A prediction market system using SIPS and generalized LMSR for collective-knowledge-based demand forecasting

Abstract

This paper proposes and verifies a prediction market system for obtaining a forecast distribution of the demand quantity for a certain product in a future time period based on the collective knowledge of employees or customers themselves of a company. It first introduces a new type of prediction securities called the self-adjustable interval prediction securities (SIPS); SIPS are a set of winner-takes-all type contracts and each of which is tied to the future event that the actual demand quantity is included in a certain interval. The prediction intervals are mutually exclusive and collectively cover the whole feasible region of the demand quantity. What is unique to SIPS is that how the whole feasible region is partitioned into a set of prediction intervals is dynamically and adaptively self-adjusted so as to improve the accuracy of the output forecast distribution. It next generalizes the logarithmic market scoring rule (LMSR), which is one of the most popular market making algorithms for a prediction market, to SIPS. The paper also tests how the proposed prediction market system works in a laboratory setting and confirms that SIPS outperform the fixed interval prediction securities (FIPS) in terms of the accuracy of the output forecast distribution.