Elections and the (first) derivative of unemployment: the turnaround strategy


By Guillaume Allègre, g_allegre

A ministerial adviser recently explained to me what he thinks is the strategy of the French President on macroeconomic management and unemployment, which could be called a turnaround strategy: “In relation to the presidential elections, the goal is to reduce unemployment in 2016-2017. The way people vote is based on the way unemployment has been changing just in the last year or even the last 6 months. Like for Jospin in 2002.” The belief that for unemployment and the economy in general what counts is the derivative, i.e. the recent evolution and not the actual level, has deep roots in the technocratic-political milieu: “it’s the derivative, stupid!” is the new “it’s the economy, stupid!“ (the maxim of Bill Clinton’s election strategist in 1992).

This belief stems in part from an intuition confirmed by a well-known psychological experiment. Participants in the study were subjected to two painful experiences during which one of their hands is immersed in ice water. One version lasts 60 seconds and the other 90 seconds. In the second experiment, the first 60 seconds are the same as in the first, while the 30 added seconds are a bit less painful (the experimenter pours some warm water into the container). Later, the participants must choose which of the two experiments to repeat: 80% chose the longer one. This seems irrational, because in the longer experiment the total amount of pain is greater. To an objective observer, this is what should count (“the area under the curve, or the integral”). But the participants have a selective memory: they are more strongly influenced by the representative moments of the experience and in particular here by the improvement at the end of the test. Daniel Kahneman, the 2002 Nobel Prize-winner in economics for his work on biases in judgment, which is popularized in a book that can be found here, distinguishes two representative moments during an unpleasant episode: the peak of suffering and the end [1].

Economists, especially in America, have developed econometric models of electoral forecasts to estimate the links between election results and the economy. The popularity of these models varies with their predictive power for the election: in 1992, half of the models predicted an easy re-election for George Bush; in 1996, the re-election of Clinton was reliably predicted; but in 2000, virtually all the models forecast a landslide victory for Al Gore … And the model that had the closest forecast in that election (0.6%) was off by 5 points in the next one. Of course, thanks to the proliferation of predictions, it is always possible to find a model with a good record for the time-being, such as Paul the Octopus (see Wiki).

Despite this motley record, these politico-econometric models have been imported into France. In their generic form, they attempt to explain the percentage of the vote going to a candidate or a party based on economic variables (GDP, unemployment, or levels or changes in income) and political variables (popularity of the President and the Prime Minister). The vast majority of models adopt as an economic variable changes in unemployment over a relatively short horizon, on average one year. The conclusion drawn from these empirical estimates is that French voters seem to have limited memories (Dubois, 2007).

But these studies are faced with a major problem: the low number of observations (nine presidential elections and thirteen legislative elections between 1958 and 2011). “We don’t vote often enough to suit the econometricians,” says Lafay (1995) [2]. In other words, the law of large numbers cannot be applied in this type of configuration. This is compounded by the fact that the number of variables that change in the context of these elections is almost as high as the number of elections (the existence of a government of multiple party “cohabitation”; legislative elections on their own or coupled to the presidential elections; the presence or absence of an incumbent in the presidential election; parliamentary elections held before the deadline; the presence or absence of a leftist candidate in the second round of the presidential elections; the importance of tactical voting when there are three candidates in the second round of legislative elections [triangulaires]; etc.).

There are other technical problems confronting the econometricians. In a comprehensive review of the literature analyzing 71 political-economic studies on voting in France between 1976 and 2006, Dubois describes the way these problems are handled – “if at all” – as “relatively frustrating”. Just as in the United States, the predictions meet with “varied success”. There is also the problem of what econometricians call “endogeneity”: the politico-economic models attempt to explain or predict the outcome of elections using economic variables (unemployment) and the popularity of the executive. However, there is little doubt that the popularity of the executive depends in part on unemployment levels and trends: given this, the lack of significance of changes in the longer-term economic variables may be explained by the fact that their impact is already included in the popularity of the executive. In short, these empirical studies are not sufficient to conclude that in economic terms, voters have short memories.

In the words of Kahneman, a machine for jumping to conclusions is at work: an intuition (the memory of voters is selective) that relies on psychological studies (whose object is distant) and is confirmed by econometric studies (not robust and therefore merely reproducing the researchers’ a priori assumptions). The story told is consistent, and it seems to be supported by facts … Upon reflection, it may seem scary that this kind of rhetorical cocktail is influencing the actions of politicians. This is even more frightening since, from an outside observer’s viewpoint and from the perspective of social welfare and hence the goals of public policy, what matters is obviously the level of unemployment over several years (its integral) and not the way it has changed in the last year (its first derivative)!

Many rules have been implemented at the European level, and now the national level too, to prevent the politicians heading up government from trying to win elections by pursuing policies that, while they may reduce short-term unemployment, also build up long-term deficits. From the Maastricht criteria (government deficit of less than 3% of GDP) to the recent European multiannual financial framework, these rules are justified by the belief that politicians are encouraged to pursue a lax fiscal policy since it does not take into account future generations, who, by construction, don’t vote. But if governments begin to believe that it is short-term economic developments that count, then the incentives are reversed, especially if it is easier to reduce unemployment after having first increased it, which would lead to a trajectory of weak growth and of excessively high unemployment. [3] In this case, the solution cannot come from governance through new binding rules, which in any case have so far proved to be ineffective. It is necessary to rely on the fact that this kind of turnaround strategy can work in electoral terms only if the citizens fail to understand that they are being manipulated. Exposing the manipulation is then more efficient than implementing rules. Duly noted.

[1] Consequently, those who follow this theory today should also deal with unemployment at its peak, and not merely with the way it is changing at the end of their mandate.

[2] Lafay J.-D. 1995, “Note sur l’élection présidentielle de 1995 et les apports de l’analyse économétrique des comportements électoraux”, mimeograph, LAEP, University of Paris 1. Cited by Dubois.

[3] This post – link – emphasizes that it was possible to achieve the same ratio of debt to GDP in 2032 by taking a path that would have reduced unemployment in the euro zone by 3 points in 2013.