Second order expansions for the moments of minimum point of an unbalanced two-sided normal random walk (Q1300764): Difference between revisions

Let \(\{ Z_i\}\) be independent and identically distributed normal variables with variance 1 and mean \(\theta_0 < 0\) for \(i < 0 \) and \(\theta > 0\) for \(i > 0\). The two-sided normal random walk is defined as \[ W_n= \begin{cases} \sum_{i=1}^n Z_i \;& \text{ for} \;n\geq 0, \cr 0 & \text{ for} \;n=0, \cr -\sum_{i=n}^{-1} Z_i \;& \text{ for} \;n\leq 0. \end{cases} \] Denote by \(W_{\nu_0}=\min_{-\infty\leq n\leq \infty}W_n.\) The second order expansion of the first two moments of \(\nu_0\) when the drift parameters \(\theta\) and \(\theta_0\) approach zero at the same order is proved. The results can be used to study the bias and variance of the maximum likelihood estimator for the change point. It is shown that the first moment of \(\nu _0\) is different from its continuous Brownian motion analog although the second moments are the same in the symmetric case.