Forecasting Failure
Samuelson was one of the economists who was convinced that, in the absence of strong government action, the end of the war would be followed by a depression. However, even though the war ended abruptly before the measures he thought necessary were in place, the predicted slump never materialized.
The result was a controversy over the methods by which these forecasts had been made, in which the focus of attention was as much on the Keynesian theory upon which it was based as the statistical techniques employed.89 Samuelson's involvement in the production of the most recent forecasts had been behind the scenes, privately advising Goldsmith, but he had clearly associated himself with them though his articles in The New Republic. Moreover, he was even more strongly committed to the methods through which they were derived. He took the opportunity to publish a response, “Unemployment Forecasts: A Failure,” in the pilot issue of The American Economist, an academic journal established in 1946 with the aim of publishing widely accessible articles.90 Articles in the journal were unsigned, so he remained anonymous.Samuelson wrote that he would not have been surprised if the reason for failure was that they had failed to forecast investment correctly, because the difficulties in forecasting investment were well known. However, the failure arose because forecasters failed to get right the relationship between disposable income and nondurable consumption, the latter being $10 billion (roughly 5 percent of 1945 national income) higher than predicted. As Samuelson put it, this error “brings into question the stability of what in economic terminology is called the ‘consumption function.' ”91 It had been assumed that, because the reconversion of industry would take time and consumer durables would be in short supply, consumers would respond by postponing their consumption until more durable goods were being produced.
Instead, they had massively increased spending on nondurables.This, however, was a narrow, technical point. The failure of forecasting also raised questions about the whole basis on which forecasting was done. The complexity of the issues is shown by exchanges Samuelson had with his one-time co-author, Everett Hagen, who after leaving the NRPB had worked at the Federal Reserve before moving in February 1945 to the Office of War Mobilization and Reconversion.92 Drawing on work he had done with Samuelson at the NRPB, Hagen wrote two articles on forecasting, about which they entered into correspondence at the end of 1944.93
This correspondence seems to have begun with some handwritten comments on a draft of one of Hagen's articles.94 After discussing technicalities relating to the estimation of business and household saving, Hagen turned to Samuelson's criticisms of “Washington economists” in The New Republic. “I don't think you got around town enough before you wrote your articles for the New Republic,” he wrote. “You certainly didn't reflect the OPA correctly.”95 Samuelson's failure, he claimed, lay in not having distinguished between two distinct concepts: the single estimate of the most probable course of events, and “the contingencies against which policy measures should be adopted.” Hagen said that he, like several other Washington economists, had been concerned only with the former and, had time permitted, he might have drafted a reply to Samuelson's criticisms.
Samuelson interpreted this as implying that his New Republic articles might have done harm, and explained that he had written them to counter what he saw as excessive optimism about the postwar situation, though he realized that they were out of date even before they were published.96
I am sorry if my New Republic articles did any harm. As you know, the OPA has certainly been as pessimistic as I; but both Walter Salant and Jack Mosak were kind enough to write that in their opinions the articles did some good.
And I know for a fact, because of informal consulting which I was doing for them, that the WPB, the agency concerned with cut-backs, was living in a fool's paradise as far as their effects were concerned.97He went on to discuss Hagen's point about different types of forecast.
Certainly there are and have been important differences in your subjective probability distribution as to the future level of national income and mine. Your single most likely expectation differs from mine and so does your spread. I suspect the same could be said about the views of Colm, Smithies, Ezekiel and myself. When I asked Smithies whether I was wrong in my belief that there were not at hand anywhere in the federal government substantial plans for alleviating unemployment, his reply was ingenious: “do we ever in this country plan for anything?”98
Hagen reassured Samuelson that he had misunderstood what was “an offhand reaction” to the comment about Washington economists (of which Hagen was one), and that the New Republic articles had indeed done some good.99
Hagen's reflections were signed and appeared, much more prominently than Samuelson's confessions, in the American Economic Review.100 Hagen admitted that forecasters should have done better. They had been overconfident owing to their great success in predicting consumption both during the years before Pearl Harbor and during the war itself, when their models had forecast better than they could have expected. More important, during the war they had correctly forecast that conversion to war production could take place much more quickly than business people believed was possible, and this experience should have taught them that reconversion might also take place very rapidly, as did in fact happen. In contrast, Samuelson argued that forecasters had failed to predict things that were unpredictable, such as the expenditures of returning veterans, or how people would react when the durable goods that they would have been expected to purchase were not available.
Hagen raised a more general question about what methods should be used, claiming that there was a need for “far more, and far more systematic econometric work—of the sort that is being done at the Cowles Commission.”101
It would be a disastrous error to conclude from the experience of the reconversion forecasts that the use of “nation's budget” models should be discontinued.
It would be folly to assume that the use of economic barometers, or of the “qualitative-historical” method, or dependence on “informed judgement,” can take the place of quantitative estimates of the relationship between aggregate demand and its components, and aggregate supply.Samuelson supported the use of such formal forecasting methods, but his support was much more guarded.
It would be as foolish to lose all faith in the worth of such careful forecasts as to make the opposite error of blindly swallowing as Gospel- truth any pseudo-scientific estimates that come wrapped up in the impressive trappings of advanced statistical, mathematical, and economic techniques. The experience of the last year has shown that economists cannot entirely rely upon routine extrapolations of past statistical curves and regressions. So much the better. One good lesson will have been snatched from the postwar forecasting debacle if it is again made clear that forecasting involves more than an assistant's turning the crank of a calculating machine and grinding out the answer.102
The Cowles Commission methods that Hagen was advocating, and about which Samuelson was more skeptical, were those about which Trygve Haavelmo had spoken when he visited the MIT statistics seminar, and which by the 1960s were to become the dominant approach to empirical work in economics.r They involved specifying formal mathematical models and then using formal statistical inference both to attach numbers to coefficients in the models and to test whether they could explain the data.
Samuelson did not respond publicly to Hagen's call for economists to adopt the Cowles Commission's methods, but not long afterward he expressed his views privately to Seymour Harris, editor of the Review of Economics and Statistics, when he acted as a referee for an article Rutledge Vining, at the National Bureau of Economic Research (NBER), had written in reply to an article by Tjalling Koopmans, from the Cowles Commission.103 Titled
r.
See chapter 17 this volume. It was the approach to empirical work that Lawrence Klein, who moved to the Cowles Commission after finishing his PhD with Samuelson, was to make his own, his macroeconomic models inspiring many others. “Measurement without Theory,” Koopmans’s article was a response to the latest NBER study of the business cycle by Wesley Mitchell and Arthur Burns (later chair of the Council of Economic Advisers).104 This was a massive volume full of data, whose authors sought to provide a detailed statistical description of the business cycle. “Measurement without Theory” was a defense of the Cowles Commission’s methods that Hagen wanted to see used more widely. In his reply, defending Burns and Mitchell, Vining criticized those methods from the point of view of the NBER. His main argument was that the Cowles Commission took it for granted that the correct theory to use was Walrasian general equilibrium theory, but their methods were not of much use if one did not accept this and needed to discover the correct theory. If one had to discover what was the correct theory, the NBER’s methods were more useful. Samuelson’s verdict on Vining’s paper was, “Many of us will say 3 cheers with Vining when he defends empiricists against the perfectionistic- formalism of the Cowles Commission.”105 Where he parted company with Vining was that he did not agree that the alternative was to revert to thirtyyear-old NBER methodological procedures that had “grown decadent.”The exchange confirms Samuelson’s ambiguous attitude toward the methods that came almost to define what later generations would call “econometrics.” A few years earlier, he had written a favorable report on an econometric paper by Gerhard Tintner, revealing his familiarity with such work and his support for it.106 He thus favored the use of Cowles Commission methods, but believed that they should be used pragmatically, and he saw a significant role for less formal work.
This attitude, of showing great interest in such work but being skeptical of the necessity of pursuing empirical work this way, was consistent with the view expressed by Wilson in a review of Haavelmo’s Probability Approach to Econometrics, published the year before Koopmans’s review of Burns and Mitchell, attacking the NBER’s methods. Wilson’s main argument was that, although it was desirable “from an ideal point of view” to specify and take into account the behavior of error terms (the essence of Haavelmo’s method), it was a puzzle that scientists in other disciplines— astronomy, physics, engineering, biology, psychology, and medicine—had not found any need to do this.107 Why, he asked, does “the backward science of econometrics” need to be “more critical with respect to its probabilistic hypotheses than other sciences need be?” Though he was open to persuasion, he found no answer in Haavelmo’s work.s It would appear that Samuelsons. Wilson also made the same criticism of Haavelmo that he had made of Samuelson’s doctoral thesis: that he had things to say that were important for economists at large and writing in a less technical style would increase the accessibility of his work. shared Wilson's ambivalent view of Cowles Commission methods, and that despite sharp differences in expression, his disagreement with Hagen was one of emphasis as much as substance.