© 2000 John Petroff
|
|
© 2000 John Petroff |
While most economist would like to reduce their science to the equivalent of physics or chemistry (as evident in the first three approaches to forecasting just outlined), the subject of their inquiry - man - behaves in ways that are far from mathematical. In fact the important variables such as attitude about the future, preference of one product design over another, are often qualitative and emotional, and do not lend themselves to measurement. To assess the unquantifiable, the only alternative is to ask for opinions of those most directly involved in the process under study, i.e. experts. For corporate sales that can be the salespersons, product managers, designers, distributors, and especially customers. For national income, that would be economists, administrators directly involved in statistical gathering, politicians and business executives, union leaders and other individuals involved in decisions that affect business activity. Naturally, the "experts" have to be carefully screened. A larger number is better, but causes logistics problems in soliciting and receiving replies from all. Sometimes, just one expert is questioned.
The technique which was introduced in Chapter 5 Section H, consists in building a probability distribution of different possible values of forecasted variables. A questionnaire can ask for lowest, highest and most likely values, which are combined into a mean giving the middle value greater weight. When questionnaires are received from a large number of experts, a mean of the entire distribution of answers is derived. This is known as PERT, i.e. Program Evaluation and Review Technique. For instance, when a process involves a sequence of events (e.g. introduction of a new product in one selected store, then introduction in a region, then advertising campaign with promotional sale, and finally, introduction of the product nationwide), the questionnaire can ask for outcomes at each stage, and a conditional probability tree can be built, which is again used to calculate the mean of the distribution.
Another technique consists in informing survey participants about partial results obtained from other participants. This encourages convergence toward a consensus forecast. To avoid stubborn outliers, the median rather than mean is chosen as the resulting forecast. This is called the Delphi process.
Judgmental forecasts have the advantage of reflecting implicitly in experts' opinions all the most up to date and sensitive data available, and allowing a multitude of assumptions. Something that neither econometric models nor time-series analysis can do systematically. Yet they lack the apparent rigor of theoretically valid and explicitly stated assumptions.
See review questions .Q-15C4.1 through Q-15C4.7
See research assignments R-15C4.1 and R-15C4.2.
| Previous: 3-Time-series |
|
Next: Indicators |