final2006sol - PROFESSOR HONG FALL 2007 TA: Jae-ho Yun...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: PROFESSOR HONG FALL 2007 TA: Jae-ho Yun Economics 319: Selected Suggested Solutions for 2006 Final 4. [10 pts] (a) The sample mean is de...ned as follows: Pn Xi X n = i=1 : n The expectation of sample mean can be derived as follows: Pn Pn E[Xi ] i=1 Xi E(X n ) = E = i=1 n n Pn i=1 = n = : Therefore, X n is unbiased for : 2 (b) For this, we have to ...gure out whether E(X n ) = 2 : We know that that V ar(X n ) = 2 E[(X n )2 ] = n , since Xi is iid. Let' use this result. s 2 E[(X n E[(X n ; n 2 )2 ] = E[X n )] = = = = = 2 E[X n ] 2 E[X n ] 2 E[X n ] 2 E[X n ] 2 2 Xn + 2 2 E[2 X n ] + E[ 2 ] 2 E[X n ] + 2 2 2 + 2 : 2 2 Therefore, we have E[X n ] biased for 2 : 5. [10 pts] 2 2 = 2 n , which implies E[X n ] = + 2 n 6= 2 : So, X n is 2 1 (a) (1 )100% con...dence interval when Xn p z n =2 = 5% is as follows: < Xn + p z n =2 ; < for every n 1, where z =2 is the upper-tail critical value of N (0; 1) at level =2 (z2:5% = 1:96). Note that we use standard normal distribuion since is known. (Otherwise, we should have used student-t distribution) Now, we have that n is 4, is 3 (Note that standard deviation should be used rather than variance), and X n = 5+10+6+7 = 7: Hence, 4 con...dence interval can be computed as follows: 7 3 p 4 1:96 < 4:06 < 3 <7+ p 4 < 9:94: 1:96; (b) The con...dence level is the probability that will lie in the interval between X n p z =2 and X n + p z =2 : The higher 1 , the more likely that will lie in the interval. n n In other words, the probability that will fall on the con...dence interval is (1 )100%: 6. [10 pts] We will use the t-statistics since 2 is unknown. Note that the degree of freedom is 1, since the sample size is 2. Compute the t-statistic as follows: t= X p 0: S= n If jtj < t1;0:005 ; then H0 is accepted, and HA is rejected. Otherwise, H0 is rejected, and HA is accepted. In order to implement this hypothesis test, we need the following: Sample mean: X= Sample variance: S 2 5 + 11 = 8: 2 2 1 = 9 + 9 = 18: = P2 i=1 (Xi X)2 = (X1 X)2 + (X2 X)2 2 Sample standard deviation: S= p S2 = p p 18 = 3 2: t1;0:005 = 63:656 from the student-t distribution table. 8 p p 4 3 2= 2 As a result, we have t = accept H0 ; and reject HA . 7. [15 pts] 2 It is known that nb21 this, we can know that E = 4=3: jtj is clearly less than t1;0:005 . Therefore, we 2 n (Note that this is wrongly stated in the question). From nb2 1 2 nb2 1 2 = n; and V ar = 2n: (a) Check whether b2 and b2 i unbiased for 2 : 1 h 22 are From the fact that E nb21 = n; we can know that E b2 = 2 : Therefore, b2 is 1 1 2 unbiased for : In order to check the unbiasedness of b2 ; we are going to use a well-known property 2 of the sample variance for iid random sample. Pn X)2 2 i=1 (Xi ; S = n 1 E(S 2 ) = 2 ; 2 4 : V ar(S 2 ) = n 1 Note that b2 = 2 n 1 2 S : n Therefore, we have that n n n n 2 V ar[b2 ] = 2 Since E b2 = 2 n 1 n E b2 2 = 1 E S2 = 1 2 n n 1 2 ; 1 n 2 2 V ar[S 2 ] = n 2 4 n 1 4 = 2 : n 1 n2 6= 2 (b) Note that MSE=bias2 +Variance. , b2 is biased for 2 . 3 result, M SE(b2 ) is 2n : 1 1 Next, consider b2 : We know that this is biased. The bias is n 2 since E b2 = 2 2 n 1 2 from part (a). The variance is nn21 2 4 (also from part (a)). Therefore, we have n 2 1 1 that M SE(b2 ) = n2 4 + nn21 2 4 = 2n 2 1 4 = n 4 n2 4 : 2 n Because M SE(b2 ) > M SE(b2 ); we can conclude that b2 is more e cient than b2 . 1 2 2 1 8. [20 pts] (a) We have to ...nd the condition under which E [b] = : " n X i=1 First, consider b2 :hThis is unbiased estimator, so bias term is zero. The variance can 1 i 4 nb2 1 be derived from V ar 2 = 2n in part (a). From this, we have V ar b2 = 2n : As a 1 4 E [b] = E = ci Xi = n X i=1 # P Hence, in order for b to be unbiased, we should have n ci = 1. i=1 (b) For the most e cient unbiased estimator, we have to solve the following minimization problem: " n # X M in V ar ci Xi ci i=1 n X i=1 n X i=1 ci E[Xi ] ci = ci : s:t: Using the indepence of Xi ; we have the following: " n # n X X V ar ci Xi = c2 V ar(Xi ) i i=1 i=1 n X i=1 ci = 1 = n X i=1 2 c2 i i2 : Let' construct the Lagrangian equation to solve the above maximization problem: s L= n X i=1 2 c2 2 i i + (1 n X i=1 ci ): 4 The ...rst order contition is as follows: 2 @L = 2 2 ci = 0; @ci i n X @L (ii) = 1 ci = 0: @ i=1 (i) for all i0 s; From (i), we can derive 2 2 ci = i2 for all i0 s: Let' sum up these equations for all i: s Then, we have the following: 2 2 n X i=1 ci = = n(n + 1)(2n + 1) 6 n X i=1 i2 (from the hint (there is a typo!)) If we use (ii), then we have the following: = 12 2 : n(n + 1)(2n + 1) Plug the above into (i), then we have: ci = = 1 2 1 2 i2 12 2 2 n(n + 1)(2n + 1) 6i2 = n(n + 1)(2n + 1) i2 2 Finally, we have the following best unbiased estimator: b = n X i=1 6i2 Xi : n(n + 1)(2n + 1) (c) We know that b is the best unbiased estimator, and that the sample mean, X n is also unbiased. Therefore, b is clearly better than X n in terms of relative e ciency. 2 Note that, for b ; the weight, ci is increasing as the variance, i2 decreases with i. That is to put more weight on less noisy observation. This improves e ciency. However, the sample mean put the same weights on all the observations although they have dierent levels of information. This fact makes b more e cient than X n : 5 ...
View Full Document

This note was uploaded on 12/05/2007 for the course ECON 3190 taught by Professor Hong during the Fall '07 term at Cornell University (Engineering School).

Ask a homework question - tutors are online