Combining forecasts for elections: Accurate, relevant, and timely

This paper increases the efficiency and understanding of forecasts for Electoral College and senatorial elections by generating forecasts based on voter intention polling, fundamental data, and prediction markets, then combining these forecasts. The paper addresses the most relevant outcome variable, the probability of victory in state-by-state elections, while also solving for the traditional outcomes, and ensuring that the forecasts are easy to update continuously over the course of the main election cycle. In an attempt to maximize both these attributes and the accuracy, I create efficient forecasts for each of these three types of raw data, with innovations in aggregating the data, then correlate the aggregated data with the outcomes. This paper demonstrates that all three data types make significant and meaningful contributions to election forecasting. Various groups of stakeholders, including researchers, election investors, and election workers, can benefit from the efficient combined forecasts defined in this paper. Finally, the forecast is tested on the 2012 elections and excels out-of-sample.