State and Politics

The results of the 2009 German federal elections were once again quite an eye-opener for demoscopic analysts who for years have been endeavouring to reliably capture the mood of the electorate. Now however there are some new methods that bode well for the future.

In order to avoid any huge losses of face on election days, institutes like Infratest-Dimap or the Forschungsgruppe Wahlen regularly send their people out onto the streets in their hundreds, armed with questionnaires, to try and find out right up to the last minute how people have voted. The aim is to ensure that as soon as the polling stations have closed at six o’clock in the evening there are at least a few reasonably accurate projections ready to be flashed across the nation’s TV screens. This does in fact work quite well as a rule. The real problem however are the prognoses before the election – they are becoming more and more unreliable.

“Election Stock Markets” – a new approach to election analysis

Improvement is on its way however in the form of virtual “Election Stock Markets” that are now being tried out everywhere. These so-called “Electronic Stock Markets” (ESMs) have been around for quite some time in the USA, but in Germany they are thought to sail too close to the wind due to the image of gambling they conjure up. They have not just borrowed the name from the realm of economics, but their methods have also been lifted from the field of business science, to be more exact, from experimental economics.

True to the theories of the Nobel laureate for economics, Friedrich August von Hayek, according to which market prices reflect individual assessments and information on possible market developments, the “Election Stock Markets” trade in futures i.e. predictions on the results of the election. The fact that real money is being “gambled with” – the stake for these securities is as a rule somewhere between 5 and 50 euros – is to guarantee that the market participants allow maximum possible rationality to prevail. After all, the more accurately one’s own portfolio reflects the actual final election result, the higher the dividend pay-out or lower the loss, as the case may be.

An astounding success

The success is astounding. After a few initial difficulties, setbacks and adjustments the results are now not to be sneezed at. The sociologists, Dr. Christian Ganser and Jochen Groß from Munich’s Ludwig-Maximilians-Universität (LMU), arrived at a particularly close prediction for the results of the federal elections. They operate the virtual stock exchange Wahlstreet, in which 724 “securities traders” took part in the run-up to the elections, and they succeeded in predicting which party got which share of the vote more accurately than all the other opinion research institutes. On 19th September, one week before the actual election, the so-called “Sunday Polls”, which are regularly used to predict party preferences, had a standard deviation of only 0.9 per cent compared to 2 per cent on “official” demoscopic surveys.

Although the Munich university “punters” even managed to narrow their result down to a lean 0.79 percentage point on 27th September, to their dismay they had to bow out to the Politikmarkt in the end. This is the “Election Stock Market” of the FAZ.net that was set up by business data processing specialists from the Technical University of Dortmund and they narrowed it down to a sensational 0.71 per cent deviation. The Munich “punters” however are not going to admit defeat so easily, “ It has to be said that the closing rates of the stock exchanges were apparently used for this,” complains Jochen Gross. “As far as I have been informed the Politikmarkt was open until 6 p.m. on election day, the “Wahlstreet” only until 4 p.m. On election day in particular there is always a lot of action on the markets. Comparing their result with our “Sunday Polls” is not fair – they were conducted at a much earlier point in time.” In fact Politikmarkt closed at 3 p.m....

The hour of the linguists has struck

Nevertheless the mood among the researchers is on the whole euphoric. The only fly in the ointment however was the poor performance of their competitor, “Betfair”. Above all because it is the exchange that trades in the highest stakes, which, if we are to believe the theory, is supposed to have a positive influence on accuracy. A more exact evaluation of the data is to shed more light on the reasons.

In the meantime the hour of the linguists has struck. This was the first time a German election underwent a systematic, computer-based matrix analysis. Semtracks is the name of the project brought into being by the Swiss linguist, Noah Bubenhofer, who set up a research group for this at the Heidelberg Center for American Studies in May 2008. In the run-up to the elections Bubenhofer, in collaboration with his colleague, Joachim Scharloth, analysed parliamentary speeches, press releases, political speeches and electoral party programs in a laboratory for computer-based meaning research.

By way of illustration the researchers developed “Word Clouds” in which words in large print are depicted that are typical of a particular party and used a lot by it. “This is why Germany is going to be governed by a coalition comprising of parties which are also close to each other on a deep semantics level,” is the verdict given by Semtracks on the 2009 German federal elections. “The coalition talks are going to be controversial in the area of financial and fiscal policy – at least that is what the semantic analysis says. From a deep semantics point of view it is good to hear that we are not going to have a revamp of the Grand Coalition. After four years of joint government the commonalities seem to have been exhausted.”