Development of empirical methods in modern economics
Another topic in the history of 20th century economics is the development of econometrics and empirical methods in economics.
Almost all economists (but not all) believe that economics must ultimately be an empirical discipline, that their theories of how the economy works must be related to (or if possible tested against) real world events and data.
However, economists differ on how one does this and what implications can be drawn from testing economic theories.
In testing or relating theories to the evidence economists used and still use different approaches, from simple common sense observation of data, through statistical analysis, to sophisticate econometric testing.
The most important approach remains econometric testing, which in itself is divided into several approaches: we have classical econometric approach, Bayesian econometrics and several approaches that are more recent.
In the second half of the 20th century, the development of computer technology heavily influenced economists' approach to empirical testing. Several modern techniques of empirical investigation are closely connected to the use of computers, for example, VAR modelling, calibration method, simulations and the like.
Since 1960s advances in computer technology made it possible to conduct extremely complicated empirical work, statistical tests that earlier would have taken days (or were simply impossible) now could be done in seconds.
During this days (1960s, 1970s) the hopes for econometrics were high. Some believed that econometrics would make economics a science in which all theories could be tested and rejected if falsified by the evidence.
Unfortunately, most of those hopes, as we know today, have not been realized so far.
Let's start with the history of empirical analysis in modern economics.
Early attempts at empirical work in economics were the exception, not the rule. In the late seventeenth century, most classical economists were satisfied with theoretical work, or in the best case used some kind of common sense empirical work, supporting their theories with examples from the real world.
In the late 19th century, during the heyday of neoclassical economics, this approach was called into question. Neoclassicals wanted economics to become a real, hard, exact science and formalized their theories in a mathematical language. Many of them thought that formalized economic theories should be tested statistically, but neoclassical economists have done not much of this kind of work.
Notable examples include some American economists, who introduced statistical methods in economics early in 20th century.
Henry L. Moore in the first two decades of 20th century used statistical methods to verify some economic theories, like marginal productivity theory. He also tried to estimate demand curves for agricultural goods and empirically measure business cycles. His work was bothered with theoretical and empirical difficulties, but he remains the first economist who empirically measured a demand curve.
From 1920s, we can observe the rise of econometrics as a part of economic science. One of the most important early econometricians was Ragnar Frisch, who was a highly trained mathematician, who contributed to both macro and microeconometrics. He coined the term `econometrics'.
He played an important role in creating a field of macroecometrics by developing a macroeconometric model of the economy in a book published in 1934.
Another person important in the development of econometrics is Trygve Haavelmo, a Norwegian economist, who has been credited with introducing the probabilistic approach to econometrics, in a paper published in 1944.
Before the introduction of the probabilistic approach, economists assumed that the underlying economic variables they were trying to measure were exact. If one could in fact hold everything else constant, one would have measure variables exactly and find exact relationships between variables.
Haavelmo argued against this assumption, contending that we should treat economic theories as probabilistic theories that do not describe exact relationships, but, instead, describe probabilistic relationships.
Acceptance of the use of the probability theory in economics, that followed Haavelmo's work, allowed the formal use of many statistical techniques and tests that previously were used without formal foundation in economics, and it lies at the heart of the modern approach to econometrics. In recognition of this Haavelmo was awarded the Nobel Prize in economics in 1989.
Haavelmo's probabilistic approach was accepted by researchers in the The Cowles Commission for Research in Economics at the University of Chicago, US, (from 1955 associated with Yale University), a centre founded in 1932 for the advancement of economic theory in its relation to mathematics and statistics. The Cowles commission contributed much to the development of econometrics and mathematical economics and it is considered the most important organization for the history of American economics (because it contributed so much to the mathematization of economics and the rise of econometrics).
The cowls commission did much of what is now considered standard econometric work. This work included estimating whether the ordinary least square estimator would be biased or not, developing The Monte Carlo approach to small data sets and working on issues of asymptotic convergence und unbiasedness of estimators.
During this time, 1940s, is should be remembered that computational difficulties were enormous, because the computer as we currently know it, did not exist. Those early econometricians had to perform the calculations manually.
The Cowles Commission followed Haavelmo in assuming that the best approach to econometrics was the probability approach, in which the structural equations of the model had an assumed distribution of error terms. This probabilistic approach became known as the Cowles Commission or standard approach to econometrics.
In following years, members of the Cowles Commission formulated several econometric models of American economy, based on probabilistic approach.
One of the most famous of these models was the Klein-Goldberger macroeconometric model, which was the first empirical representation of the Keynesian vision of the economy.
The model contained 63 variables, including 43 exogenous or lagged variables. It was a really large model.
In 1960s, several similar large econometric models of the economy were developed, to serve as predictors of the economy. These macro models remained popular through early 1970s, but in the mid-1970s, this work was losing support.
The main reasons for this were the deficiencies of classical statistical tests in the hands of economists.
The validity of those tests depends upon theory being developed independently of the data. In reality, however, most empirical economic researchers “mine the data”, looking for the best fit, that is the formulation of the theory that achieves the best R2, t and F statistics.
This procedure, data mining, misuses statistical tests and their results cannot be considered correct.
Another reason for doubting in large macroeconomic models of the economy in the mid 1970s was the limited availability of data - you had to introduce many proxies in models, which were not necessarily very appropriate.
Further, economists started to realize that almost all economic theories include some immeasurable variables, which often were relied upon to explain statistical results that did not conform to the theory.
In general, since the mid 1970, the attitude toward econometrics began to be much more reserved or even sceptical.
However, since the 1980s several new econometric techniques were develop that do not possess disadvantages of the standard (Cowles Commission) approach to econometrics, such as VAR (Vector Autoregression) approach or general to specific modelling. These approaches have their own deficiencies, but we do not have time here to discuss them.
What is interesting here is that those problems with standard econometrics have turned the attention of many empirical economists toward other, non-standard approaches, such as Bayesian econometrics or experimental economics and simulation methods.
Thomas Bayes suggested Bayesian approach, British mathematician, in 18th century. The approach is very different from classical approach to statistics.
It proposed a subjective interpretation of statistics as opposed to an objective interpretation in the classical approach. Bayesians therefore propose dropping traditional classical econometrics.
We do not have time to discuss the approach in detail, especially since for the most part economists have not used Bayesian methods, mainly for practical reasons (for example, it is hard to convince other economists to you subjective assessment of probabilities). Still a number of leading econometricians are seriously committed to this approach.
Few words on experimental economics and simulation methods.
Recently, from 1980s, a group of economists has begun to undertake a different approach to empirical work in economics.
They use animals or people to act as economic agents (buyers or sellers) and conduct experiments to verify whether various economic theories correctly predict results that occur in those experiments.
Experimental economics claims to have proved various economic propositions through their experiments.
The most well known proponent of experimental economics is American economist, Vernon Smith, b. 1927, who was engaged in conducting experiments in economics since 1950s. However, only late 1980s experimental economics became an established part of economic science.
In 2002, Smith was awarded the Nobel Prize in economics "for having established laboratory experiments as a tool in empirical economic analysis, especially in the study of alternative market mechanisms". So it can be considered as an institutional recognition that experimental economics has become an important field in economics.
Smith conducted pioneering economics experiments on the convergence of prices and quantities to their theoretical competitive equilibrium values in experimental markets.
He studied the behaviour of "buyers" and "sellers", who are told how much they "value" a fictional commodity, and then are asked to competitively "bid" or "ask" on these commodities following the rules of various real world market institutions used in many stock exchanges, as well the other models of auctions.
Smith found that in some forms of auctions, prices and quantities traded in such markets converge on the values that would be predicted by the economic theory of perfect competition; despite the well know fact that the many assumptions of this model are not met in the real world (assumptions like large numbers of agents, perfect information and the like).
Other examples of successful economic experiments would include for example experiments concerning the so-called “social preferences” of individuals.
The term "social preferences" refers to the concern that people have for each other's well-being, and it encompasses altruism, taste for equality, and tastes for reciprocity.
Experiments on social preferences generally study economic games, in which players interact in various settings. Some of those experiments have shown that people are generally willing to sacrifice monetary rewards when offered unequal allocations, thus behaving inconsistently with simple models of self-interest. People care about equality in monetary rewards in games, at least when stakes are not very high.
Given that in general problems of empirically testing economic theories are quite serious, experimental economics has gained some importance. It is now a growing subfield of empirical economics; it is very promising.
Experimental economics have opened new perspectives for economists, who want to confirm or verify economic theories.
A related development, similar to experimental economics, is the use of computer simulation. In this work, models are designed that have multiple agents that follow simple, assumed rules of behaviour. Then simulations are run, in which agents interact with each other and it is determined which rules of behaviour survive and which do not.
This allows choosing assumptions (those rules of behaviour) for economic models, which are empirically confirmed to be efficient - agents who live by those rules survive and prosper.
However, this field of simulation is even much younger than experimental economics and its future position in economic science is uncertain.
Summary of the developments in modern empirical economics.
As far as econometrics is concerned, the theoretical progress in econometrics has been enormous. Several sophisticated approaches to conducting econometric study were developed, including many quite new, such as VAR (vector autoregressive) modelling or general to specific approach. It is too early to assess whether those approaches have fulfilled the great hopes and expectations that existed when econometrics were born in 1940s and 1950s.
Yet many modern mainstream economists are disappointed with econometrics ability to explain and predict economic phenomena. For example, we are still not able to predict accurately economic growth, or many other macroeconomic aggregates, in a longer perspective, than one or two years.
So the fair evaluation of the developments in econometrics can be made only in future.
However, there is at least one very interesting and positive thing about the advance of empirical methods in economic in the 20th century - that recently economists have turned their attention to non-econometric methods of empirical verification of their theories, such as experimental economics or computer-based simulations. Those methods for sure enrich economics and make it more empirical and therefore more scientific.
5