In the past 40 years, few papers have focused on voting models of German elections. Certainly, Kirchgässner (Reference Kirchgässner1977; Reference Kirchgässner1991) or Rattinger (Reference Rattinger1991), developed Vote-Popularity (VP) functions predicting the German ruling parties’ performance based on unemployment as a dominant economic variable. Furthemore, Jérôme, Jérôme-Speziari and Lewis-Beck (Reference Jérôme, Jérôme-Speziari and Lewis-Beck2001) explored the possibility of a “yardstick” competition between French and German economic votes.
However, until 2013, only a handful of models attempted to forecast German elections. Using Vote Functions, Jérôme, Jérôme-Speziari and Lewis-Beck (Reference Jérôme, Jérôme-Speziari and Lewis-Beck1998) initially developed forecasting model where the rate of unemployment and the popularity of the FDP were the main predictors.
This seminal model correctly forecasted the CDU/CSU-FDP defeat. Thereafter, the same method has been successfully applied in 2005 and 2009 but failed in 2002 (Jérôme, Jérôme-Speziari and Lewis-Beck, Reference Jérôme, Jérôme-Speziari and Lewis-Beck2002, 2005 and Reference Jérôme, Jérôme-Speziari and Lewis-Beck2009). At last, in 2013 (see PS: Political Science & Politics 46(03)) the authors opted for a SUR (Seemingly Unrelated Regression) model in order to predict both vote shares and suitable coalitions in a context where hinge parties (FDP and Greens) and new parties (AFD) were increasingly key players. Here again, the incumbent’s political and economic accountability (Key Reference Key1966) still plays a major role except for “peripheral” political parties.
On their side, Norpoth and Gschwend (2000; Reference Norpoth and Gschwend2003) built a forecasting model where the desire for one of the main party’s to provide the Chancellor predicts the election result. This model is known as the “Chancellor model” (Norpoth and Gschwend Reference Norpoth and Gschwend2010; Reference Norpoth and Gschwend2013). While Jérôme-Speziari and Lewis-Beck is rather retrospective, Norpoth and Gschwend’s Chancellor model is mainly prospective. With a view to the next elections, the main predictor is the Chancellor support. Additionally, the model includes a second variable giving the partisan attachment and a third one reflecting the incumbent “fatigue,” in other words, the cost of ruling. Such a configuration allows provision of forecasted vote shares and probabilities for each feasible coalition.
In this 2017 mini symposium, we welcome two new teams with alternative methods. Mark Kayser and Arndt Leininger follow the tradition of pooled time series models using disaggregated or regional data (see Rosenstone (Reference Rosenstone1983) and Campbell (Reference Campbell1992) in the US case or Jérôme, Jérôme and Lewis-Beck (Reference Jérôme, Jérôme-Speziari and Lewis-Beck1999) for French legislative elections). Their political economy Länder-based model underlines the relevance of regional political and economic data to predict the partisan structure of the next Bundestag and who could be the next Chancellor.
Lastly, Andreas Graefe presents a combined forecast based on the PollyVote research project initially launched in 2004 (Cuzàn, Armstrong, and Jones Reference Cuzán, Scott Armstrong and Jones2005) to forecast US elections. In the German Case, a first attempt has been made during the 2013 elections. Gathering predictions issued from polls, prediction markets, experts and models, and treating them on average, PollyVote provides vote-share forecasts for each German party.
If German general elections were held in early March, three models out of four predict the CDU/CSU could win despite a significant loss of votes compared with 2013 (see table 1). However, Norpoth and Gschwend are dissonant to the extent that they forecast a 83% chance of winning for a coalition (red-red-green or “Traffic Light”) led by Martin Schulz.
Table 1 Forecast Models in this Symposium
On balance, the Grand Coalition could be reappointed but Angela Merkel seems to be on borrowed time.
