Formalizing judgemental adjustment of model-based forecasts

In business and in macroeconomics it is common practice to use econo- metric models to generate forecasts. These models can take any degree of sophistication. Sometimes it is felt by an expert that the model-based fore- cast needs adjustment. This paper makes a plea for a formal approach to such an adjustment, more precise, for the creation of detailed logbooks which con- tain information on why and how model-based forecasts have been adjusted. The reasons for doing so are that such logbooks allow for (i) the preservation of expert knowledge, (ii) for the possible future modi¯cation of econometric models in case adjustment is almost always needed, and (iii) for the evaluation of adjusted forecasts. In this paper I put forward an explicit mathematical expression for a judgementally adjusted model-based forecast. The key pa- rameters in the expression should enter the logbook. In a limited simulation experiment I illustrate an additional use of this expression, that is, looking with hindsight if adjustment would have led to better results. The results of the simulation suggest that always adjusting forecasts leads to very poor results. Also, it is documented that small adjustments are better that large adjustments, even in case large adjustments are felt necessary.