Oskar Morgenstern: The Economist Who Forced Mathematics to Look at Strategy

The Problem That Classical Economics Couldn’t See

There is a peculiar blindness at the heart of classical economic theory, one so foundational that it took decades before anyone adequately named it. The standard model of rational economic behavior — traceable through Marshall and Walras and back to Adam Smith’s invisible hand — imagines agents optimizing against a market that is, in a crucial sense, indifferent to them. Prices respond to aggregate supply and demand, not to the particular choices of any individual actor. The corn market does not care whether you buy or sell. The mathematical framework this generates is clean, elegant, and deeply misleading the moment you introduce a small number of strategically aware players who know that their choices will affect each other’s outcomes.

Oskar Morgenstern understood this problem viscerally, and he understood it as an economist rather than a mathematician. Born in 1902 in Görlitz and trained in Vienna during the crackling intellectual atmosphere of the interwar years, he spent the 1930s increasingly troubled by a paradox he would later formalize through a deliciously literary example: Sherlock Holmes and Professor Moriarty fleeing across Europe by train, each trying to anticipate the other’s destination. Each agent’s optimal choice depends entirely on what the other agent chooses — and each agent knows this — producing a recursive loop of anticipation that classical optimization theory simply dissolves into incoherence. This is not a technical failure; it is a conceptual failure. Optimization assumes you are maximizing against a fixed landscape. Strategy is about maximizing against a landscape that is itself strategizing.

The Collaboration That Changed Everything

When Morgenstern arrived at Princeton in 1938 as an émigré, he found himself in intellectual proximity to John von Neumann, who had already, in 1928, published a proof of the minimax theorem — showing that in two-person zero-sum games, there always exists an equilibrium strategy expressible in mixed (probabilistic) terms. The theorem was brilliant but had languished, underappreciated because its context — parlor game theory — seemed too thin to carry serious intellectual weight. Morgenstern recognized that this mathematical result was exactly the skeleton that his economic paradox needed. What followed was one of the great productive collaborations of the twentieth century: six years of work culminating in Theory of Games and Economic Behavior in 1944, a book so dense and so unusual that it arrived on the intellectual scene like an object fallen from another discipline entirely.

The book’s first major conceptual achievement is the formalization of the utility problem. Before you can talk about rational strategy, you need to know what agents are maximizing, and Morgenstern pushed hard to get this right in a way that economists had been sloppy about. The von Neumann-Morgenstern utility theorem establishes conditions — axioms of rational preference, including the crucial independence axiom — under which an agent’s preferences over lotteries (probabilistic outcomes) can be represented by an expected utility function. This is not a trivial result. It gives you a cardinal utility measure, not merely ordinal rankings, derived entirely from behavioral consistency requirements. Every subsequent debate about rationality in economics and decision theory — from Allais’s paradox to Kahneman and Tversky’s prospect theory — is in direct dialogue with this axiomatic foundation.

Games, Coalitions, and the Nature of Social Interaction

The second major contribution is the actual theory of games, and here the book takes a turn that is historically underappreciated. Von Neumann and Morgenstern were far more interested in n-person cooperative game theory than in the two-person zero-sum case that would dominate later work. The concept of the characteristic function — describing what each possible coalition of players can guarantee itself regardless of what the others do — opened a way of thinking about social and economic organization as fundamentally coalitional. The stable set (sometimes called the V-solution or von Neumann-Morgenstern solution), their answer to which outcomes might persist in a game, is not a single point but a set of imputations satisfying internal stability (no outcome in the set dominates another within it) and external stability (every outcome outside it is dominated by something inside it). It is a strange, pluralistic equilibrium concept — acknowledging that multiple stable social arrangements might coexist — and it remains philosophically richer, if mathematically harder, than the Nash equilibrium that would later become the field’s canonical answer.

John Nash’s 1950 equilibrium concept — extending equilibria to non-cooperative games in a formally simpler way — effectively redirected the field. Most working game theorists today work in Nash’s framework. But Morgenstern’s intuition that coalition formation and cooperative bargaining are central to economic and political life has never gone away. It surfaces in mechanism design, in social choice theory, in the economics of clubs and firms, in the analysis of international alliances. The cooperative tradition he and von Neumann founded is experiencing a modest renaissance in algorithmic game theory and multi-agent systems, where computational constraints make coalition structures centrally important again.

What Remains Genuinely Open

The von Neumann-Morgenstern utility axioms remain contested in a productive way. The independence axiom — that if you prefer lottery A to lottery B, you should still prefer A mixed with some third lottery C to B mixed with the same C — seems compelling as a rationality requirement until you sit with Allais’s counterexamples, which show that actual human preferences systematically violate it in ways that feel intuitively defensible rather than simply erroneous. Whether this represents a flaw in human reasoning or a flaw in the axioms is not a settled question. Morgenstern himself was always skeptical of over-mathematizing economics at the expense of empirical adequacy, a tension he held somewhat uneasily given the hyper-formal project he co-authored.

There is also the deeper question of what game theory is, fundamentally, a theory of. Is it a normative account of how rational agents should behave? A descriptive model of how they do? A framework for mechanism designers who want to build institutions producing desired outcomes? Morgenstern’s own sensibility was closest to the descriptive and the institutional — he wanted economics to be more like physics, disciplined by measurement and honest about its units — and the gap between game theory as normative logic and game theory as social science remains very much unresolved.

Why This Matters to the Technically-Minded Generalist

What I find most compelling about Morgenstern’s legacy is the way it forces a confrontation between mathematics and the irreducibly social. The logical structure of strategic interaction — the regress of I think that you think that I think — is not just an economic problem. It appears in evolutionary biology as frequency-dependent selection, in computer science as adversarial machine learning and multi-agent reinforcement learning, in philosophy of mind as theories of other-directed rationality. The formal tools Morgenstern helped build are now part of the infrastructure of almost every field that takes seriously the behavior of multiple interacting intelligent systems. That the tools originated partly in one economist’s discomfort with a thought experiment involving a detective and a criminal on a train is, I think, one of the more charming facts in the history of ideas. The best formal theories usually start as someone refusing to look away from an awkward, specific, nagging problem.