No label defined (Q4531258)

This paper solves the of goodness-of-fit test problem for the Cauchy distribution. First, the Anderson--Darling test statistics \(A^2\) is recalled as a general tool for goodness-of-fit tests [\textit{R.B. D'Agostino} and \textit{M.A. Stephens} (eds.), Goodness-of-fit techniques. (1986; Zbl 0597.62030)]. The empirical distribution function is compared with the hypothetical distribution function where the values of the unknown parameters are substituted by their estimates. In the case of the Cauchy distribution, the most common estimators of the parameters \(\mu\) of location and \(\sigma \) of the shape are constructed as linear functions of order statistics. Theorem 1 of the paper establishes the asymptotic representation of the estimator of \(\sigma \). However, it is known that when these standard estimators are used, the distribution of \(A^2\) depends on the parameter \(\mu \). A simulation study in the first part of the paper shows that the test based on these estimates does not yield good results for moderate sample sizes. Therefore, the author estimates the parameter \(\mu \) by the sample median and proposes a new test statistic based on \(k\) sample quantiles instead on full order statistics. Theorem 2 proves that when the hypothesis of Cauchy distribution holds, the test statistics converge weakly to the chi-square distribution with \(k\) degrees of freedom, provided the sample size \(n \to \infty \). A simulation study demonstrates that the test begins to be powerfull already for rather low sample sizes \(n \sim 30\).