The random average mode estimator (Q1073493)

Let \(X_ 1,X_ 2,...,X_ n\) be independent, identically distributed (i.i.d.) positive random variables with distribution F, mean \(\mu\) and finite variance \(\sigma^ 2\). The author considers the problem of estimation of mean \(\mu\) as the sample mean is not robust. Let \(U=\sum^{n}_{i=1}X_ ie_ i\), where \(e_ i,\quad 1\leq i\leq n\), are i.i.d. mean 1 exponential random variables independent of \(X_ i,\quad 1\leq i\leq n\). Define the random average mode (RAM) estimator as the mode of U/(n-1) for \(n>1\) and equal to \(X_ 1\) for \(n=1\). The author studies properties of RAM as an estimator of \(\mu\). He proves that \(n^{1/2}(RAM-\bar X)\to 0\) a.s. Another property of RAM is its logarithmic sensitivity to large values. An expression for the mean square error (MSE) of RAM is obtained. It is shown that RAM has an MSE which is smaller than that of sample mean for distributions with ''long right tails'' under the condition that \(E[\exp (tX_ 1)]<\infty\) for all t in a neighborhood of zero. Note that RAM estimator is robust against one large value but it is not robust against two large values. Some generalizations of RAM are suggested which might lead to estimators which are robust against more than one large value.