1Economists conceive most economic and business time series to consist of seasonal, trend, cyclical, and so-called irregular components such as the effects of strikes, unexpected political or international developments, or unusual weather conditions. To determine the cyclical behavior of an economic time series, they usually adjust data for seasonality, the secular trend, and irregular movements. In the mid-1950s, the analysis of a ten-year monthly series required about 2,500 computations, 5,000 if checks for correctness were counted. This, explained observers, took a clerk about a week (Shiskin, 1957, 244). The large time costs of time series decomposition caused significant delays in the availability of adjusted economic data. Figures were therefore often considered “too late” to be of value and service (Robey, 1961). This posed a particularly grave problem in business forecasting. Economists and business representatives deplored that they were forced to make forecasts without knowing whether their country or business had been getting better or worse between the moment “captured” in the data and the present.
2In the late 1940s, many economists, particularly in the United States, perceived this with growing concern. In the context of the Cold War, it seemed that any unanticipated crisis could—in the systemic competition between socialism and capitalism—tip the scales in favor of the former. Given the sense of this heightened pressure, the introduction of the electronic computer in the early 1950s was welcomed by many with great relief. For the first time, it became possible to process economic data within minutes, prompting hopes that business cycle researchers had finally found the means to foresee the approach of crises in time and, in so doing, help prevent them.
3However, as this article argues, electronic computers brought about new problems of time. First, they required and indeed created time-consuming tasks which also involved tedious organizational processes. Second, until the development of the microprocessor, computer time was scarce. Both factors pushed business cycle researchers to aim for ever greater standardization in the use of electronic computers. While this did indeed allow for an unprecedented acceleration in data processing, it ultimately called into question the very project of forecasting itself.
4Focusing on the case of the National Bureau of Economic Research (NBER), this article investigates these problems and their implications for the project of business cycle research and forecasting. The researchers of the NBER, particularly Wesley C. Mitchell (1874-1948) and Arthur Burns (1904-1987), are well-known for having developed the leading indicator technique, which up to today constitutes one of the most popular forecasting tools used all over the world. After a brief introduction of the leading indicator technique developed at the NBER in the 1930s (Section 1), I explore how researchers of the Bureau used electronic computers in the 1950s and 1960s to process monthly time series for forecasting purposes. I investigate the time-consuming routines that the use of electronic computers involved (Section 2) and argue that these time costs pushed standardization efforts at the NBER. As I show in Section 3, standardization introduced a new set of problems, which eventually prompted the researchers to cease their efforts to forecast economic fluctuations.
5Historians of science have argued that the ideal of rationality based on formal, sequential rules, immune to intuition and judgment flourished during the Cold War, shaping the American human sciences from the 1950s through the early 1970s (Erickson et al., 2015). Several historians have claimed that the advent of the electronic computer boosted and embossed this ideal, prompting the idea of rationality as rule-bound and algorithm-centered in the first place (see, e.g., Erickson et al., 2015; Kott 2018, 139-140). This article complicates this claim. As I argue, the standardization processes which computers and their time costs pushed proved that mechanical rule following was worthless in business forecasting if not supplemented with practices of judgment or interpretation which the researchers had—sometimes unconsciously—employed earlier (Section 4). The computer thus put an end to the researchers’ hope that more data and better technology would, one day, allow for a purely “mechanical” forecasting approach and, in so doing, heralded the end of the Bureau’s forecasting activities.
6The use of large-scale electronic computers in economics is usually described as the beginning of a new era in which economists could solve long-standing problems in almost no time (on this point, see Backhouse and Cherrier, 2017). While it certainly did transform economics, several scholars have qualified the role of computers in these changes. With a more general view on knowledge production, Jon Agar (2006) argued that the use of large-scale electronic computers has only been attempted in settings where material and theoretical computational practices and technologies already existed. As Agar explained, programs of human operator action were transformed to programs of computer instruction with little change in procedure. With this, Agar argued against instrument-determinists, who claim that the development of electronic computers, by allegedly making certain scientific activities possible for the first time, presented a “phase change” in the sciences (Agar, 2006; Backhouse and Cherrier, 2017). By exploring how business cycle researchers at the NBER used electronic computers to replace manual calculation practices, this case study strengthens Agar’s claim. But it does more than that. In fact, as my article claims, the use of large-scale electronic computers, which the researchers welcomed so enthusiastically at first, quickly proved the unattainability of the goals for which the procedures had been designed. As I argue in the conclusion, this article is therefore another reminder to revisit the complexities and ambiguities of the post-war period and to examine not only the planning and computerization euphoria of the 1950s and 1960s, but also the early disappointments.
7The early twentieth century saw the emergence of a veritable forecasting industry in the United States, which found imitators in Europe and on other continents after World War I (see Morgan, 1996; Tooze, 2001; European Commission and Ladiray, 2003; Friedman, 2014; Lenel et al., 2020). A series of economic crises had prompted hopes among economists, policymakers, and businesspeople that the recent expansion of economic statistics might enable them to identify certain patterns in past data that would allow them to forecast crises. This, they believed, might not only help them lay the basis to mitigate future crises but also silence calls for government intervention. One of the best-known research institutes established along these lines was the Harvard Committee on Economic Research founded in Boston in 1919 (Morgan, 1996, 56-63; Armatte, 2003, 63-65; Fayolle, 2003, 12-17; Friedman, 2014, 128-165; Friedman, 2009; Lenel, 2018; Lenel, 2021a). Their Index of General Business Conditions, which was based on the assumption of the existence of timeless laws governing the business cycle, was widely celebrated as the first “scientific” forecasting tool and was adopted on three continents.
- 1 Wesley C. Mitchell. Letter to Henry S. Dennison, February 5, 1924, Box 8, Wesley C. Mitchell Papers (...)
8And yet, only a few years after the publication of their first forecast, the Harvard Committee had to abandon its forecasting scheme. After severe deviations between their forecasts and the actual future realized, the members of the Harvard group resorted to informal expectation exchange to incorporate the plans and the mood of selected economic and political decision-makers into their forecasts (Lenel, 2021a). One of the most astute observers of this change in method was the American economist Wesley C. Mitchell, who heavily criticized his Harvard colleagues’ new approach. While in 1919, Mitchell had praised their index as “ultra-scientific,” (Mitchell, 1919a, 872) he believed they had given up their scientific status by relying on inside information rather than statistics (see Lenel, 2021a, 154).1 As Mitchell saw it, only statistical methods could claim to yield “objective” and thus “scientific” results.
9These beliefs also shaped the direction of the National Bureau of Economic Research, which Mitchell was involved in founding in 1919 and led as director of research for more than 25 years (Lenel, 2021b). Both the history of the Bureau’s formation and its design reflect the idea that economists could help unravel “economic facts” of greatest social significance by employing “scientific,” that is empirical, methods (Mitchell, 1939, 9-10). Since the Bureau’s formation, the members of the Bureau tirelessly tried to extend the—at the time of its formation still scant and inadequate—statistical coverage in the United States, devising methods for compiling data on national income, capital formation, consumer spending, capital consumption, consumer installment credit, and money flows that were later continued by government agencies. By enlarging and organizing the collection of economic data in the United States and cooperating with federal agencies to help them compile and regularly publish data, the staff hoped to serve the general welfare as well as to transform their own discipline. As Mitchell argued in 1919, the social sciences had until now been “immature, speculative, filled with controversies,” with their exponents developing, over and over again, “new ‘viewpoints’ ... instead of carrying further the analysis of their predecessors.” (Mitchell, 1919b, 230) The collection and analysis of social statistics was to offer a way forward. As the members of the NBER believed, statistics would allow them to produce “objective” knowledge that was “amenable to mathematical formulation” and “capable of forecasting group phenomena.” (Mitchell, 1919b, 231) The use of statistics, or so they hoped, could thus help advance a discipline that was “still childish” into a “robust manhood” equaling the natural sciences (Mitchell, 1919b, 230-231).
10The business cycle research conducted at the NBER, which soon became the researchers’ main focus, bore the imprint of this objective. Under the leadership of Wesley Mitchell, a group of 10-15 economists collected and analyzed an ever-growing body of time series from 1922 onward to unravel certain patterns from which they could induce a timeless “model of business cycles” (Burns, 1951, x; Gordon, 1952, 107). The work was driven by the idea that phenomena—or the data bearing witness to them—could speak for themselves. Thus, the researchers’ objective was to start not with any preconceived ideas or theories, but with “carefully screened observations” and on this basis “develop a model of business cycles” (Burns, 1951, x). This analysis, again, was to follow a set course that would warrant the objectivity of the results. Along these lines, Mitchell’s collaborator Arthur Burns described their time-series analysis as an “apparatus” that quasi-mechanically “yielded” certain results (Burns, 1949, 41).
11The apparatus was a complex one. Over one thousand time series covering a variety of economic processes in four countries were compared to so-called “reference cycles” intended to reflect “the cycles in general business activity in the country to which the series refers.” (Burns and Mitchell, 1946, 23) To this end, researchers compiled a table of so-called reference dates indicating the months and years business cycles had reached troughs and peaks. After eliminating the seasonal variations of the individual series, they broke each series into segments marked off by these reference dates. To reduce “the influence of erratic movements on the patterns,” the researchers converted the original data to percentages of their average value during a specific cycle, and determined their rise and fall during the different stages in terms of these percentages (Mitchell, 1951, 14). Based on these percentages, Mitchell and his colleagues computed an average of the cyclical fluctuation for each stage. By averaging the measures of numerous individual “cycles” of a series, the researchers hoped to reveal what behavior was “characteristic” of different economic processes (Mitchell, 1951, 14). In a second step, the researchers assessed the relation of the cyclical fluctuations of individual series to those of the reference cycles. Here, they compared the reference dates to the dates of the cyclical turns in each individual series and on this basis determined the number of months by which the turning dates of a specific series had preceded or followed the reference troughs and peaks. In this, the researchers took up one of the most important ideas of the Harvard Committee. Like the Harvard group, the members of the NBER believed in the existence of stable relationships between certain economic processes. By significantly enlarging the range and time span of the time series processed, however, they hoped for results that were more representative and thus more reliable as guides to the future.
12These measures were made for the different cycles covered by a series, and then again averaged for each set of cycles. By this means, the researchers wanted to reduce the influence of random factors. As they explained, “When averages are struck for all the cycles covered by a series, features peculiar to single cycles tend to fade away, while features common to all or most of the cycles tend to stand out prominently.” (Burns and Mitchell, 1946, 33) Although aware that each specific “business cycle is ... an individual differing in countless ways from every other,” the researchers nevertheless believed that their analysis would disclose “faithfully what cyclical behavior is typical of the process represented.” (Burns and Mitchell, 1946, 467; 33)
- 2 Wesley C. Mitchell. The Outlook for 1923. Discussion of Papers by Allyn A. Young, H. Parker Willis (...)
- 3 See Wesley C. Mitchell. Research into Business Cycles. Luncheon of Rockefeller Foundation. March 19 (...)
- 4 Group Meeting. December 9, 1937, Diaries of Henry Morgenthau, Jr., Volume 101, Henry Morgenthau, Jr (...)
- 5 See Wesley C. Mitchell and Arthur F. Burns, The National Bureau’s Measures of Cyclical Behavior. 19 (...)
13This notion made them confident enough to make cautious inferences about the future from these assumed typicalities. After all, the business cycle research conducted at the NBER had long been motivated by the goal of forecasting. As Mitchell had noted in 1923, the “task of acquiring control over the business cycle ... rests fundamentally upon our ability to foresee changes in business conditions a little before they happen and take measures to meet them.”2 From very early on, the staff members of the National Bureau hoped to play a crucial part in this venture.3 When Secretary of the Treasury Henry Morgenthau asked the researchers of the Bureau in 1937 “to set up for me the sign posts to look for as we go down a depression and as we reach the bottom of a curve, and also for the things to look for us as we go up,” the researchers therefore agreed (Moore, 1983, 370).4 During six weeks in autumn 1937, Mitchell and his colleagues compared the fluctuations of 487 monthly or quarterly series to the fluctuations of the “general cycle.” In so doing, they hoped to find out which series had commonly led the turning points of the reference cycle (Figure 1), “as clouds precede a storm,” and could thus be used as “indicators” or “barometers” of business conditions (on the expression “as clouds precede a storm,” see Alexander, 1958, 290).5
Figure 1. Overview of Different Time Series That Had Led (Sign “-”) During Past American Business Cycles
Source: Wesley C. Mitchell and Arthur F. Burns. Timing at Reference Troughs. 1937, Box 51, Manuscript of Bulletin on Cyclical Timing, Arthur F. Burns Papers, Eisenhower Presidential Library.
- 6 Wesley C. Mitchell and Arthur F. Burns. Draft for a Letter to Henry Morgenthau, Jr. December 10, 19 (...)
14Most of the series covered the post-war period alone, several ran back to the 1880s or earlier. While some series seemed to lead the turning dates “with considerable regularity,” some coincided with them, and others “commonly or always” lagged behind.6 Out of the American monthly and quarterly series analyzed, the researchers drew up a list of 71 statistical series that had usually led the upward movements of the reference cycles and had thus proved to be “relatively consistent” indicators of business cycle revivals in the past (Burns and Mitchell, 1938, 4). From this list, they again selected those that had shown, among others, the longest and most uniform average lead at past revivals and the smallest and most regular seasonal variations that had to be corrected for (Burns and Mitchell, 1938, 5-6). Their movements, it was hoped, might also indicate revivals in the future. Among the series selected were, for instance, data on industrial building contracts, on passenger car, truck, and paper production, and on department store sales (Burns and Mitchell, 1938, 8-9).
- 7 Wesley C. Mitchell and Arthur F. Burns. Draft for a Letter to Henry Morgenthau, Jr. December 10, 19 (...)
- 8 Frank W. Graham. The Objectives of Business Cycle Research. December 13, 1940, Box 6, Wesley C. Mit (...)
15However, the researchers were cautious in their recommendation of using the indicators for forecasting purposes. Pointing to their finding that “no one who knows the past expects that what happened during any earlier business revival will repeat itself exactly during the next revival,” Mitchell and his colleagues warned that their indicators presented not an “automatic guide,” but merely a “registering device that may be useful to those who are trying to interpret the general drift of current fluctuations in different types of business activity.” (Burns and Mitchell, 1938, 1) As the researchers pointed out, the indicators were not ready to use as is, but required “the exercise of the user’s judgment.”7 The many warnings were undoubtedly due to Mitchell’s instigation, who remained cautious and non-committal even during the crisis and its aftermath when economic expertise seemed increasingly needed. When Mitchell was asked in 1930 to advise the Chamber of Commerce in the face of an increasingly desolate economic situation, he refused any definite advice. Even ten years later, jokes circulated about Mitchell’s dithering attitude, quoting him as saying, “We ought first to commence to get ready to begin to prepare to study the problem of approaching the task of analyzing more scientifically the possibility of undertaking to examine the question.”8
16It was this hesitant attitude that prevented the Leading Indicators from being used at first. While Morgenthau had hoped for expert advice which he could trust and base his decisions on, Mitchell and his colleagues insisted that their indicators were not a “forecasting machine” that produced future knowledge (Burns and Mitchell, 1938, 1). When, in 1937, Morgenthau asked them to interpret the indicators for him, Mitchell declined:
- 9 Transcript of a Call. Re Data on Cyclical Expansions and Contractions of Business Activity. Decembe (...)
I always think that when we have a call from any government—state, municipal, or national—that we should do what we could to provide anything that is asked. But we are debarred by our by-laws, by the very character of our organization, from giving any advice as to policy.9
17Mitchell therefore refused to interpret the indicators “as a member of the staff of the National Bureau. As a member of the staff of the National Bureau, I can give you this set of judgments as to what we think the most reliable indicators are.”10 As Mitchell saw it, business cycle research had “not yet” reached a stage that would “enable them to make consistently successful business forecasts”—a fact that to him presented “reason, not for giving up [the] work, but for pressing it further.” (Mitchell, 1927, 467)
18It was only after Arthur Burns had replaced Mitchell as director of research that the Leading Indicators found widespread adoption. Then, their rise happened all the more rapidly. In 1953, Arthur Burns became chairman of Eisenhower’s Council of Economic Advisers. As “economic watchman for the White House,” Burns had the chance to demonstrate the practical value of the Leading Indicators and to significantly enhance their prominence (Harris, 1953, 65; Business Week, 1953, 45). During his election campaign in 1952, Dwight D. Eisenhower had declared that “never again shall we allow a depression in the United States,” pledging that he would mobilize “the full power of private industry, of municipal government, of state government, of the federal government” if signs of a depression appeared (Burns, 1953, 5). Burns, convinced that “our own economy is on trial before the jury of world opinion against the false claims of communism that it can meet human needs and aspirations better than we can,” was determined to support the president in this endeavor (Harris, 1953, 65). As Burns saw it, “The old Marxist dogma that capitalism is doomed to collapse on the rocks of economic crisis has become a weapon of propaganda, used adroitly and energetically to confuse the uninformed and to stir discontent the world over” (Burns, 1950, 3). The prevention of severe depressions was thus “the crucial problem of our times,” a task in which Burns hoped to play a major role by watching the Bureau’s Leading Indicators (Business Week, 1953).
19In his “Measurement without Theory” critique voiced in 1947, Tjalling Koopmans had claimed that the Bureau’s work on business cycles was of no practical value:
The movements of economic variables are studied as if they were the eruptions of a mysterious volcano whose boiling caldron can never be penetrated. There is no explicit discussion at all of the problem of prediction, ... although surely the history of the volcano is important primarily as a key to its future activities. (Koopmans, 1947, 167)
20Possibly, Koopmans’ charge had further motivated the NBER researchers to demonstrate the practical significance of the Leading Indicators. With Burns and his colleagues of the Council regularly using and publicly commenting on the latest movements of the indicators, the system became known throughout the country, with simple graphs displaying the behavior of the indicators to the common reader. Only a few months after Burns had been sworn in as chairman of the Council of Economic Advisers, the system of Leading Indicators was used by banks, corporations, and research agencies alike.
- 11 Geoffrey H. Moore. Letter to Arthur F. Burns, October 8, 1953, Box 101, National Bureau of Economic (...)
21In the observation and analysis of the indicators, Burns closely cooperated with his colleagues from the NBER. From summer 1953 onward, Geoffrey Moore and other colleagues gathered, analyzed and interpreted current economic data in an ever-increasing pace.11 The “applicability of the National Bureau’s business cycle investigations to the problem of forecasting” became of foremost importance, fundamentally shaping the research agenda of the NBER (Fabricant, 1954, 20; see also Moore, 1958, 28). The researchers were excited. As Moore put it in 1955, they hoped to have discovered “certain basic economic ‘facts of life’ which, while not necessarily impervious to change, might nevertheless prove to be lasting.” (Moore, 1955, 25)
- 12 Arthur F. Burns. Letter to Geoffrey H. Moore, July 8, 1953, Box 101, National Bureau of Economic Re (...)
- 13 Memorandum on Statistical Needs. July 7, 1954, Box 3, Arthur F. Burns Papers, David M. Rubenstein R (...)
22Burns repeatedly asked his colleagues to speed up the times series analysis to be as “up to date as it can be,” as “economic events move so swiftly” that indicators run quickly out of date.12 As a memorandum explained in 1954, “The Council has been handicapped in its work of analyzing and anticipating economic developments by the fact that some important statistics are not available until weeks or months after the period to which they refer.”13 The reduction of the “‘blind’ interval between the time the indicators move and the time data about them can be recorded, assembled, and funneled up” to Burns by his staff therefore became of foremost importance (Harris, 1953, 64).
23In this context, the advent of large-scale electronic computers in the early 1950s seemed like a blessing. Computer programs developed by programmers and researchers of the NBER could adjust time series for seasonal variations and smooth so-called irregular movements within almost no time, it seemed. “Through the speed of computers,” the researchers heralded, statistics could be assembled, analyzed and published faster than ever before (Francis, 1961, 4). The NBER staff was amazed by this “extraordinary speed,” claiming that the availability of electronic computers would make it seem “reasonable to expect better diagnoses of the current state of the business cycle and improvement in businessmen’s forecasts of their own companies’ activities.” (Francis, 1961, 4; Shiskin, 1957, 551)
24The optimism that electronic computers would solve the problem of timeliness in business forecasting, however, soon became clouded by the fact that their utilization necessitated and indeed created time-consuming tasks—a fact that has also been erased from the standard narratives of technological progress. For example, programs had to be written in an abstruse machine-dependent language. While programmers had to design the algorithms on paper, typists then translated the sequences of commands into holes on punch cards. These were fed into a reader to be then executed by the computer which produced the results via a printer.
25But that was not all. In the 1950s, computers were still scarce. Researchers had to obtain (purchase or apply for) so-called “programming time” or “machine time” on computers in different locations. Most of them were located outside metropolitan areas, thus necessitating long-distance travel. In the 1950s, researchers of the Bureau, which, by then, was still located in New York City, had to switch between locations in Washington, D.C., Massachusetts, and Upstate New York. In a 1961 application for support in the acquisition of an electronic computer, the researchers called the “physical distance of the computers for which we have grants of free machine time” a “vexing problem:”
- 14 Application for Support in the Acquisition of an Electronic Computer. April 1961, Box 11, Folder 98 (...)
For some time we operated with the IBM-704 at the Vanguard Computing Center in Washington. Then we obtained machine time in Cambridge, Mass. With the dismantling of that machine we were shifted to Poughkeepsie, a location from which we have recently been dislodged. At our present location in Mohansic (Westchester County, N.Y.) we have night-shift privileges only, and these will end by mid-April. After April we will have to shift to another location. The cost of long-distance operations are not only in the form of traveling time and expense: testing and debugging operations are unduly delayed, we have to buy expensive snatches of time at commercial computer centers in order to avoid extreme delays and constant breaking of commitments; and repeated overnight or odd-hour duty puts unreasonable strain on the members of our programming unit.14
26The use of large-scale electronic computers required additional, even more time-consuming adjustments. As the researchers had to shift from location to location, they also had to work with different computers. For each kind of computer system in use, the researchers had to adjust their programs based on the system’s specifications. In fact, even nominally similar computers differed in features, auxiliary equipment, as well as in input, output, testing and control systems, creating program incompatibilities. As the researchers explained, “These differences necessitate program adjustments with their unavoidable learning costs and adaptation errors.”15 To make matters worse, the computers they developed their programs for quickly became obsolete. While there existed convertibility routines which permitted the use of certain programs on new generations of computer systems, available memory size limited their use in combination with the programs developed at the NBER. The need to develop the necessary program adjustments caused additional delays.
- 16 Improved use of IBM-1130. Electronic Computer Memorandum No. 30. Box 139, Electronic Computer Memor (...)
27September 1966 saw the installation of the first large-scale electronic computer at the NBER. This saved much time, as researchers no longer needed to switch between different locations and computer systems. But even then, machine time was scarce. In numerous memoranda circulated in the 1960s, researchers of the NBER were asked to help reduce time at the computer and between runs. “Prime-shift use of the IBM 1130 has become very tight and turn-around time has increased considerably. Also, we are somewhat pressed for space and it is no longer possible for 1130 users to work on the fifth floor while waiting for their turn on the machine. As a result, more prime time is lost between runs,” explained a memorandum circulated after the installation of the IBM-1130.16 Users were therefore asked to submit their runs, “particularly long ones,” to be run after hours by a machine operator. Even after the installation of the IBM-1130 at the NBER, the need to reduce “machine time” remained of utmost importance.
28Forecasters had always tried to collect and analyze economic data as quickly as possible. After all, their outlooks had to keep up with economic and policy changes. However, the complex and highly time-consuming task of programming and the scarcity of machine time—both before and after the installation of the first electronic computer at the NBER—pushed for new degrees of efficiency. As a result, time series analysis became a highly standardized process, both concerning the algorithms themselves and the usage of the computer.
29From the mid-1950s onward, electronic computer programs could perform all the tasks involved in time series decomposition that were before computed manually. The researchers celebrated this, explaining that electronic computer programs would provide “for all these adjustments systematically and extremely rapidly.” (Moore, 1957, 219) Indeed, once all the preparatory steps had been taken, monthly series could for the first time be corrected and adjusted within minutes. Observations could be turned into “more useful data,” that is indicators, at “lightning speed,” the researchers explained (Shiskin, 1957, 224; Shiskin and Eisenpress, 1958, 438). And the process became increasingly faster. In 1957, Julius Shiskin noted that a full run of the program, for a ten-year monthly series, required less than five minutes on a large-scale computer of the Univac class (Shiskin, 1957, 231). One year later, he reported that it took only 2.3 minutes to analyze a similar series (Shiskin and Eisenpress, 1958, 436). The new degree of speed allowed for a different application of the indicator technique. Determining “the stage of the business cycle at which the economy stands” now seemed to be possible almost instantaneously and continuously (Shiskin, 1957, 222). For the first time, researchers could increase the granularity of their analysis and analyze monthly series, fueling hopes that these changes would allow for new degrees of forecasting accuracy.
- 17 Dwight D. Eisenhower. Letter to Arthur F. Burns, November 10, 1956, Box 1, Arthur F. Burns Papers, (...)
30Incumbent president Eisenhower had won the election in November 1956 by a landslide. As Eisenhower wrote in a letter to Arthur Burns four days after the election, this was, in his view, to no small degree thanks to Burns: “That ‘grateful nation’ that went to the polls was, without knowing it, thanking you for the wise advice you have given to the Administration during the past four years and for the assistance you have been in keeping our nation prosperous and our credit stable. I am a ‘grateful citizen’ as far as you are concerned.”17 Indeed, Burns’s involvement in government had been unprecedented in scale. This also dramatically increased the expectation on Eisenhower and future policymakers to prevent and mitigate future recessions, exerting heightened pressure on the Council of Economic Advisers to accurately predict economic conditions.
- 18 Julius Shiskin. Letters to Fortune. Signals and Noise, February 1964, Box 34, Julian Shiskin, Arthu (...)
- 19 Ibid.
31For a brief period, it seemed as if the computer would deliver on these aspirations. This hope, however, turned out to be in vain. As soon became clear, the computer was unable to separate “meaningful ‘signals’ from random ‘noise’” in data sets.18 Statistically, there was no proper way to adjust the data for the “various non-systematic irregularities in the behavior of the economy-erratic real-world occurrences” as the researchers explained.19 In their search for patterns matching their preconceived notion of the business cycle, the researchers had, until then, relied on judgment to retrospectively correct economic data for those components that did not fit what they had identified to represent cyclical movements. Based on the analysis of past data, they had carved out average cyclical patterns for each time series against which they had matched and “corrected” current data (Burns and Mitchell, 1946, 151-154). Indeed, as Mitchell and Burns had explained in a side note to their 1946 volume on Measuring Business Cycles, they had excluded data from the analysis “when some exceptionally powerful random factor, such as a great strike, has warped an individual cycle out of resemblance to other cycles in the array.” (Burns and Mitchell, 1946, 33-34) Also, they changed their course of analysis “[w]hen the cyclical behavior of a long series gives definite indications of having undergone a secular or structural change.” (Burns and Mitchell, 1946, 34)
32Such exercise of professional judgment was obviously impossible for an electronic computer. The computer could detect neither “unrepresentative periods” nor “highly extreme individual items, such as arise from strikes,” as the researchers had not found any formula or automatic technique that would allow the computer to delineate the underlying cyclical movements (Shiskin and Eisenpress, 1958, 437; 419). This, however, was made all the more necessary by the recent increase in granularity which the computer had brought about. As is turned out, the irregular factor was disproportionately large in the newly available monthly series, dominating the cyclical factor to a high degree (Shiskin, 1957, 237; 239). As a result, most month-to-month movements of seasonally adjusted economic series were not “cyclically significant” (ibid.). As the researchers noted somewhat disillusioned, the indicators were “continuously fluctuating” (Francis, 1961, 4).
33To mitigate this and to nevertheless allow for forecasts based on the monthly data, the researchers introduced an analysis based on moving averages of time series spanning multiple months. As Julius Shiskin explained, “For most series, meaningful economic trends are revealed only by comparing changes over three month or longer spans.”20 While the use of moving averages smoothed the irregular factors, it introduced a new set of problems, though. Using averages entailed the risk of eliminating whole cyclical movements if they were centered on the averaging window (Shiskin, 1957, 239). Also, the principle of moving averages also made it impossible to generate statements about current business conditions. As Shiskin pointed out, “[A]ll moving averages suffer from the handicap that they do not reach to the current month. Since they are centered at the middle month of the interval covered by the average, there are no values for the last month(s).” (Shiskin, 1957, 239) As a result, forecasts could still not be based on data from the immediate past, that is on the most recent data. All efforts to increase the speed of time series analysis had proved futile in the researchers’ attempt to “capture” the present and predict the future. The motion picture of the present that the researchers had hoped for still only showed the past.
- 21 Wesley C. Mitchell. Letter to Henry S. Dennison, February 5, 1924, Box 8, Wesley C. Mitchell Papers (...)
34Since the formation of the NBER in 1919, Wesley Mitchell and his colleagues had advocated a strictly empiricist approach, dismissing any form of intervention or interpretation as “smack[ing] more of metaphysics than of science.” (Mitchell, 1919b, 231; see also Mitchell, 1925, 4-5) Along these lines, quantitative analysis was described as a means to safeguard economists from “deserting the firm ground of measurable phenomena for excursions into the subjective.” (Mitchell, 1925, 4) When their colleagues of the Harvard Committee of Economic Research started to base their forecasts not solely on their index anymore but on the information gained from personal contact, Mitchell had expressed irritation. As Mitchell, an early admirer of the work of the Harvard Committee (Mitchell, 1919a), warned in 1924, “[T]he two types of forecasts”—the “mechanical” one given by their statistical series and the one “relying on general information or bread observations or tips or ‘hunches’”—“ought to be kept separate.”21 (see also Lenel, 2021a, 154)
35Their “struggle against subjectivity” (Porter, 1995) and the push for what Mitchell called “mechanical” procedures had defined their research on business cycles and forecasting from its very beginning. Mitchell’s dithering attitude and the researchers’ strict empiricism had formed the target of jokes and criticism during the 1930s and 1940s. And yet, the researchers’ approach and rhetoric of standardized, “objective” business cycle analysis had also served as “technologies of trust” that helped establish the NBER’s remarkable and long-lasting authority when other business forecasting agencies had to shut their doors (Porter, 1995).
36The use of electronic computers in business cycle analysis proved, however, that the statistical methods developed by the researchers of the NBER to forecast business conditions were useless if not complemented by judgment. Business fluctuations, it turned out, became recognizable only retrospectively by manual procedures of selection and weighting. As the use of computers showed, the Leading Indicators, in themselves, provided no knowledge about the future. They not only required the additional exercise of the user’s judgment, as explained by Mitchell and Burns in 1938, but indeed heavily depended on human judgment. This, however, was incompatible with the researchers’ ideal of mechanical rule-following.
37An unforeseen recession put the final nail in the coffin. When the Leading Indicators failed to forecast the 1960 recession, the researchers began to issue warnings to not overestimate the forecasting power of their indicators. As they explained, random factors such as the steel strike in 1959 could distort the indicators (Shea, 1961; Moore, 1961). What is more, the time by which the Leading Indicators led a recession had varied considerably, signaling the 1960 recession twelve months and the 1957-58 recession 18 months before they began (Cameron, 1961). Moore therefore stressed that the indicators merely helped to recognize business fluctuations “at about the time they occur” or even after a peak or trough had been reached (The Weekly Bond Buyer, 1961, 4; Shea, 1961). Indeed, only a combined analysis of the leading, the coincident, and the lagging indicators was reported as yielding reliable results (Shea, 1961). The researchers therefore cautioned against basing policy-decisions on a hasty interpretation of the indicators. “The business forecaster operates in a continuum, and the evidence for or against the judgments he must make from time to time accumulates months after month,” noted Moore (The Weekly Bond Buyer, 1961, 5). As a consequence of their new cautious stance, the researchers ceased their efforts to forecast economic fluctuations. Instead, they confined themselves to keeping their “indicators up to date for the benefit of other business analysts.” (Business Week, 1961, 10)
38In 1961, the National Bureau became the government’s “accepted judge of U.S. recessions,” announcing peaks and troughs in economic activity for the Commerce Department, which published the turning point dates in Commerce Department publications (Hall, 1979). Seventeen years later, the Business Cycle Dating Committee was created, thereby formalizing the process of “track[ing] the country’s economic cycles.” In their role as “arbiter of business cycles,” the researchers did not attempt to forecast economic fluctuations anymore, but instead announced a peak in economic activity and the onset of a recession well after they had begun (McClain, 1990; Hall, 1979). As Robert E. Hall, chairman of the Bureau’s Business Cycle Dating Committee, explained in 1979, “By the time we make our announcement, much of our audience may have lost interest in the subject, but making news is not our primary purpose.” (Hall, 1979)
- 22 Wesley C. Mitchell. Types of Business Cycle Theory. Midday Club, Meeting No. 7. February 25, 1934, (...)
39Although valid for 1979, this had not always been the case. In the 1930s, shortly before publishing the first report on the indicator technique, Mitchell had explained that the hope to find a way to forecast and control business fluctuations was “what really justifies all the time I have put in on this job.”22 The computer, however, had turned the researchers’ attention away from the future and towards the past and thus fundamentally changed their focus.
- 23 Wesley C. Mitchell. Research into Business Cycles. Luncheon of Rockefeller Foundation. March 19, 19 (...)
40As this article argues, the use of electronic computers at the NBER demonstrated that the forecasting validity of the Leading Indicators depended on human judgment correcting for the inherently irregular nature of the business cycle. This made the use of the Leading Indicators for forecasting purposes incompatible with the researchers’ ideal of objectivity and prompted the researchers to ultimately abandon or outsource their forecasting practices. In the United States, the Leading Indicators are nowadays most prominently used by the Economic Cycle Research Institute established by the former NBER researcher Geoffrey Moore, which presents itself as “the ‘secret weapon’ of companies from Disney to DuPont.” (Achuthan and Banerji, 2004) Back in 1928, Wesley Mitchell had called it “unfortunate” that forecasting was largely carried on “as a commercial venture.”23 Fifteen years prior, he had claimed that “[o]ne way of increasing social control over economic activity is ... to democratize the knowledge of current business conditions already possessed by a few.” (Mitchell, 1913, 588) And yet, their own ideals prevented the researchers from continuing their forecasting efforts.
41To this day, economic forecasts are often being presented as the result of mechanical rule-following. Practices of expectation exchange, of “foretalk,” on the other hand, usually take place on the backstage (on this term, see Gibson, 2011, 504; Gibson, 2012; Reichmann, 2013). The importance of judgment, intuition, and imagination is equally downplayed. Neither fits the image of the objective, impartial economic expert that forecasters usually seek to convey. The history of the end of forecasting at the NBER demonstrates that early on, researchers were well aware of the limits to mechanical rule-following. Rather than taking the presentation at face value, the rhetorics of standardized, rigid analysis should be understood as what Theodore Porter called a weak and vulnerable community’s response to conditions of distrust (Porter, 1995, xi).
42The faith in mechanical rule following and the general planning euphoria of the post-war period are often described as the features of high modernity (on the planning euphoria of the postwar period, see, e.g., van Laak, 2008; Christian et al., 2018). Here, the years of the “boom” are portrayed as a strange place, populated by researchers with an “obsessive narrowing of vision,” “mind-bogglingly implausible assumptions,” and “towering ambitions” (Erickson et al., 2015, 24). Only the crises of the 1970s, according to this view, ended the “downright bizarre” faith in progress, science, and planning (Erickson et al., 2015, 10; see also Doering-Manteuffel and Raphael, 2008). This article revisits an instance which illustrates that researchers realized the inaptitude of their approach of mechanical rule following already in the early 1960s and indeed abandoned their forecasting practice when faced with the unattainability of their ideal. The pursuit of predictability and control, it seems, already contained its disappointment. This case study thus presents another reminder to question the dichotomous view of the post-war period and instead examine what Jenny Andersson has dubbed “the many contradictions in Cold War science” (Andersson, 2018, 27; see also Engerman, 2015, 575). Rather than embracing the self-portrayal of historical actors, we should examine more closely the limits of this representation. The past may not be such a strange place after all.
I am grateful for the criticism and advice from participants in the workshop on “The Computerization of Economics,” my colleagues at the department of economic history at Humboldt-Universität zu Berlin, and the members of the German-Swiss network “Wissen und Wirtschaft.” I especially thank Verena Halsmayer, Marcus Mikulcak, and two anonymous referees, whose close and critical reading contributed greatly to the clarification of my argument.
Flux RSS
