“Constrained No Regret Learning”

Ye Du, Southwestern University of Finance and Economics

Ehud Lehrer, Tel Aviv University and INSEAD

In this paper, we investigate the dynamic decision making problem when there are a set of linear constraints on the empirical frequency of actions actually played. To cope with this, we introduce the constrained no-regret learning model.  In this model, the set of alternative strategies, with which a dynamic decision mechanism is compared, is the set of stationary mixed actions that satisfy the set of linear constraints. We show that there is indeed a constrained no-regret dynamics.