How to deal with unobservable variables in economics

The paper discusses different methods to deal with unobservable variables: Kalman-Filtering, principal components, factor analysis, LISREL, MIMIC, DYMIMIC, PLS with respect to parameter estimation and forecasting. We got very good results by an extension of Kalman-Filtering called AS (general stationary parameter model). LISREL proved to be superior to PLS in parameter estimation. Explicit introduction of the latent variables "mood" of the economic agents, the "political trend" and "social stability" improved the forecasting performance of an econometric model of the FRG.