Problem
Electoral data, e.g. polls, are subject to a variety of biases and uncertainties that generate challenges to combining them into coherent, probabilistic forecasts.
Solution
We design an election forecasting engine that delivers probability statements for arbitrary electoral scenarios. It is trained on the aforementioned heterogeneous electoral data and accommodates different levels of granularity, an evolving party landscape and local seat assignment rules.
We used the engine to forecast recent Spanish general elections. The engine proved to be adept at capturing uncertainy in the distribution of parliamentary seats both at national and regional level.
