2
$\begingroup$

Let $Y_n$ denote the maximum (the last order statistic) of a random sample of size $n$ from a distribution of the continuous type that has c.d.f $F(x)$ and pdf $f(x) = F'(x)$. Find the limiting distribution of $Z_n=n[1-F(Y_n)]$

I dont have the solution to this but the answer my teacher gave us is that the limiting distribution of $Z_n$ should be exponential with mean $1$. If you define $Z_n=n[1-Y_n]$ then that limiting distribution is exponential with mean $1$ but I dont know how to work with $F(Y_n)$

$\endgroup$
2
$\begingroup$

I'm going to denote your "maximum" $Y_n$ as $Y_{(n)}$. Observe that \begin{align*} F_{Z_n}(z) &= P(Z_n \le z) \\ &= P([1 - F_{}(Y_{(n)})] \le z/n ) \\ &= P (F_{}(Y_{(n)}) \ge 1-z/n) \\ &= P(Y_{(n)} \ge F^{-1}(1-z/n)) \\ &= 1- P(Y_{(n)} \le F^{-1}(1-z/n)) \quad \text{(continuous dist.)} \\ \end{align*}

Suppose our random sample was $Y_1,\dots,Y_n$ from a distribution with cdf $F$. Then $$ P(Y_{(n)} \le c) = P((Y_1 \le c) \ \cap \cdots \cap (Y_n \le c)) \stackrel{ind.}{=} P(Y_1 \le c) \cdots P(Y_n \le c) \stackrel{i.d.}{=} [P(Y_1 \le c)]^n = [F(c)]^n $$

Returning and using this in the main problem, we have $$ F_{Z_n}(z) = 1 - [F \left( F^{-1}(1-z/n) \right)]^n = 1- [1-z/n]^n \xrightarrow{n \to \infty} 1- e^{- z}. $$ which we recognize as Exponential rate $=1$.

$\endgroup$
1
$\begingroup$

This is essentially the same idea given by @user365239, but stated a little differently.

Let $X_1,\dots,X_n$ be IID with DF $F$. Let $Y_n =\max_{j=1\dots n} X_j$.

Let $Z_j = n(1-F(X_j))$. Then $Z_1,\dots,Z_n$ are IID, and observe that

$$n(1-F(Y_n)) = \min_{j=1\dots n}Z_j.$$

Also, since $F(X_j)\sim U[0,1]$, we also have $1-F(X_j)\sim U[0,1]$, and so $Z_j\sim U[0,n]$.

From this we obtain:

$$ P( \min_{j=1...n} Z_n > x) = P( U[0,n]>x)^n = (1-x/n)^n \to e^{-x},$$

where the second equality holds when $n>x$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.