2. FRISCH AND THE INSTITUTE OF ECONOMICS

The University of Oslo was at the time the only university in Norway. There were in 1930 two professors in economics but no department of economics. Economics belonged to the faculty of law. A separate economics study had been established in 1908 as a two-year study, primarily meant as a complementary study for law graduates and others. Economics study alone did not give access to higher positions in the government administration. Doctoral degrees were awarded but not many. From when the university was founded in 1811 until 1930 there had only been six or seven doctoral dissertations on economic topics and only one in statistics, namely, that of Frisch in 1926.

Until Ragnar Frisch appeared on the scene, Norway did not have economists and statisticians of international renown such as Sweden's Knut Wicksell and Gustav Cassel and Denmark's Harald Westergaard. There was a national tradition in economics; academic economists had played an important role in economic policy, in social issues, and in promoting statistics. It was a small but influential profession. The economics study around 1930 does not in retrospect seem to have attracted many talented students.²

Frisch, born in 1895, had graduated in economics in 1919 while being apprenticed to become a silversmith. After completing both educations he spent two to three years abroad studying mathematics and statistics in Paris, also visiting for study purposes other European countries. The object of his studies was, as he had told an acquaintance in 1926, to find “methods of submitting the laws of pure economic theory to a numerical verification by an intensive study of economic statistics.”³

Even after such preparations for a career in economics Frisch may, however, as late as 1929 have been in a dilemma as to whether it was possible for him to pursue his scientific interests in economics and statistics, in view of his responsibility toward his family, which meant taking over the family jeweler's business. A point of no return came when in the same year Irving Fisher instigated an invitation for Frisch to spend one year at Yale. The year and a half he spent in the United States in 1930–1931 gave him the opportunity to develop all his scientific ideas and prepare a comprehensive research program. At the University of Oslo the visit created the imminent fear that Frisch would accept an offer from Yale for a permanent position. While Frisch was in New Haven, the two professors in economics maneuvered shrewdly to have the Norwegian Storting (parliament) by an unusual act appoint Frisch as full professor at the University of Oslo, effective from July 1, 1931.

After his return to Norway in June 1931 Frisch negotiated with the Rockefeller Foundation about funding a research institute at the university as it had done in Denmark and Sweden. The Rockefeller Foundation was impressed with Frisch's plans and after a new round of negotiations in September the same year donated funds to support an empirical research institute.⁴

Frisch had a lifelong passion for numerical calculations and was exceedingly adept at developing efficient algorithms and organizing large-scale numerical work. The availability of computational tools was limited; the most expensive equipment was beyond what Frisch could afford. But simple tools used by assistants adhering strictly to meticulous computing schemes masterminded by Frisch, usually with specially designed calculation sheets, went a long way toward satisfying the need. Frisch even invented some computation equipment and had it built to order. The process of numerical analysis and experimentation was a sine qua non in Frisch's laboratory; it was crucial for his most ambitious projects and often included stochastic elements, using random numbers from lotteries.

Frisch's position had since 1931 been that of a university professor. His main responsibilities were teaching and other academic duties. The institute was a separate responsibility, an institute at the university but not part of it. Somewhat separate from both university and institute was Frisch's active international role. He had played a key role in the establishment of the Econometric Society and in organizing the first two European meetings of the society in 1931 and 1932. He became the first editor of Econometrica in 1933. Of course, to Frisch his different roles were quite intertwined.

3. FRISCH'S RESEARCH PROGRAM FOR THE INSTITUTE OF ECONOMICS

The research program for the institute in the early years comprised problems and research areas rooted in Frisch's work in the 1920s, on which he had worked intensively while in the United States from 1930 to 1931. They were all econometric problems in the sense of the word Frisch had written into the constitution of the Econometric Society.

In 1926 Frisch had demonstrated how marginal utility, and more particularly the income elasticity of the marginal utility of income, could be “quantified” by making the theoretical assumptions precise by means of mathematical formulation and then confronting them with statistical data, adapted and interpreted by economic theory. In this work he expressed one of his key tenets in empirical analysis, namely, the primacy of theory, and demonstrated how that would lead the way for numerical estimation. At Yale he reworked his ideas and extended the range of applications of his approach to estimate marginal utility in the New Methods monograph.⁶

A second theme was to establish an appropriate framework for data analysis as a foundation for econometric analysis. This was really Frisch's motivation and rationale for pursuing economics, inspired not least by the lack of rigor he found in others.⁷

Frisch had also in the 1920s put much effort into the key research topic in economics at the time, the analysis of business cycles. He was familiar with Wesley Mitchell's work and had become acquainted with him during his visit in 1927. To Frisch the mechanical periodogram analysis of early researchers was much too rigid and simplistic. He first aimed at developing more flexible methods allowing for cycles of different lengths and had during his first stay in the United States written a methodological paper, which Wesley Mitchell showed interest in.¹⁰

By 1930 Frisch's ambition in business cycle analysis had advanced far beyond his 1927 paper, which he also recognized had shortcomings. He had been in touch since 1925 with Eugen Slutsky, who alerted him about his 1927 contribution as soon as it appeared. Slutsky also made Frisch aware of Yule's article the same year.¹¹

To corroborate, explore, and extend the idea Frisch drew up a comprehensive research program. While he was in the United States in 1930–1931, he had engaged students at Yale and the University of Minnesota in numerical exercises, e.g., identifying cycles in constructed data sets with superimposed cycles and random disturbances. He also collected various long time series for experimental purposes and offered his time series methods for use in other fields. The institute gave him the chance to embark on the research program he envisioned.¹²

Yet another topic at the institute rooted in the 1920s and worked on at Yale was production theory, or “productivity theory” as Frisch called it at the time. Frisch had taught production theory for the first time in 1926, found flaws in earlier presentations, and decided to incorporate production theory as one of his econometric topics. That implied “quantification,” i.e., mathematization of the theoretical concepts as the basis for confrontation with data. Frisch had attended the American Economic Association meeting at which Paul Douglas presented the first empirical application of the Cobb–Douglas function. Frisch's enthusiasm about the topic was mixed with dismay over shortcomings in the Cobb–Douglas article.¹³

As the Institute of Economics at the outset had only one source for its funds, the research program of the institute coincided with the research program agreed upon in the contract with the Rockefeller Foundation. Frisch and his colleague Wedervang had negotiated with John Van Sickle from the New York office of the Rockefeller Foundation and committed themselves to letting the institute serve “as a tool of advancing scientific economic research in Norway, as means of making the teaching of economics more effective, … [and] as a means of organizing cooperation between Norwegian industry and commerce and the economic research work at the University.”¹⁴

Frisch had his research agenda already set. He coped with Rockefeller's requirements by emphasizing only the more appealing parts of his planned research in the proposal. The reports he gave about research done, included, however, all activities.

In 1936 Frisch reported on the work of the first four years of the institute using the following subdivisions:

These headings covered the portfolio of projects at the institute, including those financed from other sources. Projects under II and III had figured prominently in the original proposal submitted to Rockefeller, whereas I had been somewhat disguised as time series analysis of long series of price and wage data and IV not mentioned at all.

Frisch's high-flying methodological research ambitions, pertaining particularly to I and IV, were thus a bit camouflaged in his communication with the Rockefeller officials but not for long. John Van Sickle's and the Rockefeller Foundation's high enthusiasm at the outset, influenced by Schumpeter's praise of Frisch's talent, waned quickly. After one year Van Sickle found to his dismay that Frisch devoted himself to “highly abstract mathematical theory.”¹⁶

Shortly after the institute was founded Frisch had hit upon the idea of national accounting and in 1934 asked the Rockefeller Foundation's Paris office for an additional grant to pursue national accounts for Norway it. It was a pity, perhaps, that he did not get it, as this could have become a pioneering effort of great usefulness and helped to advance macroeconomic research. But Frisch had already ruined his chances.¹⁷

In 1936 Van Sickle referred to what went on in Oslo as “abstruse mathematical theoretical work” in an internal memo on how to disengage from further support.¹⁸

Of particular interest for the theme of this paper is Frisch's thinking at the time about the role of probability in econometrics. His two major research ideas to be pursued at the institute comprised on the one hand the impact of random shocks on a dynamic economic system and on the other how to identify (simultaneous) relationships among variables with errors of measurement. In both he may be said to have come near to considering probability theory at center stage. He expressed at the same time a very general view of the role of probability in economic relationships, as can be gauged from a letter Frisch wrote from Yale to Joseph Schumpeter at Harvard in December 1930 a few days before they would both meet in Cleveland at the foundation of the Econometric Society:¹⁹

Here Frisch thus viewed economic relations as having deeply embedded stochastic elements. The formulation allowed different interpretations as to the source of the stochastic variation. His very general formulation of the role of stochastic influences in economic relations seems in retrospect to have been an excellent starting point for elaborating a probability approach to the estimation of the parameters. It is almost, but not quite, something that can be imputed to Frisch on the basis of the last paragraph quoted here. Frisch did not choose that route, and that in itself is worthy of a question why. Instead he returned to the framework of his “Correlation and Scatter” essay and adapted it to become the confluence analysis cum bunch maps.

4. CONFLUENCE ANALYSIS

Frisch's Confluence Analysis aimed at dealing with regressions when there was more than one (linear) relationship connecting the variables under consideration. With no errors, either in equations or in observed variables, an attempt to determine regression coefficients when there were multiple relationships would give 0/0 expressions. When the relations or observations were contaminated by random errors, the same expressions would be determinate but really without meaning. To use a favorite expression of Frisch, one would have “fictitious determinateness created by random errors.”²⁰

Frisch assumed that the observed variables generally were measured with errors but did not include error terms in the equations. His analytic framework was computation intensive as the “complete tilling” of the data that he firmly recommended at the outset comprised finding the regression slopes of all subsets of the variables in all minimizing directions. Then his highly inventive graphical diagnostic tool, the “bunch map,” would display the computation results in a way that revealed simultaneity problems caused by confluent relationships and multicollinearity and pave the way for the selection of regression equations to be included in a model.²¹

The confluence analysis was rooted in the cluster theory set out in the “Correlation and Scatter” essay.²²

The first long-term visitor at the new institute was Frederick V. Waugh, who stayed in Oslo most of the academic year 1932–1933, working closely with Frisch.²³

The Confluence Analysis book, rooted in Frisch's 1920s work and practical experiences during the visits of Waugh and Belz, was written over the course of three months in the spring of 1934. It was not premeditated as a monograph, and its content partly gives the impression of being a laboratory progress report.²⁶

Frisch tackled head on his abstention from probability considerations in his methodological approach, as if the issue had been discussed between him and others, which rather likely may have occurred. He argued that “sampling theory” (i.e., probability theory in inference) did not work when a regression equation included variables fulfilling two simultaneous equations. An attempted regression would have indeterminate parameters, but the standard formulas could very well give fictitious determinateness created by random errors.²⁸

In the introduction to Confluence Analysis Frisch set out the nature of his investigation, referring to the weakness of the method in his earlier work as its lack of criteria for judging the significance of the “scatterances,” as defined in “Correlation and Scatter”:

He might well have had in mind here the “general set of assumptions” in the letter to Schumpeter. Instead of basing his approach on an explicit consideration of the probability aspect Frisch offered his method as an “attack on the problem more from the experimental side, working out numerically—on actual economic data as well as on constructed examples—various other types of criteria which intuitively and heuristically may suggest themselves. These experiments converged toward a definite method which after applications to various kinds of data was found to give satisfactory and plausible results” (pp. 7–8).

After these initial remarks he did not discuss probability further, except in a passage in the middle of the treatise where he defended his decision not to give a rigorous definition of probability by referring to the arbitrariness of choosing one specific probability scheme, adding, “I believe it will be a better application of time and energy to work experimentally with the method and rely on one's intuitive judgement of whether a given spread in the various determinations of a given regression coefficient is reasonable or not.”²⁹

The confluence analysis attracted, as the New Methods monograph had done, both students and scholars to Frisch as a generator of new and fruitful ideas, but his approach also met with resistance. His presentation was criticized for being too hard to penetrate and his methods for being too computation intensive and—more important—also for the lack of exact criteria. Frisch was not moved by the computation argument; it was only a question of organizing the calculations in an efficient way, and in this he was right.³⁰

This and other passages thus show Frisch shying away from adopting an overall probability framework but not as the outcome of a decidedly antiprobabilistic stand, more as a temporary position of pursuing one research idea at a time. Frisch's rationale was influenced on the one hand by the urgency he felt to provide tools that could be helpful and on the other hand by the conviction that there did not exist an all-encompassing probability framework as an operational tool for the problems Frisch had chosen to deal with.³¹

5. APPRENTICED IN THE ECONOMETRIC WORKSHOP, 1933–1936

Trygve Haavelmo, born in 1911, had begun his study of economics in 1930. He said in an interview after being awarded the Nobel Prize that his study inclinations had really been toward engineering or languages. Economics was chosen because only such a short study could be afforded. Frisch was in the United States in the first year of his study but was present and teaching after that. Haavelmo graduated in June 1933 and was hired by Frisch as an assistant at the one-year-old institute, still in its establishing phase, and which, despite its name, was not part of the university.

Frisch had hired a number of assistants, students and young graduates, and acquired the computational equipment he could afford. He stretched the Rockefeller means by offering relatively low pay. The assistants were primarily “computers,” and much of the work can surely be described as endless and tedious calculations. Perhaps the laboratory was run with some similarities to the silversmith's workshop that Frisch also managed.

Thus Haavelmo was thus hired to be a “computer.” It may seem likely that Frisch had discovered Haavelmo's suitability for laboratory work before he graduated, as he had taken part in Frisch's statistical seminar, but he was not offered particularly favorable conditions.

Frisch was 38 years old and Haavelmo 21 when they entered into the master-apprentice relation in 1933. The distance between the two men in terms of social and cultural background, and also cumulated intellectual capital, was at the outset large and remained large throughout the 1930s. Frisch was from a well-established liberal-bourgeois background with a forefather who, according to the family tradition, at the king's invitation came to Norway from Germany as a mining specialist in the seventeenth century. Haavelmo had rural roots; his father had chosen a teachers' college rather than emigration to escape the impoverished overpopulated countryside.³²

Haavelmo's work assignment in the first period at the institute, in addition to his tasks as Frisch's personal assistant and secretary, included calculations, much of them related to confluence analytic problems, which meant “tilling” of data and construction of bunch maps. Haavelmo was involved in the efforts undertaken in Frisch's work with Maurice Belz in 1933–1934. Confluence analytic computations thus became Haavelmo's first introduction to econometric problems. Haavelmo became extremely proficient in the kind of calculations that the bunch map analysis required. He was actually the first person to read the Confluence Analysis monograph from beginning to end, as Frisch passed the responsibility for proofreading on to Haavelmo as he rushed off to visit the Cowles Commission at Colorado Springs in July 1934.³⁸

Haavelmo's apprenticeship with Frisch has left few documentary traces of Haavelmo's activities at the institute in the period 1933–1935. He did not publish anything, nor has there survived much in terms of internal documents with Haavelmo's name on them. Any assistance that he extended to Frisch with regard to his publications, such as, e.g., Confluence Analysis, was not acknowledged. Why was this so? Haavelmo's intellect at the age of 21 to 23 must have been as bright as it turned out to be later, although his fund of knowledge naturally was smaller at the outset. With an element of pure speculation, one might surmise that it was due to Haavelmo's time being fully absorbed in computing and studying or due to Frisch not fully having recognized Haavelmo's potential. A third reason could be a lack of a definite commitment on Haavelmo's part to become a laboratory-trained econometrician, as the institute neither offered career prospects nor much pay.³⁹

The evidence on Haavelmo's motivation and commitment is that in the spring of 1935 he applied for a job as clerical assistant in the social security administration.⁴⁰

Very shortly after Haavelmo got this offer Frisch redefined Haavelmo's position to become “chief computer” and doubled his pay. It could be tempting to read a game play into this outcome. Although Haavelmo is known to have enjoyed poker games with other assistants, this seems out of character. The pay raise (with a transfer of some administrative duties) perhaps reflected a (belated) recognition on Frisch's part that Haavelmo was as good an apprentice as he could ever hope for. Most important is the impression one has that from 1935 on Haavelmo's position relative to Frisch had moved at least a little in the direction of becoming more that of a co-worker rather than a lowly assistant.

Haavelmo's motivation and commitment did not seem to falter after 1935, but what made that year a point of no return for Haavelmo may well have been something else, such as the visit by Tjalling C. Koopmans. In March 1935 Koopmans wrote to Frisch, whom he had never met, and asked to come to stay for the autumn term. He indicated his chief interest as being “the problems, arising from the circumstances that classical sampling theory does not regard cases in which observational series develop in time in such a way, that the probability distribution of the second term is not independent of the value attained by the first.”⁴²

Frisch gave an elaborate and somewhat critical answer to Koopmans's suggestions:

Then Frisch's current concern with confluence analytic problems came to the forefront, as well as an outright invitation to Koopmans to become a student of confluence analysis:

Thus it happened that Tjalling Koopmans spent the autumn of 1935 at the institute, where he gave a series of lectures under the title On Modern Sampling Theory.⁴⁴

Frisch's skepticism toward probability reasoning as a fruitful approach in practical econometric work thus must have come under scrutiny during the visit. Further testimony on Frisch's view during Koopmans's stay can be gathered from a letter he sent to an American statistician who wanted to visit Oslo.⁴⁵

One may find it a pity that Frisch did not choose to give a “fuller statement.”

Koopmans's doctoral dissertation thesis gave a thorough discussion of Frisch's confluence analytic approach.⁴⁷

Koopmans's lectures were attended by Frisch, Haavelmo, and Reiersøl.⁵⁰

One should perhaps also consider the effect that Koopmans's visit had upon Frisch's thinking. First, he was clearly very enthusiastic about Koopmans's lectures and insisted on having them mimeographed and distributed to various of his acquaintances.⁵²

Frisch involved Haavelmo in his most ambitious and certainly most time and effort consuming project, namely, the study of time series and business cycles. It was the “economic dynamics” project, in the institute jargon just “shock theory.” Frisch had completely reoriented his original time series analysis after studying Slutsky and Yule, and he utilized their ideas in his propagation-impulse cycle-generating model of the macroeconomy. The macrodynamic structural model of the economy worked according to Frisch's idea as a linear operator of the random disturbances to which it was exposed and produced cycles through the mechanism that Slutsky had described.⁵⁴

Frisch's method in the economic dynamics project was mathematical analysis and numerical simulations, endless numerical simulations.⁵⁵

Haavelmo and Reiersøl were thus Frisch's two key operators in trying out a large number of hypotheses and conducting enormously extensive calculations. To show that the Slutsky–Yule–Wicksell mechanism could produce cycles that simulated observed ones, as shown in the propagation-impulse model, was by itself an achievement but only the first step on the road. Frisch aimed at establishing a general theory for determining “the exact nature of the cycles which are created when a linear operator is applied to a random series” and professed on occasion to have found the general solution to that. Further along the road was the “inversion problem,” i.e., to determine from a given time series produced by such a mechanism the weight curves by which the random disturbances have been accumulated.⁵⁶

In addition to working within Frisch's two main scientific paradigms, Haavelmo took part in a number of empirical studies falling under demand and production studies. He was also Frisch's teaching assistant, drafting and editing lecture notes. Parts of Frisch's lecturing were on his research frontier, such as Macrodynamics (1933–1934) and Time Series Analysis (1934–1935).⁵⁹

In the autumn of 1935 Frisch would lecture again on monetary theory. From his visit to Cambridge in 1934 and contacts with Cambridge economists he was well aware that Keynes was working on a new theory. Motivated by a wish to give his students the most updated theory, he had the following brief letter exchange with Keynes:

Keynes left Frisch with no choice. He decided not to heed Keynes's admonition and with Haavelmo as teaching assistant dissected the Treatise of Money once again.⁶¹

Frisch solicited financial support from Norwegian sources, offering statistical and economic analyses using the new tool of econometrics, partly for financial reasons but also, and perhaps more important, to prove to the Rockefeller Foundation and also to the Norwegian public and authorities the social usefulness of econometrics. Haavelmo was the key investigator in several of the empirical studies that Frisch contracted to do at the institute.

One study was undertaken in 1934 for the breweries' association, whose concern was that deflation under nominal taxation had caused a doubling of the real price of beer whereas the consumption was halved. The breweries wanted corroborated evidence to convince the government that lower taxes would increase tax revenues. The task seemed simple, a question of determining the price elasticity. Direct regressions gave highly uncertain estimates around −1 and revealed high multicollinearity. Frisch used his connections with the statistical bureau and borrowed the entire data files for the most recent household surveys, from which Haavelmo estimated a short- and long-term price elasticity of −1.74 and −1.55, respectively. But Frisch was still not satisfied; he produced a questionnaire and sent students out to interview acquaintances about their reactions to hypothetical changes in the price of beer.⁶²

The report to the Rockefeller Foundation on findings resulting from a study of the demand for eggs in Oslo in 1935 may have been written by Frisch tongue in cheek to suit the foundation's request for “inductive research of a more realistic nature”:

On his appointment to “chief computer” in 1935 Haavelmo had bought a used Harley-Davidson motorbike, an unusual means of transportation among the institute employees. It saved time going to work, Haavelmo's home was about 20 kilometers from Oslo. He was fond then, as during the rest of his life, of the wilderness, as, indeed, was Frisch. But it was not really a common interest. Whereas Haavelmo's ideal pastime was fishing and pipe smoking miles away from other human beings, Frisch's outdoor activity was hiking and climbing mountains. Hours of solitude at a lake may have been when Haavelmo conceived his original ideas, whereas Frisch seemed to derive from hard physical exercise in mountain air his legendary ability to work on a problem for days on end. (Cross-country skiing was, however, an activity they both enjoyed into old age.) Haavelmo was highly regarded by the other staff at the institute in the supervisor position in which Frisch had placed him. All of them would surely have concurred in what Frisch wrote in introduction letters in 1937—that he was convinced Haavelmo “in the future [will] do excellent work in his chosen field,” adding “that Mr. Haavelmo is a perfect gentleman, whom I have always trusted in all matters.”

6. PROBATION WORK COMPLETED, 1936

As apprenticed to become an econometrician, Haavelmo was in a very young profession! The mark of completed apprenticeship was, naturally, membership in the Econometric Society.⁶⁵

Early in 1936 Haavelmo must have worked on the paper “Confluent Relations as Means of Connecting a Macrodynamic Subsystem with the Total System,” to be presented at the sixth European meeting of the Econometric Society at Oxford in September 1936. Frisch and Haavelmo shared a cabin on the voyage across the North Sea on their way to Oxford. Frisch had been prominently present at all but one of the five previous meetings. Haavelmo had never been outside Scandinavia. The Oxford meeting was the largest meeting so far, with 64 participants. Among participants Haavelmo would have more contact with later in his career were only a few apart from Frisch: J. Marschak and J. Neyman, primarily the Geneva group of H. Mendershausen, H. Staehle, and J. Tinbergen. The meeting opened on Saturday, September 26.

Top billing on the agenda was the discussion of Keynes' General Theory, a topic of great interest to Haavelmo.⁶⁶

On Sunday morning R.G.D. Allen presented “The Assumptions of Linear Regression,” a paper in a Frischian vein (even Frischian notation!) on the determination of limits for the true regression coefficient between two variables with measurement errors.⁶⁹

On Monday morning it was Haavelmo's turn. The topic was chosen entirely within Frisch's paradigm and more precisely was meant to show the applicability of Frisch's macrodynamic modeling scheme. The problem Haavelmo posed was what to do when the structural equations representing the issue under consideration were a subsystem and mathematically underdetermined. Haavelmo discarded the alternatives of adding enough structural equations (“press more theory into the subsystem”) and letting endogenous variables become exogenous, which meant they had to be specified rather arbitrarily as time functions. Instead he argued for adding sufficient “confluent relations” that fitted the data reasonably well. The underlying rationale was that the macrodynamic methodology called for eliminating all but one variable to get to the “final confluent relation” and only then to estimate. The numerical values of the estimates would reveal the dynamic properties of the model. In an empirical illustration Haavelmo also employed bunch maps.⁷¹

Before Frisch left England he spent an evening at Neyman's home. Frisch showed a great interest in the Neyman–Pearson theory of testing hypotheses and had drafted a note on it to make sure he had got it right.⁷²

After the meeting Haavelmo remained in England for a couple of months, until the beginning of December, to study statistics at the Department of Statistics, London University College. In the report he submitted to the university after his return he stated the motivation for the study visit as his being “specially interested in the problem of using sampling theory in economic statistics.” He followed lectures by Egon S. Pearson on general statistical theory and by Jerzy Neyman on the topic “Testing Statistical Hypotheses” and on orthogonal polynomials. In his report to the university Haavelmo also mentioned that the library facilities and document collections at the London School of Economics had given him the opportunity during the visit to go through the most important parts of the modern literature on statistics and probability theory, which presumably also included the first volume of Neyman's Statistical Research Memoirs, just published.

Haavelmo did some numerical work with Pearson, who also gave him referee assignments for Biometrika. He worked with Robert Jackson, a research worker at the department, with some participation also by Neyman on numerical tests for regression coefficients in confluence analytic problems. Jackson and Haavelmo planned a joint paper, but the only result from the stay seems to be a memorandum by Haavelmo entitled “Standard Errors on Regression Coefficients in Multivariate Sets.”⁷³

Shortly after Haavelmo's return Frisch proposed him as a member of the Econometric Society. The apprentice had come to a new stage; surely the meeting and the two months in London must have whetted Haavelmo's appetite for spending time with other econometricians, not least people such as Tinbergen and Marschak.

From 1937 on the institute had become involved in a government-financed project to explore Norway's production possibilities. Frisch had offered to participate and had gotten approved as part of the project, an attempt to develop national accounts, which ranked high on his own agenda. Haavelmo did not get very deeply involved but also had his share of the initial stage of that project.

Frisch's fourth group of research topics, productivity studies, had been relatively neglected. Frisch had demonstrated his ideas when he pioneered an engineering production function approach in a study of chocolate production in 1935. Estimation of production functions became, however, a topic in the laboratory in 1937. Oslo's biggest bakery commissioned a project to investigate the quality of its bread; the econometric analysis revealed that the dough needed more water!⁷⁵

The Cobb–Douglas production function also turned up on the agenda. When Douglas reminisced toward the end of his life about his 1928 article with Cobb, he mentioned “such critics of the production analysis as Horst Mendershausen and his mentor, Ragnar Frisch.” Without reference he quoted the two critics as holding that Douglas's study had so few observations that any mathematical relationship was purely accidental and not causal and that all past work should be torn up and consigned to the wastepaper basket!⁷⁶

A study of the demand for milk was published as a joint paper.⁷⁸

7. JOURNEYMAN TRAVELS, 1937–1939

Haavelmo had expressed to Frisch his interest in a longer study visit abroad. Frisch advised a stay of at least two years. In the spring of 1937 Haavelmo was awarded a Norwegian grant for “further study of statistical theory and techniques” abroad. The amount was small, hardly sufficient for one year. Frisch hoped to get him a Rockefeller fellowship also.

While attending the Third Cowles Commission Research Conference in July 1937 Frisch gave some thought to Haavelmo's plan for studying abroad. His advice was to “read a considerable amount of mathematics before leaving Norway” and then to concentrate in the first part of the stay on mathematical statistics by going back to work with Neyman in London (“Stay there as long as you think is necessary in order to get a good foundation in sampling theory”).⁷⁹

In the autumn of 1937 Tinbergen visited the institute again. Tinbergen was learning to speak Norwegian; he had a gift for languages and also an interest in being updated with work notes on the progress in the laboratory. The topic for discussion would naturally be Tinbergen's project at the League of Nations, which had progressed much since Oxford. Tinbergen was the world's foremost macroeconomic model builder (in fact, he was still practically the only one) and used bunch maps intensively in the investigation.⁸¹

A pertinent question with regard to Tinbergen's work, which prior to the League of Nations project had comprised a model of the Dutch economy, is why it did not ignite any interest in model building from Frisch's side. The macrodynamic model figured, indeed, prominently in Frisch's “ideal programme” as presented in Oxford, but it seemed as if Frisch was not interested in the model for its own sake but only in the transmission mechanism that explained the properties of data generated from the economy represented as a macrodynamic system of equations and random shocks. Even the propagation and impulse model, often celebrated as the first macro model, can be read as if Frisch's point is really only the exemplification it gives of the transmission mechanism.⁸³

Although the Rockefeller fellowship and a visit to the United States still were pie in the sky, Haavelmo intended to use his Norwegian grant to visit several institutions in Europe. Tinbergen in Geneva was the prime target. He did not heed Frisch's advice about going back to Neyman, but he included a visit to Marschak in Oxford in his plan.⁸⁴

Haavelmo's first stop was Berlin, where he spent a month and a half (including Christmas and New Year's) as a visitor at the well-known Institut für Konjunkturforschung, directed by Ernst Wagemann. Part of the purpose of the Berlin visit seems to have been to study the equipment and report back to Frisch. Haavelmo spent much of the time in Berlin at the Meteorologisches Institut's department for Periodenforschung (i.e., time series analysis), directed by Professor Karl Stumpff with a range of advanced equipment for harmonic analysis. Haavelmo was given access to Stumpff's equipment and tried it out on data already analyzed in Oslo by Frisch's methods to compare the efficiency.⁸⁵

Stumpff's methods for harmonic analysis were described in a note by Haavelmo.⁸⁶

From Berlin Haavelmo traveled to Geneva in mid-January 1938, primarily to work with Tinbergen at the League of Nations' Financial Section.⁸⁹

Haavelmo found Geneva a great place to continue his studies. He was taken care of by Tinbergen, who gave him a thorough introduction to the ongoing work and arranged office space for him with access to the financial section's computer equipment. Meetings were held in Tinbergen's office facing south toward the park of the Palais des Nations, the lake, and on clear days the majesty of Mont Blanc. Haavelmo conveyed the impression that there could hardly be a more ideal place for him to visit, and clearly he regarded Geneva at the time as the leading center in modern economic research.⁹²

He was certainly right about that; no other place in the world had a project that could be of more interest to an econometrician just then than Tinbergen's League of Nations project.⁹³

Being an observer and a participant in this landmark of an econometric project must have left an indelible impression upon Haavelmo. It was also great timing; Tinbergen was close to completion of his first league volume when Haavelmo arrived and was working hard on the second volume in preparation for the special conference to be convened in Cambridge in July 1938 to discuss the results. Tinbergen brought Haavelmo into a little informal group of six or seven “econometricians” he had gathered in Geneva and met with almost daily. In a letter home to Frisch Haavelmo could report that confluence analysis had gained firm ground in Geneva and terms such as bunch map and multicollinearity were used in daily communication without explanation. While he was in Geneva Haavelmo also found time to review for Weltwirtschaftliches Archiv Tinbergen's 1936 publication and Koopmans's dissertation.⁹⁵

Frisch had surely encouraged Haavelmo to submit the Oxford paper to Econometrica after revision. A year and a half after the Oxford meeting it was still not completed. Haavelmo had brought it along to Geneva, and from there he submitted the finished paper to Frisch, acknowledging valuable suggestions from Tinbergen.⁹⁶

From April 1, 1938, Haavelmo was in Paris, apparently without a very specific purpose. Frisch had put him in touch with his old friend François Divisia and also sent with him introduction letters to the mathematician/physicist Phillippe Le Corbeiller and the statistician Georges Darmois. Haavelmo visited the statistical department of Institut H. Poincaré in Paris, where he had the opportunity to follow at close range a major study of production efficiency in French manufacturing. While in Paris Haavelmo wrote a note entitled “The Seasonal Movements Considered as a Periodic Acting Force.” In mathematical terms the idea was to start from a general dynamic equation of a system mobile around an axis with a stable equilibrium and add seasonal movements as a linear function of the state of the system. The economic inspiration was from having noted that the seasonal amplitudes of unemployment in Norway varied inversely with the cycle.⁹⁸

After one month in Paris Haavelmo spent the last couple of weeks of his trip abroad at the Institute of Statistics in Oxford, which had been directed by Jakob Marschak since it was established in 1935. He took part in colloquia organized by Marschak, who also invited him to lecture on confluence analysis. At the time Haavelmo may have considered Marschak primarily a demand and supply analyst. There is no further information about the interaction with Marschak. In retrospect, it could be viewed as an important meeting. Marschak was still only one-third into his 60-year-long remarkable career. The most colorful and dramatic events of his life were behind him. He was certainly one of the brightest and most inventive and enterprising persons in the entire econometric circuit. Frisch had been in close contact with Marschak since 1932 and valued him very highly as reflected in his advice to Haavelmo to spend half a year rather than two weeks with him.¹⁰⁰

While Haavelmo was in Oxford in May 1938 Frisch was asked by the head of the newly established Department of Economics of Aarhus University to find a statistics teacher for the coming academic year. Without thinking twice Frisch offered Haavelmo, who was thus called to a position as a teacher of statistics in Aarhus and accepted without hesitation when he learned about it. We can hardly interpret this otherwise than that Frisch found it a good idea for Haavelmo to be away from Oslo for awhile. Haavelmo might even have expressed an interest in teaching statistics; in London he had made a point of studying the teaching programs in statistics. A sojourn at Aarhus University can hardly have been rated by Frisch as of much interest in itself. It gave Frisch more time to secure Rockefeller support for Haavelmo's visit to the United States, which he surely regarded as a necessity to complete Haavelmo's education.

In the summer of 1938 Frisch recuperated as usual in the mountains, hiking in the daytime, working at night. He had been invited by Alexander Loveday at the League of Nations to take part in the conference convened in Cambridge to discuss Tinbergen's work.¹⁰¹

Aarhus provided a nice break for Haavelmo. The teaching burden was light and gave him ample time for reflecting on the econometric problems he had struggled with for five years. The break came at a convenient time. He had behind him four years of experience with Frisch in the laboratory, the short autumn term with Neyman in London, and his recent Berlin–Geneva–Paris–Oxford tour. He had with him Frisch's 1938 memorandum for the Cambridge conference, which hardly contained anything that was new to Haavelmo. (Did he have the proof version of Tinbergen's two volumes too?) There still was a chance for a one-year visit to the United States. Then time would tell.

At the Department of Economics in Aarhus there were two members of the Econometric Society, the chairman Professor Jørgen Pedersen and Professor Erich Schneider, who was German.¹⁰⁴

Jørgen Pedersen had initiated a series of empirical studies related to Denmark's most important industry, agriculture, and invited Haavelmo to contribute. Haavelmo rose to the challenge and embarked on two econometric studies, one on pig production in Denmark and one on the demand for pork, both completed in the spring of 1939.¹⁰⁷

The pig production paper was not so much about pig production as it was econometrics of regulation (for the first time?). Haavelmo apparently learned a lot about pigs in Denmark. He had determined the structure of pig production, summarized in one of the relations as the outflow of finished pigs per month as a lagged function (by 8.18 months!) of a linear combination of the stocks of first-time-breeding sows and other breeding sows. But what made the paper really interesting derived from the quota regulation, which implied a two-price system for pigs brought to the slaughterhouse, quota price and without-quota price. The quotas were transferable among farmers at market price. Haavelmo was on the frontier in discussing the interaction between production lags and the effects of the regulation with regard to how “shockproof” the system was, concluding that it was indeed not very shockproof. Shocks in the quota price affected the price of small pigs, resulting in shocks reverberating through the production system.¹⁰⁸

8. GETTING READY FOR COLORADO SPRINGS

In the spring of 1939 Haavelmo received confirmation that he had been granted some means from the Norway-America Foundation and thus could plan a departure for the United States. The means would not suffice, however, for more than a few months. At Frisch's insistence Haavelmo drafted an application to the Rockefeller Foundation in which he described his research interest as follows:

Little can be read into the quite generally formulated first sentence with regard to how Haavelmo's thinking had progressed on the issue he had worked on since 1933. The second topic reflected his struggle with the problem that arose from his Oxford paper. The third topic was inspired by Keynes and was an attempt to develop a microeconomic underpinning of the consumption function.

The reaction from the Rockefeller Foundation's Paris office was rather cool. The application was too late; the study plan too vague; and, worst of all, Haavelmo, without a university position, did not fit into the foundation's institution-building policy. Frisch had good standing with the Paris office. He had been consultant to the Paris office on a number of European applicants for Rockefeller fellowships. He now rose to the occasion and did his utmost to convince the foundation officials, expressing his conviction that Haavelmo would have a future at the University of Oslo.¹¹⁰

The outcome was that Haavelmo got a fellowship for one year beginning in 1940. That would allow a total stay in the United States of a year and a half. Haavelmo had no wish to stay any longer. Relatively soon he would also obtain a formal affiliation with the University of Oslo.¹¹¹

The topic for the paper Haavelmo prepared for his presentation at the Fifth Cowles Commission Research Conference in 1939, immediately after his arrival in the United States, was nothing less than the “inversion problem.” Haavelmo dealt with “shock cumulants,” i.e., observations generated by a dynamic model and contaminated by erratic shocks. Classical regression methods would then not give unbiased estimates of structural coefficients. As in his 1938 paper Haavelmo asserted the equivalence between erratic shocks and stochastic coefficients. It was thus a topic chosen from within Frisch's research agenda, but it also reflected the increasing interest Haavelmo had taken in the confrontation between observations and theory.¹¹²

Before leaving Denmark, Haavelmo took part in the Third Nordic Meeting for Younger Economists in May 1939 in Copenhagen, after being asked by colleagues in Oslo to contribute on behalf of the Norwegian association. He chose to present a paper, “On the Statistical Testing of Hypotheses in Economic Theory.” The technical level of the presentation was quite elementary, and the audience was perhaps not exactly erudite in modern statistical theory. Haavelmo aimed nevertheless at presenting a highly sophisticated lecture on verification in economics. The section headings of the paper, which was only 18 pages long, were as follows:¹¹³

1. Introduction.
2. The hypotheses of economic theory are of statistical nature.
3. About the general principles for statistical testing of hypotheses.
4. Free and system bound variations. “Visible” and “invisible” hypotheses.
5. The “ceteris paribus” clause as a statistical problem.
6. The specification problem.
7. The trend problem.
8. The distinction between average explanation and momentaneous explanation.

It was the experienced researcher who presented the modern theory of verification in economics. The opening section set the tone:

He moved on to assert the statistical nature of the hypotheses of economic theory, emphasizing that testing was not an easy task:

He dealt briefly with the principles of statistical testing, not even mentioning Neyman–Pearson, and explained “free” and “system bound,” using formulations similar to those he would later use in the “Autonomy” chapter of the Probability Approach, forewarning that “[t]his is precisely one of the main reasons why refined techniques must receive such a prominent position in modern economic research. Here, there is no use to come with ‘sledgehammer’ methods; we need the statistical technique's finest tools to come to grips with the problems.”

Section 7 on the trend problem touched upon the often-mentioned barrier for probability in economics, that economic time series are not recurrent events to which probability laws apply, but Haavelmo did not bring up probability explicitly. The question of trend elimination, he began, is often conceived as a purely technical-statistical problem but is in reality of far more profound character:

It was not the probability approach; neither was it the occasion for it. Haavelmo's journey had not yet brought him to that stage, but he was getting closer. His experiences and studies since 1933 had advanced his thinking in leaps and bounds and prepared him for further achievement. His carefully phrased and pedagogic sentences in Copenhagen did not fully do justice to his scholarly level; he was after all not addressing econometricians. His presentation would even today serve its purpose as an excellent introduction to the fundamental problems of econometrics. Unlike much of Frisch's work it was imbued with the spirit of the experienced empirical researcher. Less than half a year later he would write home to Frisch about the need for making probability considerations about the deviations between theory and data to decide ultimately whether a theory was “good” or “bad” but still with mixed feelings about the range of applicability of this idea. It was an idea for which he on the eve of his departure was well prepared.

Haavelmo left Denmark in June 1939. His next stop was the Cowles Commission Research Conference at Colorado Springs, where Haavelmo would rejoin Jakob Marschak, who had moved to the United States at the end of 1938, and for the first time meet with Abraham Wald and also with Gerhard Tintner, M. Ezekiel, H.T. Davis, the Working brothers, Charles F. Roos, and others.