Room C816, Building of the Faculty of Science Graduate school of Science
Prof. Shintaro HASHIMOTO(Hiroshima University)
Robust estimation for location and scale parameters using heavy-tailed distribution
We often use the sample mean and standard derivation to estimate the location and scale parameters. However, inference can be contaminated by the presence of outliers if we use light-tailed distributions such as the normal distribution. To overcome this, we often use the t-distribution. By using the t-distribution, we can remove the effect of outliers for location parameter, however, we can not remove completely the effect of outliers for scale parameter. Recently, Desgagne (2015) proposed the log-Pareto-tailed symmetric distribution and show that this distribution work well for robust estimation of scale parameter. In this talk, we review his paper and show some problems on using the log-Pareto-tailed symmetric distribution.