A1546
Title: Fixed event forecasting at multiple lead times
Authors: Markus Fritsch - University of Passau (Germany) [presenting]
Harry Haupt - University of Passau (Germany)
Joachim Schnurbus - University of Passau (Germany)
Abstract: Fixed event forecasts target a particular event at a specific future date. Examples are the outcomes of sports events, wind speeds of hurricanes when they hit inhabited areas, or the results of political elections. As rolling event forecasting, where each forecast refers to a different time period, is by far more common, we first detail the particularities of fixed event forecasting. Second, we review testable and theoretical conditions for the efficiency of fixed event forecasts. Third, a nonparametric framework for fixed event forecasting is developed and forecasting the results of German state elections is illustrated for multiple election cycles in selected states. We compare the results to traditional election polls and methods from modern time series analysis. Additionally, the benefits of averaging across multiple estimation windows are investigated for the different approaches. Our results suggest that off-the-shelf statistical models outperform conventional election polls, even on days when new polls become available.
