Walter Salant and Forecasting Employment
In November, Samuelson began a correspondence with Bill Salant’s brother, Walter, who was working with Hans Neisser at the Office of Price Administration (OPA), concerning a paper by Richard Gilbert and Victor Perlo, also of the OPA, that had been presented to the Econometric Society in December 1941 and that had recently been published in Econometricati That paper argued that the traditional method of forecasting, which involved combining separate forecasts for each industrial sector, had not proved successful.
The problem was not lack of knowledge about the industries, for forecasts were typically produced by people who were very well informed about the industries concerned. Instead, they proposed to approach the problem the other way round, starting with “general economic factors” and only then constructing forecasts for individual industries. The theoretical framework Gilbert and Perlo used was an extended version of the multiplier/ Using least squares regression analysis, they had estimated ten relationships they believed were stable. This formed a set of simultaneous equations that could be used to generate forecasts that, they claimed, were much more successful than traditional methods.gSamuelson, who was by this time involved in post-defense forecasting at the NRPB, argued that their multipliers were much too large, for he
f. The crucial difference between their model and Samuelson’s multiplier-accelerator model was that, instead of finding a relationship between the level of investment and the change in income, they found relationships between changes in investment and changes in income. This difference completely changed the dynamic properties of the system.
g. The time path of income was explained by three sets of equations:
(ι) The multiplier: the increase in gross national product (GNP) was three times the increase in “weighted offsets to saving” (variables such as investment and net government spending).
(2) Equations for each of three types of investment: the increase in private investment in equipment was 0.1 times the increase in GNP; the increase in private investment in plant was 0.06 times the increase in GNP; and the increase in housing investment was 0.05 times the increase in GNP.
(3) Equations to determine federal receipts (a nonlinear relationship) and state and local government net receipts.
(4) The rise in imports was 0.03 times the rise in GNP.
Once they had calculated the change in GNP, they could use two further equations to calculate changes in industrial production and nonagricultural employment. Though these equations had been calculated using regression analysis, they did not report the results of any statistical tests, as would be routine in modern econometric work (Gilbert and Perlo 1942, p. 312).
was skeptical of the propensities to invest that were central to their calculations. “My own inclination,” he wrote, “is to distrust mechanical relationships between income and investment; at best I would use propensities to invest only in the very short run.”15 He suggested that their methods had worked during the previous two years simply because private investment had risen during “the defense period,” irrespective of whether or not there was a “reversible, repeatable relationship like that between consumption and income.” This makes very clear Samuelson’s strong belief that, despite the formal similarity between saving and investment functions in Keynesian models, their behavior was fundamentally different. He then went on to explain his view of the business cycle.
My own view is that the business cycle represents contagious swings of investment, set off by autonomous factors and changes in the level of income. How far these quasi-induced bursts will go depends upon the accumulation of investment opportunities due to technological change, changes in capital stock, outlook for extensive growth, etc.
He clearly thought that investment was a dynamic problem and, like Hansen, emphasized long-term structural factors.
However, despite his skepticism of the OPA forecasts, Samuelson felt that Perlo and Gilbert had avoided many of the pitfalls in the forecasting being undertaken in other government agencies, such as Agriculture and Commerce.Samuelson’s criticisms of the Gilbert-Perlo study were discussed within the OPA, and in December 1942, Salant sent Samuelson a memorandum from a colleague, Murray Geisler, explaining that a crucial equation had been left out of the Econometrica article.16 This stated that “weighted offsets” (the spending to which the multiplier was applied) was in fact a weighted average comprising 0.6 times current “offsets” and 0.4 times the previous year’s figure. This introduced a lag into the system that greatly reduced the multiplier—to a more normal level of 2¼. If no account were taken of the lag, the unreasonably high figures Samuelson had found would be correct. The problem Samuelson had found was due to a typing error.
However, Samuelson was not satisfied and looked at the problem again. What bothered him was that, if correct, this explanation undermined a basic idea—that lags could not affect the final value of the multiplier after all induced spending had taken place. This was an idea he had held since his early work on the business cycle, when he viewed the multiplier as determining the level of income and the accelerator, in which a lag is essential, determined fluctuations. He rejected the idea that Geisler had not allowed the multiplier process to work through, and he found another mistake: they had forgotten to include the lagged term, 0.4 times last year's offsets, in their calculations.17 Perlo wrote a memorandum to Salant, passed on to Samuelson, in which he said that Samuelson's view that lags could not affect the final equilibrium was correct, but “the point is that under the conditions which have prevailed in recent years, the system has never settled down.”18 In a corrigendum, Gilbert and Perlo explained that an equation had been left out of their paper, noting that, “[t]he total effect [of a change in public spending] over a whole business cycle is just as great as if there were no lag, but the timing is considerably modified.”
They continued to argue about technical details of the calculations, Samuelson not convinced by Perlo's explanation. Their discussion led to more substantial points concerning the methodology of forecasting and the statistical issues being confronted by forecasters at this time.
Perlo noted Samuelson's claim that, were it not for federal taxes, the system would be explosive: it was dangerous to assume that the system would remain the same whether federal expenditures were high or low. Wartime had led to such a large increase in government spending that applying the same system to pre- and postwar spending was hazardous. Samuelson responded that his main criticism still stood.19 They obtained high multipliers by assuming a mechanical relationship between saving and investment. The correlation between investment and income proved nothing, for it said nothing about causation and because the same data were being used to deduce saving and investment schedules. This would produce marginal propensities to save and invest that were essentially the same and “almost infinite” multipliers. The same problem applied to results obtained by other forecasters, Mordecai Ezekiel, Richard Bissell, and James Tobin.Here, Samuelson made two methodological points. If the OPA's methods produced good results for the period in which they were interested, they were justifiable. However, correct forecasts did not imply that the theory was correct: to prove that, it would be necessary to prove that the investment function was invariant, reversible, and that shifts over time were either negligible or predictable. This went against the trend of modern business cycle theory, which was “in the direction of placing the greatest emphasis upon the capricious, volatile, and shifting character of the net investment schedule.” In saying that, he was concurring with Keynes.
Perlo's comments, Samuelson concluded, bore on the basis upon which postwar forecasting was being undertaken, for if people believed that it was possible to find a reversible relationship holding in the twenties and thirties, one could not blame them for extrapolating it to the postwar world.h This was very important because “plenty of people are using what I consider to be a false part of the theory in their policy thinking concerning the post-war outlook and it will not do to tell them that the post-war world will assuredly be different from the pre-war world.” Much more was needed.
The following summer, Samuelson sent Walter Salant a copy of his chapter in the Harris volume, eliciting the following remark.
Your comment that a budget balanced at a higher level with nonprogressive taxes is employment-creating was certainly put in a tantalizing way. Can you let me know briefly what you have in mind? It sounds something like a point that my brother was making about a year ago, but I can't remember exactly the line of reasoning.20
Samuelson replied that it was the same point, and that he and Bill Salant had arrived at it independently. He added that it was a conclusion he had “been resisting subconsciously for some years.”21 He told Walter about his discussions with Bill, discussed earlier in this chapter, and that he was feeling guilty about being too busy to work on their joint paper. He said it would have been much better had he not interfered and left Bill's manuscript in its original form, and so he suggested that, if Bill agreed, Walter should now see Bill's original paper through to publication.
He also pointed to another way of presenting the result. The budget deficit measures the stimulus to private sector national income, so if there is a balanced-budget increase in public spending, private employment will not change; but there is nothing to cancel out the rise in public sector employment, resulting in an increase in total employment. Samuelson also explained why he thought the result important. Its wartime significance was that it showed the inflationary gap to be larger than the budget deficit. In peacetime, it could be important for discussions of the burden of taxation, potentially changing the “whole analysis of the incidence of taxes.”
Possibly in the postwar world it may become necessary to levy nonprogressive, dead-weight taxes. When we come to ask who shares the burden of these, it may turn out that nobody does; the government gets resources which would otherwise be idle, and their work is the only burden of the tax.22
h.
Samuelson used the term “reversible,” underlining it for emphasis. He appears to mean a stable structural relationship rather than a correlation that would disappear as soon as circumstances changed.The reference to the possible need to impose dead-weight (lump-sum) taxes shows the way that, even a year later, they were not thinking of this as a general proposition but, rather, as something relating to a specific tax regime.
The idea of the balanced-budget multiplier also occurred to Hansen, who after discussing it with Walter Salant and Abba Lerner, wrote to Samuelson about it in September 1943.23 By then, Samuelson had stopped commuting to Washington and they had not met for a while, and Hansen expressed the hope that they could meet on one of the days when he was at Harvard. Hesitant about the mathematics, Hansen wanted reassurance that the idea was right, and when Walter Salant had said that his brother had come up with the idea, he remembered that Samuelson had mentioned it to him several months previously, as well as stating it in the Harris volume. Samuelson and Bill Salant may not have finished their paper, but the main idea in the paper was circulating in Washington.[47]