Heavy tailed probability distributions for non-Gaussian simulations with higher-order cumulant parameters predicted from sample data

Abstract

Forecasting of extreme events and phenomena that respond to non-Gaussian heavy-tailed distributions (e.g., extreme environmental events, rock permeability, rock fracture intensity, earthquake magnitudes) is essential to environmental and geoscience risk analysis. In this paper, new parametric heavy-tailed distributions are devised starting from the exponential power probability density function (pdf) which is modified by explicitly including higher-order “cumulant parameters” into the pdf. Instead of dealing with whole power random variables, novel “residual” random variables are proposed to reconstruct the cumulant generating function. The expected value of a residual random variable with the corresponding pdf for order G, gives the input higher-order cumulant parameter. Thus, each parametric pdf is used to simulate a random variable containing residuals that yield, in average, the expected cumulant parameter. The cumulant parameters allow the formulation of heavy-tailed skewed pdfs beyond the lognormal to handle extreme events. Monte Carlo simulation of heavy-tailed distributions with higher-order parameters is demonstrated with a simple example for permeability.