Oskar Morgenstern, Ökonometrische Berechnungen im Großen (1952)

Morgenstern’s journal article examines how large-scale electronic computation could transform economic theory and econometric research. Its argument moves from the history of equilibrium systems through the methodological relationship between calculation and experiment to five prospective research applications. The central claim is that dramatically increased computing power changes more than the speed of established procedures: it makes previously inaccessible investigations possible. Yet computational capacity does not itself guarantee meaningful results. Adequate data, mathematically justified models, and careful analysis of error remain indispensable.

Morgenstern distinguishes three stages in the development of calculable economic theory. Walras assembled a comprehensive system of variables and equations, but matching the number of equations to the number of unknowns did not establish solvability. Pareto’s example of 70,699 equations for just 100 people exchanging 700 commodities exposed the enormous computational burden. Morgenstern also rejects the reassurance that economic reality solves such equations every day, supposedly rendering numerical solution unnecessary. That claim would require proof that the model uniquely represents reality and that the actual economy possesses an identifiable stability.

Welche Gleichungen die Wirklichkeit tatsächlich löst, kann heute niemand angeben.

English translation: No one today can specify which equations reality actually solves.

This objection separates economic activity from its theoretical representation: the existence of an economy cannot certify the validity of a particular equation system. The second stage therefore concerns proofs of existence. Wald’s demonstration of a positive solution for a modified Walrasian system and von Neumann’s proof for a dynamic system establish a stronger standard. Authors must demonstrate that their systems admit solutions with economic meaning, rather than merely count equations. Morgenstern particularly values von Neumann’s contribution for its dynamic character and mathematical innovation.

The third stage is numerical solution. Leontief and Mitchell’s 1947 calculation of a 38-equation system required approximately forty hours on Harvard’s Mark II; subsequent electronic machines enabled systems of roughly 200 equations. But the progression is not simply toward ever-larger models. Observation errors can propagate through an otherwise correct calculation, while the precision needed depends on both the operations performed and the mathematical methods employed. Economic statistics often display more digits than their reliability warrants. Data collection may consequently remain the principal and most expensive obstacle, and computing costs must also be considered alongside competing scientific demands.

The article’s decisive conceptual turn comes when Morgenstern shifts from computation as an application of existing theory to computation as an instrument for developing theory.

Es besteht nämlich eine fast vollständige Äquivalenz zwischen Berechnung und Experiment.

English translation: There is, in fact, an almost complete equivalence between computation and experiment.

He explains this equivalence through analog computation: a physical experiment embodies a mathematical process, even when that process is not yet understood. A wind tunnel can reveal a critical failure point that an adequate, computationally manageable theory might instead calculate. Economic analog machines, including hydraulic models, make relationships visible but remain tied to relatively rigid built-in models and serve chiefly pedagogical purposes. Digital machines lack that visual immediacy but permit numerical exploration of a theoretical system without imposing the corresponding pressures on the actual economy. Linear programming exemplifies this possibility by identifying optimal programmes and potential bottlenecks under linear constraints.

Morgenstern then challenges the assumption that empirical work must always follow a strong theory. Theoretical guidance is especially valuable when research resources are scarce; greatly enlarged computational capacity creates a different situation. Time savings make extensive exploratory investigations feasible, potentially revealing relationships that established theory has not anticipated. His five proposed applications connect this exploratory ambition with specific mathematical procedures.

Solving large linear systems could support retrospective tests: researchers might introduce observed changes in demand into a model based on an earlier year and compare calculated outcomes with later production figures. Discrepancies could guide model revision before prospective interventions are attempted. This proposal remains qualified by criticism of prevailing Walrasian models, which inadequately accommodate monopoly and reduce economic situations too readily to maximization problems.

Fourier and spectral analysis could investigate rhythms across hundreds of long economic time series, moving beyond the removal of isolated components toward their structural interrelationships. Autocorrelation analysis would examine how current values depend on past values, addressing an empirical weakness in dynamic modelling.

Solange wir keine zahlenmäßigen Ausdrücke hierfür im großen haben, ruhen die meisten dynamischen Modelle der Wirtschaft auf unsicherem Fundament.

English translation: As long as we lack large-scale numerical expressions for this, most dynamic models of the economy rest on an uncertain foundation.

Lag correlations extend the programme to relationships among series. Morgenstern explicitly anticipates spurious correlations, yet suggests treating tens of thousands of coefficients as a statistical population whose own structure might be informative. This is an exploratory proposal, not an assurance that genuine relationships will emerge.

Niemand kann dessen sicher sein oder jetzt das Gegenteil beweisen.

English translation: No one can be certain of this or now prove the contrary.

Finally, computing optimal strategies in game-theoretic problems would bring duopoly and oligopoly into the numerical programme. This complements his criticism of equilibrium models by addressing strategic interaction directly. The article’s enduring methodological significance lies in combining computational optimism with epistemic restraint: electronic calculation can test models, investigate dynamics, and open unexpected research paths, but only researchers’ judgement about data, mathematical validity, and economic interpretation can turn technical capacity into knowledge.