Bias of a Maximum Likelihood Estimator

112 Views Asked by At

Question:

For $\lambda \subset \mathbb{R}$, define the function $$ g_\lambda(y)=e^{-(y-\lambda)} _{(y\geq\lambda)} $$ Let $Y=(Y_1, Y_2, \ldots, Y_n)$ with $Y_1, Y_2, \ldots, Y_n$ iid random variables. Assume that each $Y_i$ has a continuous distribution pdf $g_\lambda$ where the parameter $\lambda$ is unknown.

(a) Find the maximum likelihood estimator $T(Y)$ for $\lambda$.

I have found out that the $\hat{\lambda_{MLE}}= max(Y_1,...,Y_n)$.

(b) Compute $P_\lambda(T(Y)>t)$ for $t\in\mathbb{R}$.

My attempt:

$P_\lambda(T(Y)>t)$

$=P_\lambda(max(Y_1,...,Y_n>t))$

$=1-P_\lambda(max(Y_1,...,Y_n)\leq t)$

$=1-P_\lambda(Y_1\leq t, ...,Y_n\leq t)$

since $Y_1,...,Y_n$ are iid,

$=1-[P_\lambda(Y_1\leq t)...P_\lambda(Y_n\leq t)]$

$=1-[(1-e^{-(t-\lambda)})...(1-e^{-(t-\lambda)})]$

$=1-(1-e^{-(t-\lambda)})^{n}$

(c) Based on the expression found in (b), compute the cdf $F_\lambda$ of $T(Y)$.

My attempt:

$F_\lambda(t)$

$=P_\lambda(T(Y)\leq t)$

$=1-P_\lambda(T(Y)>t)$

$=1-[1-(1-e^{-(t-\lambda)})^{n}]$

$=(1-e^{-(t-\lambda)})^{n}$

(d) Based on $F_\lambda$, compute the pdf of $T(Y)$.

$f_\lambda(t)$

$=\frac{\partial}{\partial t}F_\lambda(t)$

$=n(1-e^{-(t-\lambda)})^{n-1}e^{-(t-\lambda)}$

(e) Compute the bias of the estimator $T(Y)$. You may use known expressions for the mean and variance of an exponential distribution without proof.

My attempt:

bias($T(Y)$)

$=\mathbb{E}[T(Y)]-T(Y)$

$=\mathbb{E}[max(Y_1,...,Y_n)]-max(Y_1,...,Y_n)$

I'm having difficulty determining the distribution of max(Y_1, ..., Y_n) by examining the cumulative distribution function constructed in (c). I'm wondering if I made any mistakes in the previous parts of this question. Any help would be greatly appreciated!"