S IN CE the end of the war, most of the widely publicized forecasts of the level and direction of change in national income (or gross national product) have been derived from models built around the consumption function. As these estimates resulted in a series of incorrect predictions, the literature has been filled with reconsiderations designed to show that the mistakes resulted from an incomplete statistical treatment of models which were in themselves essentially correct. This paper does not examine the validity of Keynesian-type models for longer analysis, but it does suggest that certain variables which have been suppressed in constructing simplified models cannot be ignored in short economic forecasting. 2 A common cause of difficulty in econometric forecasts has been the assumption that a consumption function, derived from data extending over a period of time, can be applied to successive short cases. Whether or not such a function may be correct for analysis, it will probably yield erroneous results when used to predict changes in income and consumption over two, three, or four quarterly periods in the future -because the consumption-income relation appears to shift from one short period to another. Since the needs in both business and governmental planning are primarily for short forecasts, quarter by quarter, an approximation of the varying, short schedule of consumption-income relationships would seem to be needed. The only concession by most writers and forecasters has been to make provision for a regular upward shift in the consumption function in small annual increments. Others, notably Modigliani, have gone further to distinguish conceptually between a varying short relation and the long run function customarily used.' Having made this logical distinction, even Modigliani has, however, ruled aside most of the short influences upon the relation, in the interest of statistical manageability. He has identified short movements with a cyclical pattern, measured by the percentage variation of disposable income (real, per capita,) from its previous highest level. Cyclical uniformity is thus implied by his assumption that the savings ratio will always be the same whenever disposable income is a given percentage below previous highest income. Differences attributable to the direction of cyclical income change are ignored. Nor is there allowance for the possibility that some significant short changes might not be uniquely related to a repetitive cyclical pattern. While the aggregative techniques used in forecasting have, implicitly or explicitly, disregarded short influences on the consumption-income relation, consumer budget studies have been focusing increased attention upon such influences. Changes in income distribution, by income-size groups and among occupational groups; the associated effect of changes in tax rates; recent changes in price levels and in relative prices; the distribution and aggregate volume of liquid assets; and the state of consumer expectations all these, and other influences, have been highlighted by the ' While this paper has gained considerably from the personal comments of many colleagues at the Federal Reserve Bank of New York, and those of other economists both within and outside the Federal Reserve System, the writer alone is responsible for the views expressed. The writer is especially grateful to Professor John H. Williams and Dr. P. A. Baran for major criticisms of earlier drafts. 2 It must be recognized from the outset that most of the emphasis upon the consumption function in economic literature, apart from actual forecasting, has been directed toward problems not discussed in this paper. 3 Franco Modigliani, Fluctuations in the Savings Ratio: A Problem in Economic Forecasting, a paper to be published by the National Bureau of Economic Research in a volume of Studies on Research in Income and Wealth. Both Modigliani and James S. Duesenberry have constructed equations which yield reasonably accurate forecasts of the current rate of savings.