Scribd
Upload
38 views2 pages

Solution To Exercise 10.19

The loglikelihood function (th) depends on one scalar parameter th. Consider the distribution of the LM statistic under the drifting DGP. This DGP drifts toward the fixed DGP with th = 0, which we think of as representing the null hypothesis.

Uploaded by

Leonard Gonzalo Saavedra Astopilco
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
38 views2 pages

Solution To Exercise 10.19

The loglikelihood function (th) depends on one scalar parameter th. Consider the distribution of the LM statistic under the drifting DGP. This DGP drifts toward the fixed DGP with th = 0, which we think of as representing the null hypothesis.

Uploaded by

Leonard Gonzalo Saavedra Astopilco
Copyright
© Attribution Non-Commercial (BY-NC)
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Econometric Theory and Methods

Answers to Starred Exercises

76

Solution to Exercise 10.19


10.19 Let the loglikelihood function () depend on one scalar parameter . For this special case, consider the distribution of the LM statistic (10.69) under the drifting DGP characterized by the parameter = n1/2 for a xed . This DGP drifts toward the xed DGP with = 0, which we think of as representing the null hypothesis. Show rst that n1I (n1/2 ) I(0) as n . Here the asymptotic information matrix I() is just a scalar, since there is only one parameter. Next, show that n1/2 times the gradient, evaluated at = 0, which we may write as n1/2 g (0), is asymptotically normally distributed with mean I(0) and variance I(0). Finally, show that the LM statistic is asymptotically distributed as 2 (1) with a nite noncentrality parameter, and give the value of that noncentrality parameter.

We see by a Taylor expansion of the denition (10.31) that n


1

I (n

1 /2

) = n
1 = n

It (n1/2 )
t=1 n t=1

1 It (0) + n

n t=1

) n1/2 , It (n1/2

where It () is the derivative of the contribution It (), which is just a scalar . Taking the limit as n function of a scalar argument, and 0 of the above relation gives the rst result, since lim n1 It (0) = I(0) by ) is bounded above as n . denition, and n1 It (n1/2 Another Taylor expansion, of n1/2 g (n1/2 ) this time, gives ) , n1/2 g (n1/2 ) = n1/2 g (0) + n1 H (n1/2 (S10.44)

where H () is the Hessian, which is also just a scalar in this special case. For each value of n, the true value of is n1/2 , and so the expectation of n1/2 g (n1/2 ) is zero for all n. Similarly, the variance of n1/2 g (n1/2 ) is n1 I (n1/2 ). Thus the plim of n1/2 g (n1/2 ) has expectation 0 and variance the limit of n1 I (n1/2 ), which is I(0) by the rst part of the exercise. A central limit theorem can be used to show that the plim is also asymptotically normal. Thus we see that the limit of the left-hand side of (S10.44) is a variable distributed as N 0, I(0) The argument used in rst part of the exercise shows that the limit of n1 H (n1/2 ) as n is H(0), which is equal to I(0) by the information matrix equality. Thus the limit of the second term on the right-hand side of (S10.44) is the deterministic quantity I(0). It follows that the plim Copyright c 2003, Russell Davidson and James G. MacKinnon

Econometric Theory and Methods

Answers to Starred Exercises

77

of n1/2 g (0), the rst term on the right-hand side of (S10.44), is distributed as N I(0), I(0) , as we were asked to show. The LM statistic (10.69) for the null hypothesis = 0 can be written as g 2 (0) LM = = I (0) n1/2 g (0) n1I (0)
1 /2 2

The random variable in the numerator of the rightmost expression here has a plim that is distributed as N I(0), I(0) . When this variable is divided by the square root of its variance, the result is asymptotically distributed as N I1/2 (0), 1 . The LM statistic, which is the square of this result, is therefore asymptotically distributed as noncentral 2 (1) with noncentrality parameter 2 I(0); see Section 4.7.

Copyright c 2003, Russell Davidson and James G. MacKinnon

You might also like