4.3 Sampling from a Multivariate Normal Distribution and Maximum Likelihood Estimation

4.3 Sampling from a Multivariate Normal Distribution and Maximum Likelihood Estimation

<p> 1</p><p>4.3 Sampling From A Multivariate Normal Distribution And Maximum Likelihood Estimation (a) The Multivariate Normal Likelihood</p><p>Let X1, X 2 ,, X n ~ N, be a random sample from a multivariate normal population. Then, the joint density function of</p><p>X1, X 2 ,, X n</p><p> n   t 1   1  x j    x j    f x ,, x  exp  1 n   p 1   2 2 j1 2  2    n  t 1  .  x j    x j   1  exp j1  np n 2  2  2  2     </p><p>The following results will be used to obtain the maximum likelihood estimate of  and  . </p><p>Result: Let A be a p  p symmetric matrix and x be a p 1 vector. Then, </p><p> t t t (a) x Ax  trx Ax  trAxx , where A  aij  and</p><p> p trA  aii . i1</p><p> p  (b) trA  i , where i are the eigenvalues of A. i1</p><p>Based on the result, we have 2</p><p> n n t 1 t 1 x j    x j   trx j    x j   j1 j1 n 1 t  tr x j  x j    j1   n  2A.12, p. 98 tr 1 x   x   t    j  j     j1  Then, n n t t t x j  x j    x j  x  x   x j  x  x   j1 j1 n n t t  x j  xx j  x  x  x   j1 j1 n t t  x j  xx j  x  nx  x   j1</p><p> n where  xi . Further, x  i1 n</p><p>  n  tr 1 x  x x  x t  n x   x   t     j  j        j1    n   tr 1 x  x x  x t   ntr 1 x   x   t   j  j          j1    n   tr 1 x  x x  x t   n x   t 1 x     j  j         j1 </p><p>Therefore, the likelihood function of X1, X 2 ,, X n can be simplified to 3</p><p> n  t 1   x j    x j   1 j 1 L ,  f x ,, x  exp      1 n  np n 2  2  2  2      .    n   tr  1  x  x x  x t   n x   t  1 x      j  j       1    j1    exp    np n   2  2  2  2      (b) Maximum Likelihood Estimation of  and  To obtain the maximum likelihood estimate, the following result will be used. Result: Given a p  p symmetric positive definite matrix B and a scalar b  0 , it follows that 1   tr 1B  1 exp    2b pb exp  bp b   b       2  B</p><p>  1 B for all positive definite  , with equality holding only for  2b . </p><p>Import Result (MLE of  and  )</p><p>Let X1, X 2 ,, X n ~ N, be a random sample from a multivariate normal population. Then, n t X j  X X j  X  ˆ X and j1 n 1S   ˆ   n n are the maximum likelihood estimators of  and  , respectively, where n t X j  X X j  X  S  j1 n 1 4</p><p> is a unbiased estimate of  . Their observed values, x and</p><p> n 1 x  x x  x t  n j  j  , are called the maximum likelihood j1 estimates of  and  . [proof:] ˆ maximizing the function </p><p>   n   tr  1  x  x x  x t   n x   t  1 x      j  j       1    j1   exp    np n   2  2  2  2      also minimizes the function </p><p> n  1  t  t 1 tr   x  x x  x   n x    x   .    j  j         j1 </p><p>Since  1 is positive definite, so that nx  t  1 x    0 . However, as</p><p>  x , the function </p><p>  n  tr 1 x  x x  x t   n x   t 1 x      j  j         j1    n   tr 1 x  x x  x t     j  j     j1  achieves its minimum. It remains to find ˆ maximizing 5</p><p>   n   tr  1  x  x x  x t     j  j    1    j1   L ˆ,  exp      np n   2  2  2  2     </p><p> n t b  n By the previous result with 2 and B   x j  xx j  x , the j1</p><p> n x  xx  xt maximum occurs at  j j . ˆ  j1 n</p><p>Note: Maximum likelihood estimators possess an invariance property. Let ˆ be the maximum likelihood estimator of  , and consider estimating the parameter h , which is a function of  , the maximum likelihood estimate of h  is given by hˆ. For examples, </p><p>1. The maximum likelihood estimator of  t  1 is ˆ t ˆ 1ˆ . </p><p>2. The maximum likelihood estimator of  ii is ˆ ii , where </p><p> n 2 X  X   ij i . ˆ  j1 ii n Note: </p><p>Let X1, X 2 ,, X n ~ N, be a random sample from a multivariate normal population. Then, n t X j  X X j  X  ˆ  X and j1 ˆ  n are sufficient statistics. 6</p><p>4.4 The Sampling Distribution of X and S Definition of the Wishart Distribution: </p><p>Let Z1,Z2 ,,Zn ~ N p 0, be independently distributed. Then,</p><p> n t the random matrix M   Z j Z j is distributed as a Wishart j1 distribution with n d.f., Wn, . The density of a Wishart distribution with n d.f. is (n p1)  trm 1  m 2 exp  2   . f m |   p pn p( p1) n 2 4 2 1  2    n 1 i i1 2 </p><p>Properties of the Wishart Distribution: 1. If M ~ W and M ~ W , then 1 n1 , 2 n2 ,</p><p>M  M ~ W . 1 2 n1 n2 ,</p><p> t 2. If M ~ W , then CM1C ~ W t . 1 n1 , n1 ,CC</p><p>Import Result:    1. X ~ N p  ,  .  n  n t 2. n 1S  X j  X X j  X  ~ Wn1, . j 1</p><p>3. X and n 1S are independent. </p><p>4.5 Large-Sample Behavior of X and S Import Result: 7</p><p>Let X1, X 2 ,, X n be a random sample from a population with mean  and finite (nonsingular) covariance  . Then, </p><p> nX    N p 0, and </p><p> t 1 2 nX   S X    p for n  p large. </p>

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    7 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us