Analyse et prévision: Leurs rôles respectifs dans la Maîtrise de nos Destins

Abstract

Resume Dealing with forecasting as an autonomous scientific discipline may lead to unwarranted expectations. Deep analysis of the phenomena commonly is the true source of progress in the art of forecasting. Indeed, in each subject matter field, good theoretical foundations improve forecasting ability and they can only be the result of intensive analytical work, using alternatively both the deductive and the inductive method. Moreover, in the present world, the forecasting activity has a need for stricter ethical principles. Pure observation of past regularities is sometimes sufficient for good forecasting, and this thanks to now wellknown techniques of time series analysis. But most often the difficulty does not lie in the forecasting phase, but rather in the identification of the structure of the complex multidimensional stochastic process ruling the phenomenon. This is illustrated by reference to macroeconomics. Simple observation of regularities in business activity was given high priority during the interwar period. But in the 1940's the approach was vigorously challenged as “measurement without theory” and a new methodology emerged that required identification and estimation of a structural system in which the main equations explaining the determination of the variables of interest where explicitly specified. The methodology was widely accepted and has been almost exclusively used since then. This situation prevails in macroeconomics even though the methodology makes the results somewhat dependent on prior theoretical views of model builders, and even though the accuracy achieved remains lower than one would wish. Recent questioning of the methodology and the advocacy, by C. Sims in particular, of direct fitting and extrapolation of multidimensional processes on a few main macroeconomics series does not seem likely to change the situation. When the system ruling the phenomena was effectively built by men and is therefore known, methods of system analysis are appropriate. But usually knowledge of the structure requires detailed research on many of its components, once these have been identified. As an example illustrating this point the problem of forecasting French unemployment is considered. It is shown that, in the situation prevailing in 1986, a crucial step is to determine how and with which lags the demand for labor will react to a major shift in the structure of prices and remuneration rates. But at present knowledge of the determinants of this demand for labor, and of the productive capacity building that goes with it, is still too uncertain to permit even an approximate assessment. Only detailed theoretical and econometric research can alleviate the difficulty. For the progress of this essential grass-root scientific work, the demand for forecasts, particularly by the media, often plays a detrimental role. Pressed to answer, some scientist may forget their standards of rigor, and this may even bring to them some definite rewards. This is why professional ethics of forecasting should be made explicit. Their main principles should be fairly obvious: forecasts should not be intentionally misleading, forecasts should be completently produced, using the appropriate stock of knowledge and one of the best available techniques; moreover systematic ex post assessment of the quality of past forecasts should be made, and made public. Again, this is illustrated by the practice that tends to emerge for ethics of macroeconomic forecasting.