Are Nancial Crashes Predictable?

{ We critically review recent claims that nancial crashes can be predicted using the idea of log-periodic oscillations or by other methods inspired by the physics of critical phenomena. In particular, the October 1997 \correction" does not appear to be the accumulation point of a geometric series of local minima. It is rather tempting to see nancial crashes as the analogue of critical points in statistical mechanics, where the response to a small external perturbation becomes innnite, because all the subparts of the system respond cooperatively. Similarly, during crashes, a large proportion of the actors in a market decide simultaneously to sell their stocks. If one furthermore postulates that this critical point is decorated by \log-periodic" oscillations (for which there is a recent upsurge of interest in a wider context 1]), then one can interpret the oscillations seen on markets as precursors to predict the crash time t c , which should be the point where these oscillations accumulate. Intriguing hints supporting this scenario have initially been reported in 2, 3], and more recently in 4, 5], where it was explicitly claimed that the October 1997 correction was predicted ex-ante (see also 6]). As a proof of this, the implementation of a winning strategy was reported in two papers published in physics journals 7]. In view of the considerable echo that these claims have generated, in particular in the physics community 8], we feel that it is important to temper the growing enthusiasm by discussing a few facts. In general, the unveiling of a new phenomenon either results from a strong theoretical argument suggesting its existence, or from compelling experimental evidence. In the present case, there is no convincing theoretical model which substantiates the idea that crashes are c EDP Sciences