Expansions for the distribution of the maximum from distributions with a power tail when a trend is present (Q370889): Difference between revisions

This paper deals with expansions about the Fréchet distribution in inverse powers of the sample size, \(n+1\), for the sample maximum \(M_{n}\) when the \(j\)th observation is \(\mu\left( j/n\right) +e_{j}\), where \(\mu\left( \cdot\right) \) is a bounded trend function on \([0,1]\) with finite derivatives at \(0,1\). The residuals \(e_{0},\dots,e_{n}\) are independent with a common type Pareto\((\alpha,\beta)\) distribution \(F(x)\) whose upper tail can be expandend as \[ 1-F(x)=x^{-\alpha}\left( a_{1}+a_{2}x^{-\beta}+a_{3}x^{-2\beta}+\cdot \cdot\cdot\right) \] as \(x\rightarrow\infty\) with \(a_{1}>0\). Set \(M_{n}^{^{\prime}}=\left( a_{1}n\right) ^{-1/\alpha}\max_{0\leq j\leq n}Y_{jn}\), \(P_{n}(x)=P\left( M_{n}^{^{\prime}}\leq x\right) \) and \(G_{\alpha}(x)=\exp\left( -x^{-\alpha}\right) \) for \(x>0\), the Fréchet distribution. The main results give expansions for \(P_{n}(x)\) about \(G_{\alpha}(x)\) in inverse powers of \(n^{-1}\), \(n^{-1/\alpha}\), \(n^{-\beta/\alpha}\). The case when \(F(x)\) is Cauchy or Student's \(t\) is studied in more detail. Finally, the problem of inference from \textit{N }years{\ }of annual maxima in the presence of a trend is considered. The likelihood, when \(\alpha\) and \(\beta\) are rational, is given.