Journal article

Publication year: 1995

Pages: 477-494

Journal: Control and Cybernetics

Volume number: 24

Issue number: 4

URL: http://control.ibspan.waw.pl:3000/contents/show/97?year=1995

Publisher: Sciendo

Abstract: 

In econometric models, the variables of economic theory are typically replaced by a much smaller set of aggregated variables, while the structure of the model remains unchanged. The manner in which the variables are partitioned into groups to be aggregated is usually based on intuition or convenience. In this paper we propose to carry this out in an optimal manner, the criterion being minimization of mean-square forecast error. This leads to an integer programming problem of high computational complexity. The optimization heuristic Threshold Accepting is implemented for the optimal partition and aggregation of a long monthly series of Swedish internal and external price indices. To correct for heteroskedasticity resulting from inflation, the sample variance matrix is assumed proportional to a diagonal matrix whose diagonal elements are the sums of squares of the external prices. This is compared with results obtained by using
a scalar variance matrix and replacing Euclidean by Mahalanobis distance. The algorithm and the resulting groupings are presented.

Harvard Citation style: Chipman, J. and Winker, P. (1995) Optimal Industrial Classification by Threshold Accepting, Control and Cybernetics, 24(4), pp. 477-494. http://control.ibspan.waw.pl:3000/contents/show/97?year=1995

APA Citation style: Chipman, J., & Winker, P. (1995). Optimal Industrial Classification by Threshold Accepting. Control and Cybernetics. 24(4), 477-494. http://control.ibspan.waw.pl:3000/contents/show/97?year=1995