A highly connected world opens plenty of new possibilities. One rising phenomena is the “wisdom of the crowds” that can be efficiently used, since the web lowered cost of information collecting and the reaching of large groups. An application of this potential is for example group forecasting. Group forecasting can happen from relatively small numbers as in Delphi Studies to an extensive set of participants as in polls. However there exist a number of other methods, with advantages and disadvantages each. Prediction markets are one new method for information aggregation and generates sometimes impressively accurate results, but also recently perform only average or below. SciCast (http://blog.scicast.org) is an example of a recent prediction market that had over 11,000 registrations and aggregated more than 129,000 single estimations on smaller events as well as major breakthroughs in science.
Based on the publicly available data set of SciCast, the aim of the thesis is to identify and proof the existence of several biases in the forecasts of the prediction market or in single trader. Using methods of descriptive and inferential statistics as well as data mining methodologies, the effects of biases such as the conjunction fallacy or the bandwagon effect shall be found and their effect on the market forecast interpreted.
The applicant should be comfortable to work a lot with and in data, should be used to relational databases and bring experience in common tools to analyze data such as e.g. R. A good knowledge in statistics as well as well as endurance an autonomy are expected.
Language: English or German