Bandwidth selection for kernel density estimation of fat-tailed and skewed distributions

Applied researchers using kernel density estimation have worked with optimal bandwidth rules that invariably assumed the reference is Normal (optimal only if true underlying Normal). We offer four new rules-of-thumb based on other infinitely supported distributions: Logistic, Laplace, Student's t and Asymmetric Laplace. Additionally, we propose a psuedo rule-of-thumb (ROT) Gram-Charlier expansion of unknown linked to empirical skewness kurtosis data. The intellectual investment needed implement these bandwidths practically zero. discuss behaviour as it links differences in ROT. further model selection criteria for choice when unknown. performance ROT are assessed variety Monte Carlo simulations well two illustrations, known data set annual snowfall Buffalo, New York, timely example stock market trading.