Forecasting election outcomes using market-based indicators

The use of market-based indicators to forecast closed door decisions can be traced as far back as the 1503 papal conclave. In terms of open national elections, markets on the outcome of the US Presidential election can be traced to at least 1868. It is only since the 1930s, however, that scientific opinion surveys have been used to help predict election outcomes. More recently econometric models, citizen forecasts (asking citizens who they expect to win, as distinct from who they intend to vote for), expert judgement techniques, index models (determining variable weights using candidate qualities and biographies) and social media sentiment indices have gained rising prominence. Growing attention has also in recent years been paid to the value of combining forecasts derived from different forecasting methodologies. Even within a distinct forecasting methodology, however, such as market-based forecasting, there are a range of platforms, including exchange market models, academic crowd wisdom markets and traditional wagering markets. There is also a growing emphasis on the use of weighted and de-biased forecasting methods. These techniques have in addition been used outside of traditional elections to help forecast uncertain outcomes as diverse as the Annual Academy Awards and decisions of the US Supreme Court. In this paper, quantitative and qualitative data is collected from a range of sources to compare and contrast the predicted outcome of a range of high-profile event outcomes between and including 2012 and 2016, with the purpose of distinguishing the relative contribution and value added of each of these forecasting methodologies, considered individually and in combination. The data is collected from primary sources, including as noted a range of prediction markets, opinion polls, citizen forecasts, index models, econometric models and social media flags. In this way, the paper seeks to maximize predictive power using a wide set of alternative forecasting methodologies, while asking whether market-based predictors can act as a close proxy for the combined forecasting power of all.