JK MacKie-Mason, A Osepayshvili, DM Reeves, and MP Wellman

Fourteenth International Conference on Automated Planning and Scheduling, pages 244-252, 2004.
Copyright (c) 2004, American Association for Artificial Intelligence. All rights reserved.


In a market-based scheduling mechanism, the allocation of time-specific resources to tasks is governed by a competitive bidding process. Agents bidding for multiple, separately allocated time slots face the risk that they will succeed in obtaining only part of their requirement, incurring expenses for potentially worthless slots. We investigate the use of price prediction strategies to manage such risk. Given an uncertain price forecast, agents follow simple rules for choosing whether and on which time slots to bid. We find that employing price predictions can indeed improve performance over a straightforward baseline in some settings. Using an empirical game-theoretic methodology, we establish Nash equilibrium profiles for restricted strategy sets. This allows us to confirm the stability of price-predicting strategies, and measure overall efficiency. We further experiment with variant strategies to analyze the source of prediction’s power, demonstrate the existence of self-confirming predictions, and compare the performance of alternative prediction methods.