Econometric Theory and Methods

Answers to Starred Exercises

62

Solution to Exercise 10.2

10.2 Let the n --vector y be a vector of mutually independent realizations from the uniform distribution on the interval [1 , 2 ], usually denoted by U(1 , 2 ). 1 be the ML estimator of 1 given Thus, yt U(1 , 2 ) for t = 1, . . . , n. Let in (10.13), and suppose that the true values of the parameters are 1 = 0 and 1 is 2 = 1. Show that the CDF of 1 ) = 1 (1 )n. F ( ) Pr( 1 10 ), which in this case is just n 1 , is Use this result to show that n( asymptotically exponentially distributed with = 1. Note that the PDF of the exponential distribution was given in (10.03). (Hint: The limit as n of (1 + x/n)n, for arbitrary real x, is ex.) Show that, for arbitrary given 10 and 20 , with 20 > 10 , the asymp1 10 ) is characterized by the density (10.03) with totic distribution of n( 1 = (20 10 ) .

The probability that any single yt is greater than is 1 . Thus, for n independent observations, the probability that every one of them is greater 1 , the smallest yt , must be than is (1 )n. This is the probability that greater than . Therefore, 1 ) = 1 (1 )n, Pr( as we were asked to show. 1 and let z denote n. Since For the second part, let Z denote n 1 ), Pr(Z z ) = Pr( we have just shown that the CDF of Z is F (z ) = 1 (1 z/n)n. By the result alluded to in the hint, the limit of this as n is F (z ) lim F (z ) = 1 ez,
n

(S10.04)

(S10.05)

(S10.06)

which is the CDF of the exponential distribution with = 1. The PDF that corresponds to this CDF is then f (z ) = ez = ez , z

which is just a special case of (10.03) with = 1. Copyright c 2003, Russell Davidson and James G. MacKinnon

Econometric Theory and Methods

Answers to Starred Exercises

63

When 10 and 20 are arbitrary, the probability that any single yt is greater than is 1 ( 10 ), where (20 10 )1. Therefore, the equivalent of equation (S10.04) is 1 ) = 1 1 ( 10 ) , Pr( and the equivalent of equation (S10.05) is F (z ) = 1 (1 z/n)n. 1 10 ). By the same result that led to Here we have redened Z as n( (S10.06), we nd that F (z ) lim F (z ) = 1 ez,
n n

which is the CDF of the exponential distribution with arbitrary . The PDF that corresponds to this CDF is then f (z ) = ez = ez , z

which is the density (10.03), namely, the PDF of the exponential distribution. Since we have already dened as (20 10 )1, this is the result we were required to show.

Copyright c 2003, Russell Davidson and James G. MacKinnon