Skew-Adjusted Extremized-Mean: A Simple Method for Identifying and Learning From Contrarian Minorities in Groups of Forecasters

27 Pages Posted: 1 Mar 2022

See all articles by Ben Powell

Ben Powell

University of York

Ville Satopää

INSEAD - Technology and Operations Management

Niall J. MacKay

University of York

Philip Tetlock

University of Pennsylvania

Date Written: January 8, 2022

Abstract

Recent work in forecast aggregation has demonstrated that paying attention to contrarian minorities among larger groups of forecasters can improve aggregated probabilistic forecasts. In those papers, the minorities are identified using `meta-questions' that ask forecasters about their forecasting abilities or those of others. In the current paper, we explain how contrarian minorities can be identified without the meta-questions by inspecting the skewness of the distribution of the forecasts. Inspired by this observation, we introduce a new forecast aggregation tool called Skew-Adjusted Extremized-Mean and demonstrate its superior predictive power on a large set of geopolitical and general knowledge forecasting data.

Suggested Citation

Powell, Ben and Satopää, Ville and MacKay, Niall J. and Tetlock, Philip, Skew-Adjusted Extremized-Mean: A Simple Method for Identifying and Learning From Contrarian Minorities in Groups of Forecasters (January 8, 2022). Available at SSRN: https://ssrn.com/abstract=4004029 or http://dx.doi.org/10.2139/ssrn.4004029

Ben Powell (Contact Author)

University of York ( email )

Heslington
University of York
York, YO10 5DD
United Kingdom

Ville Satopää

INSEAD - Technology and Operations Management ( email )

Boulevard de Constance
77 305 Fontainebleau Cedex
France

Niall J. MacKay

University of York ( email )

Heslington
University of York
York, YO10 5DD
United Kingdom

Philip Tetlock

University of Pennsylvania ( email )

Philadelphia, PA 19104
United States

Do you have negative results from your research you’d like to share?

Paper statistics

Downloads
274
Abstract Views
1,310
Rank
205,434
PlumX Metrics