Predicting the probability of uncertain future events is critical in many decision scenarios, especially when we deal with complex organizational and social problems in a large scale involving a number of people. However, the information related to the event to be forecasted is typically dispersed in the group as individual opinions, knowledge and other forms. It is critically needed to aggregate these dispersed information to improve prediction accuracy. In efficient markets, market price can be utilized to aggregate all of the commodity related information. Prediction Markets (PMs) thus can be a group information aggregation mechanism by providing a market place for individuals to trading on the event related contracts whose payoff are defined as dependent on the realization of the event. In PMs, all the participants trade on the clearly defined event futures. The event related information can be reflected in the contract price through the trading behavior of participant. Some experiments have witness the power of PMs as a efficient way of predicting future events. However, most of trading mechanisms employed in current PMs are based on either Continuous Double Auction (CDA) or Logarithmic Marketing Scoring Rules (LMSR). PMs with CDA often fall victim to thin market problem and PMs with LMSR et al. cause additional cognitive barriers for non-experienced participants. Potential market manipulation also decreases the decision makers’ credibility on PMs. We propose Fixed Odds Betting as a mechanism of Prediction Markets in this thesis. The main contributions of this thesis are as follows: First, we propose a new prediction market based on fixed odds betting and analyze its properties. Fixed Odds Betting Prediction Markets are easy to use and free of manipulation. Second, we propose a method based on paradigm of computational experiment to compute the mean belief of the crowds from the given market odds and final bet share on each side of the binary event. This means belief can serve as an indicator of the event probability. Third, we hypothesize that there exists a mapping between event probability and collective belief distribution on the event. We formulate this mapping by analyzing its statistical properties and propose an approximate approach to estimate the parameters in this model. Fourth, we verify these two different approaches using empirical data from an online fixed odds betting data set. The results give us some implications: (1) collect...
修改评论