This paper discusses the question of electoral prediction based on politico-econometric models. It presents a brief historical sketch of this specific research domain in “political economy”, and a synthesis of the present model-based predictions for the French 2007 presidential elections. Because the reviewed models predict different and somewhat opposed results, we suggest to use an ex ante arbitrage, based on a simple indicator, the Figaro-Sofrès popularity index for the socialist party. The arbitrage between potential winners appears to be very clear. As this paper is written six weeks ante eventum, it can be seen as a kind of “natural experiment” in itself, useful to test the predictive capacity of our selected indicator.Cet article tra...