Beschreibung

"In the paper an innovative method is applied to generate a forecast of regional employment. In common extrapolation problems the entries of a base matrix are used to estimate the elements of an unknown matrix. In the present problem of forecasting the base matrix is given for a time period in the past and the matrix to be estimated is related to a future period. In this paper a generalisation of this approach is used since the data about the past is given for a number of points in time and not only for a single period. <br> The Generalized Extrapolation Method (GEM) developed to take this information into account is related to well-known methods of panel analysis based on models with fixed effects. Additionally, it is similar to the techniques of entropy optimization. The objective function to compare the past and the future is a modified chi-squared statistic, which is approximately equal to the relative entropy function often used in gravity models and in input-output analysis. <br> The technique is demonstrated with respect to the example of employment in all 327 (western) German districts for a time span of two years. The method chosen uses any available information extensively. Therefore, the estimates are reliable, as is shown in an ex-post forecast." (Author's abstract, IAB-Doku) ((en))