11 October 1989
THIS YEAR'S LAUREATE IN ECONOMICS SHOWED
HOW ECONOMIC THEORIES CAN BE TESTED
The Royal Swedish Academy of Sciences has decided to award the 1989 Alfred Nobel Memorial Prize in Economic Sciences to
Professor Trygve Haavelmo, Oslo, Norway,
for his clarification of the probability theory foundations of econometrics and his analyses of simultaneous economic structures.
This year's prize in economic sciences is awarded to Trygve Haavelmo for his fundamental contributions to econometrics.
During the 1930s, noteworthy attempts were made to test economic theories empirically. The results of these attempts called attention to two fundamental problems associated with the possibility of testing economic theories. First, economic relations often refer to large aggregates of individuals or firms. Theories regarding such relations can never be expected to conform fully with available data, even in the absence of measurement errors. The difficult question then is to determine what should be considered sufficiently good, or better conformity. Second, economists can seldom or never carry out controlled experiments in the same way as natural scientists. Available observations of market outcomes, etc., are results of a multitude of different behavior and relations which have mutually interacting effects. This gives rise to interdependence problems, i.e., difficulties in using observed data to identify, estimate, and test the underlying relations in an unambiguous way.
In his dissertation from 1941 and a number of subsequent studies, Trygve Haavelmo was able to show convincingly that both fundamental problems could be solved if economic theories were formulated in probabilistic terms. Methods used in mathematical statistics could then be applied to draw stringent conclusions about underlying relations from the random sample of empirical observations. Haavelrno demonstrated how these methods can be utilized to estimate and test economic theories and use them in forecasting. He also showed that misleading interpretations of individual relations due to interdependence cannot be avoided unless all relations in a theoretical model are estimated simultaneously.
Haavelmo's doctoral thesis had a swift and pathbreaking influence on the development of econometrics. His probability theory research prograrn attracted a number of outstanding economists - among them, subsequent Nobel laureates such as Koopmans and Klein. This gave rise to extraordinarily rapid methodological development, primarily during the 1940s. The foundation of modern econometric methods had thus been established
The Probability Theory Revolution in Econometrics
Econometric research has been carried out since the beginning of this century. Initially, U.S. economists such as Moore and Schultz worked on econometric determination of supply and demand on individual markets. During the 1930s, Tinbergen and Haavelmo's own teacher, Ragnar Frisch, made the first attempts to apply corresponding methods to test various macrodynamic relations. These estimates touched on several problems which Haavelmo later analyzed in his dissertation.
Prior to Haavelmo's thesis, researchers lacked a common conceptual system for formulating, analyzing and solving econometric problems. At the time, few econometric methods were based on probability theory and therefore could not utilize statistical inference to draw conclusions from data. To the extent that calculations contained any random variations at all, they usually referred to measurement errors in the variables. Simple statistical methods - mainly regression analysis - were used in most instances, without any clear probability theory assumptions whatsoever. During this period, most prominent economists including Keynes - rejected more extensive use of probability theory in empirical research on the grounds that, e.g., economic processes were irreversible. Many of the leading econometricians of the day - such as Frisch - were also skeptical about the possibility of applying statistical inference methods to economic data.
In his dissertation, Haavelmo refuted these various objections and showed that in order for economic theories to be testable, probability theory formulation is not only a prerequisite, but also extremely reasonable. Economists analyze results of millions of decisions made by individuals and firms. According to Haavelmo, it is unreasonable to believe that economists could ever "fully" explain or predict such individual decisions on the basis of necessarily simplified assumptions. Decisions are in fact affected by individual characteristics and numerous temporary conditions which change over time. Therefore, economists' explanations of decisions always have to encompass a stochastic term which summarizes these different kinds of "disturbances". As economic theories in general do not refer to individual decisions but are concerned with relations which comprise long sequences of decisions and a multitude of decision-makers, there are frequent opportunities to make relatively simple assumptions about the probability distribution of these aggregate relations.
Haavelmo also demonstrated that by formulating theories in probability theory terms, statistical inference methods could be applied to estimate and test economic theories and use them in forecasting. Most of the problems he dealt with and analyzed are associated with interdependence in economic relations.
In economic life, every individual decision may be regarded as affecting all other decisions through a chain of market relations. This economic interdependence creates problems in empirical research because an observed market outcome is the result of a large number of simultaneous or previous decisions and behavioral relations. Thus, an underlying relation can never be observed, as it were, in isolation, but only as conditioned by a number of other simultaneous relations and circumstances in the economy. As Haavelmo showed, interdependence gives rise to difficulties in specifying, identifying and estimating economic relations.
The difficulty of specifying economic explanatory models, lies in choosing among numerous models or systems of relations which might explain the observed market outcome. When the relations in the model are interdependent, then a set of model equations can be used to derive a multitude of other equation systems which produce the same observable result. Haavelmo emphasized the importance of trying to choose a set of relations which are each as "autonomous" as possible, i.e., which are not affected by changes in other parts of the system. For example, in order to determine the effect of a decrease in household incomes, due to changes in fiscal policy, on private consumption, then obviously the estimate of the propensity to consume used in the computation should not be conditioned by previous fiscal policy. The choice of autonomous relations in explanatory models is primarily a matter of adequate knowledge and intuition regarding the basic mechanisms of the economy. However, Haavelmo also discussed the need for statistical invariance tests and not too long ago, researchers succeeded in developing a method whereby the autonomy in different relations can be tested statistically.
The fact that many different types and forms of explanatory models can explain observed data, also gives rise to an identification problem. For example, if a theory is intended to explain the observed relations between price and sales on a market, the relations have to be sufficiently specified so as to be identifiable as demand and supply relations, respectively, with some given form of probability distribution. Haavelmo was the first to provide an explicit mathematical formulation and solution of the identification problem. Further development of identification criteria has been based on his formulation.
Interdependence also creates what Haavelmo called simultaneity problems in estimating models with several different structural relations. Since the combined relations limit the possible variations in the input variables, isolated estimates of individual relations can be highly misleading. Using a probability theory framework, Haavelmo provided a generally valid formulation and method of measuring this bias in isolated estimates of individual relations in an interdependent system. He also showed that the problem could be avoided by using a method of simultaneous estimates of interdependent models. Haavelmo's analysis of simultaneity problems has had a decisive influence on later work with econometric models.
From Econometrics to Economic Theory
Once the foundation of probabilistic econometrics had been established, Haavelmo's next important research effort involved attempts to transform various components of economic theory so that the new econometric methods would be applicable. According to Haavelmo, the prerequisites for achieving this purpose were not only additional assumptions about probability distributions, but also in many instances a more dynamic theoretical formulation. There are two areas in particular - investment theory and economic development theory - where Haavelmo's approach has resulted in influential and far-reaching contributions. In addition to these main lines of research, Haavelmo's achievements include valuable contributions in numerous areas - from analysis of macroeconomic fluctuations and fiscal policy to price theory and the history of economic thought.
Haavelmo's most influential study is his doctoral dissertation entitled, The Probability Approach in Econometrics , presented at Harvard University in April 1941, although not published until 1944 as a supplement to Econometrica. The argument in his thesis was later extended and exemplified in numerous publications, among which may be mentioned two articles in Econometrica: "The Statistical Implications of a System of Simultaneous Equations" (1943) and "Statistical Analysis of the Demand for Food" (1947, co-authored by M.A. Girshick).
Haavelmo's early research on economic development theory is summarized in his book entitled, A Study in the Theory of Economic Evolution (1954). A corresponding resume of Haavelmo's contribution to investment theory is given in A Study in the Theory of Investment (1960).