It is assumed that demand for information that subjectively appears to be relevant for forecasting improves forecasting quality. To study this hypothesis a number of forecasting experiments were conducted. Fifty managers from the housing business, from banking, and from a research institution were asked to forecast interest rates, using a Delphi process. They communicated via a computer system, and, to support their judgements, they had access to a data bank that was stored in the same system. Their communication with the system was automatically recorded. Part of the data collected in these experiments is used to study the existence of a relationship between information activities and forecasting results. A weak positive relationship is found if non-linear functions are tested, where information demand is corrected by those data retrievals that seem to have resulted from an inability to handle the information system. For further research a more general, albeit less informative, Boolean model is suggested.
KEY WORDS Forecasting quality Information demand Delphi groups
Computer dialogue Interest rate forecasting Practitioners often assume that decisions and forecasts may be improved by increasing the amount of information to be incorporated into these processes. On the contrary, some forecasts may suffer from information overload, as has been shown in a number of empirical studies. This should induce selective screening for 'relevant' information. This appears to be reasonable, as complete, piecewise processing of all incoming messages may consume too much time, effort and resources to put the relevant bits of the information to their use. Theoretical studies on the optimal level of information have been based primarily on Bayesian analysis to study the consequences of new information, and on an explicit model of information usage to evaluate its consequences. This approach is not discussed here, as it has been shown that people have difficulties in correctly estimating subjective probabilities (Hogarth and Makridakis (198 1) present a review ofjudgemental biases).
In the following section three approaches are mentioned that have been taken in the past with respect to empirical explanations for the use of information. Then, a laboratory experiment that served to generate the data to test one of these approaches is described. After that a number of hypothesis are presented and tested. The final section suggests further research.