Fundamental models for forecasting elections at the state level

This paper develops new fundamental models for forecasting presidential, senatorial, and gubernatorial elections at the state level using fundamental data. Despite the fact that our models can be used to make forecasts of elections earlier than existing models and they do not use data from polls on voting intentions, our models have lower out-of-sample forecasting errors than existing models. Our models also provide early and accurate probabilities of victory. We obtain this accuracy by constructing new methods of incorporating various economic and political indicators into forecasting models. We also obtain new results about the relative importance of approval ratings, economic indicators, and midterm effects in the different types of races, how economic data can be most meaningfully incorporated in forecasting models, the effects of different types of candidate experience on election outcomes, and that second quarter data is as predictive of election outcomes as third quarter data.