A Split-Sample Approach for Estimating the Stability Index of a Stable Distribution (1705.09840v1)

Abstract: The class of stable distributions is used in practice to model data that exhibit heavy tails and/or skewness. The stability index $\alpha$ of a stable distribution is a measure of tail heaviness and is often of primary interest. Existing methods for estimating the index parameter include maximum likelihood and methods based on the sample quantiles. In this paper, a new approach for estimating the index parameter of a stable distribution is proposed. This new approach relies on the location-scale family representation of the class of stable distributions and involves repeatedly partitioning the single observed sample into two independent samples. An asymptotic likelihood method based on sample order statistics, previously used for estimating location and scale parameters in two independent samples, is adapted for estimating the stability index. The properties of the proposed method of estimation are explored and the resulting estimators are evaluated using a simulation study.