Extreme events are one of the main channels through which climate and socio-economic systems interact, and it is likely that climate change will modify the probability distribution of the losses they generate. The long-term growth models used in climate change assessments, however, cannot capture the effects of such short-term shocks. To investigate this issue, a non-equilibrium dynamic model (NEDyM) is used to assess the macroeconomic consequences of extreme events. This exercise allowed us to define the Economic Amplification Ratio, as the ratio of the overall production loss due to an event to its direct costs. This ratio could be used to improve the cost-benefit analysis of prevention measures. We found also that, unlike a Solow-like model, NEDyM exhibits a bifurcation in GDP losses : for each value of the capacity to fund reconstruction, GDP losses remain moderate if the intensity and frequency of extremes remain under a threshold value, beyond which GDP losses increase sharply. This bifurcation may partly explain why some poor countries that experience repeated natural disasters cannot develop. Applied to the specific issue of climate change, this model suggests that changes in the distribution of extremes may entail significant GDP losses in absence of specific adaptation. It suggests, therefore, that to avoid inaccurately low assessments of damages, researchers must take into account the distribution of extremes instead of their average cost and make explicit assumptions on the organization of future economies.