Navigation:  Menu Commands > Tools Menu >

 

Information Matrix

 

Print this Topic Previous pageReturn to chapter overviewNext page

Calculate Fisher's information matrix of the log-likehood function at initial or posterior mode parameter values. This can be performed either in the frequency domain  or in the time domain. In the former case, there computation is based on Whittle's (1953) estimator and, hence, doesn't require any data, while the latter does. The information matrix can be used when investigating identification issues; see Andrle (2009) or Iskrev (2007).

YADA attempts to calculate the correlations from the inverse of Fisher's information matrix. All correlations that in absolute terms are greater than or equal to 0.7 are written to the text file. Moreover, YADA logs the rank of the information matrix.

In addition, YADA calculates a singular value decomposition of the information matrix to determine which linear combinations of the parameters (eigenvector) that are connected with the largest to the smallest eigenvalues.

Furthermore, YADA performs a sequential QR factorization with column pivoting of the information matrix. This is used to obtain an ordering of the parameters from the "most" identified to the "least" identified.

 

Additional Information

A more detailed description about the information matrix from a time domain perspective can also be found in Section 11.11 of the YADA Manual, while the information matrix from a frequency domain perspective is covered in Section 13.8.
The rank revealing algorithm is discussed in Section 11.12.

 

 


Page url: http://www.texlips.net/yada/help/index.html?information_matrix.htm