François Lévêque (Ecole des Mines de Paris)

How to predict the probability of a nuclear accident using past observations? How Fukushima Dai-ichi accident changed the probability of nuclear accidents?. Many models and approaches can be used to answer these questions, however their suitability depends two critical elements: the definition of nuclear accident and the time period chosen. In this paper, we have compared three Poisson models taken two definitions of nuclear accidents from 1952 to 2012. We found that if we use a dynamic model, the estimates change dramatically. Thus ignoring the temporal nature of the data can lead to wrong estimates. In fact, using a Poisson Exponentially Weighted Moving Average (PEWMA) we found a substantial increase in the risk of a core meltdown accident for the next year in the world, owing to the new major accident that took place in Japan in 2011. Keywords: nuclear risk, nuclear safety, time series count data