Abstract:
Literature on climate change and extreme events has found conflicting and often weak results on the evolution of economic damages related to natural disasters, although climate change is likely to bring about an increase in their magnitude (Van Aalst, 2006; IPCC, 2007, 2012). These studies usually focus on trend detection, typically employing mean regression techniques on yearly summed data. Using EM-DAT data, we enrich the analysis of natural disasters' risk by characterizing the behavior of the entire distribution of economic (and human) losses, especially high quantiles. We also envisage a novel normalization procedure to control for exposure (e.g. number and value of assets at risk, inflation), so to ensure spatial and temporal comparability of hazards. Employing moments and quantiles analysis and non-parametric kernel density estimations, we find a rightward shift and a progressive right-tail fattening process of the global distribution of economic damages both on yearly and decade aggregated data. Moreover, a battery of quantile regressions provide evidence supporting a substantial increase in the upper quantiles of the economic damage distribution (upper quantiles of human losses tend to decrease globally over time, mostly due to adaptation to storms and floods, but with a worrying polarization between rich and poor countries). Such estimates might be even conservative, given the nature of biases possibly affecting the dataset. Our results shows that mean regressions underestimate systematically the real contribution of the right tail of the damage distribution in shaping the trend itself.
