Abstract
To date, most demand forecasting methods study some historical data on the demand (and maybe other related variables as well), find certain patterns or trends, and extrapolate them into the future to obtain a demand forecast. In the recent rapidly changing market environment, however, it is often difficult to prepare sufficient historical data, and even when it is available, the patterns or trends extracted from the data rarely last long enough to be used to hit the target in the future. Thus, this paper proposes a novel demand forecasting method which will work effectively even in such circumstances where extrapolateable demand patterns are hardly available. The proposed method uses a market mechanism, called the prediction market system, to aggregate dispersed knowledge of a company's sales people regarding the future demand of a product into a continuous forecasted demand distribution. In order to make it work effectively and smoothly, the paper introduces a new type of prediction security and an original market maker algorithm suitable for the security type, and furnishes them to an intra-firm prediction market system. As a result, the transactions can be conducted with the computerized market maker whenever necessary, and hence sufficient liquidity is supplied into the market even when the number of traders is small. Further, the market maker can output an aggregated demand forecast of the sales people as a continuous distribution at any time. An agent simulation model, where each trader has a log-utility function, is also developed to show how the proposed method works, and running it reveals the parameters upon which it depends how quickly the forecasted demand distribution converges and how stable the forecasting process is.
