🔧 A Dynamic Bayesian Model That Blends Polls and Fundamentals

A dynamic Bayesian forecasting model is introduced that explicitly combines published pre-election public-opinion polls with information from fundamentals-based forecasting models. The approach is built for multiparty systems and is structured to produce probabilistic statements about election-relevant quantities beyond simple vote shares.

🧾 What the Model Can Estimate

• Probabilities of a party winning a plurality of votes
• Probabilities that particular coalitions secure a parliamentary majority
• Uncertainty around those probabilities, calibrated by observed polling variation

📊 Tested on Two 2017 Elections

The model was used to generate two ex ante forecasts for elections that took place in 2017. These forecasts combine historical and current polling data with fundamentals to produce forward-looking probability statements about electoral outcomes.

📈 Key Results: Improved Accuracy and Better-Calibrated Uncertainty

• Outperforms fundamentals-only forecasting models on point-prediction accuracy
• Produces better-calibrated measures of uncertainty (probabilistic statements match realized outcomes more closely)

🌍 Where This Applies

The model can be applied to any multiparty electoral setting, provided that historical and current polling data are available. It is particularly useful when probabilities for pluralities or coalition majorities are the primary objects of interest.