflation-adjusted consumption.

Contribution

It develops a novel MPC-based retirement funding policy that uses convex optimization for planning and adapts yearly to uncertainties and policy changes.

Findings

The MPC policy effectively manages retirement funds under uncertainty.

Simulation shows improved sustainability of inflation-adjusted consumption.

The approach is computationally efficient and adaptable to changing conditions.

Abstract

The retirement funding problem addresses the question of how to manage a retiree's savings to provide her with a constant post-tax inflation adjusted consumption throughout her lifetime. This consists of choosing withdrawals and transfers from and between several accounts with different tax treatments, taking into account basic rules such as required minimum distributions and limits on Roth conversions, additional income, liabilities, taxes, and the bequest when the retiree dies. We develop a retirement funding policy in two steps. In the first step, we consider a simplified planning problem in which various future quantities, such as the retiree's remaining lifetime, future investment returns, and future inflation, are known. Using a simplified model of taxes, we pose this planning problem as a convex optimization problem, where we maximize the bequest subject to providing a constant inflation adjusted consumption target. Since this problem is convex, it can be solved quickly and reliably. We leverage this planning method to form a retirement funding policy that determines the actions to take each year, based on information known at that time. Each year the retiree forms a new plan for the future years, using the current account values and life expectancy, and optionally, updated information such as changes in tax rates or rules. The retiree then carries out the actions from the first year of the current plan. This update-plan-act cycle is repeated each year, a general policy called model predictive control (MPC). The MPC retirement policy reacts to the effects of uncertain investment returns and inflation, changes in the retiree's expected lifetime or external income and liabilities, and changes in tax rules and rates. We demonstrate the effectiveness of the MPC retirement policy using Monte Carlo simulation.