LOOKAHEAD AGENT ANALYSIS IN LONG-RUN FAIRNESS SIMULATION

Abstract

Bank lending has been used as an exploratory scenario to understand how "fair" short term lending policies fare in the long-run. Unfortunately, policies implemented in the short-run are not necessarily fair in the long-run. In this paper, we build off previous simulation studies to propose four one step lookahead lending agents. Lookahead checks the expected outcome of a lending policy before making a lending decision. We compare these lookahead policies against each other under relative improvement and active harm scenarios after multiple time steps. We find that the advantaged group and bank always benefit most from a maximum utility agent -- that maximizes bank profit -- while the disadvantaged group always benefits most from an unbounded equality of opportunity agent -- which maximizes bank profit under lending thresholds with equal true positive rate across the groups.