Analyzing prediction market trading behaviour to select Delphi-experts

The selection of experts for Delphi studies is crucial for the quality of the forecast results and the information taken into account. In the past, this has usually been done by selecting participants according to their reputation, although this approach is questionable in terms of reaching the most knowledgeable participants having new, relevant and valid information. In this context, this paper aims to propose to operate a prediction market alongside Delphi studies and select participants based on their trading behaviour in the market for the Delphi study.

Based on more than three years of historical prediction market trading data, the authors verify attributes that indicate insightful trades, as previously discussed in the finance literature, by using regression and classification trees.

The paper contributes attributes of trading behaviour that are theoretically derived from literature and potentially related to informed traders. These are tested and evaluated on historical prediction market data. Especially, the trading volume, the spread at the moment of trading and the market maker attribute seem to predict informed traders the best.

Algorithms based on identified attributes can be used to objectify the selection of experts for Delphi studies with potential gains in terms of the amount of information considered.