SFB 303 Discussion Paper No. B - 104

Author: Krelle, Wilhelm
Title: Latent Variables in Econometric Models
Abstract: As to errors in the variables, not much can be done if one does not have suitable indicators for the true variables. In this case the method of latent variables could be applied to the variables subject to errors, but these indicators usually do not exist. Monte Carlo experiments may help to assess the degree of distortion induced by these errors so that one has at least a hint as to the importance of this error. But this does not lead to better results. The best way out, of course, is to make the statistics more reliable.The introduction of latent variables which have a definite meaning and which may be estimated by suitable indicators seems to be the next step in improving the performance of econometric forecasting models. This would also connect economics with the social and political domain and take into account the socio-psychological and organizational impacts on the economic sphere. A lot of work remains to be done here, but the results so far reached are promising.As to time dependent variables, the results obtained at Bonn University and published in Kirchen (1988) are so much in favor of time dependent parameters that one should use this method throughout. Not more information is needed for that purpose, but the information is transformed more efficiently into the econometric model. Moreover, wrong specifications of the behavioral equations may be corrected to a certain degree by this method. Of course, the best approach would be to introduce latent variables and use time dependent parameters simultaneously. This should be the next step.
Creation-Date: October 1988
Unfortunately this paper is not available online. Please contact us to order a hardcopy.

SFB 303 Homepage

28.09.1999, Webmaster