Working Paper: CEPR ID: DP62
Authors: Colin Mayer; Matthias Mors
Abstract: Kalman filtering is used as a method of analysing the expectation formulation procedures employed by companies. The filter permits a number of alternative representations of managers' models of their firms activities to be compared. Applying the procedure to United Kingdom data on expectations, straightforward representations are found to provide reasonable descriptions of company predictions that frequently outperform contending hypotheses. However, there is evidence of changes in the forecasting rules employed and indications that multi-model versions of the filter provide better descriptions. A very straightforward forecasting rule that in some cases outperforms companies' own expectations is also revealed.
Keywords: kalman filtering; company expectations; learning; survey data
JEL Codes: 130; 210; 600
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| use of basic forecasting models (C53) | improved prediction accuracy (C52) |
| evolution of forecasting rules (C53) | accuracy of predictions (C52) |
| application of simple forecasting rule (C53) | accuracy of forecasts (C53) |
| combining multiple Kalman filters (C32) | accuracy of predictions (C52) |
| external economic conditions (E66) | evolution of forecasting practices (C53) |