David Schmeidler (1939 – 17 March 2022) was an Israeli mathematician and economic theorist. He was a Professor Emeritus at Tel Aviv University and the Ohio State University.
Schmeidler's early contributions were in game theory and general equilibrium theory. He suggested a new approach to solving cooperative games – the nucleolus – based on equity as well as feasibility considerations. This concept, originating from Schmeidler's PhD dissertation, was used to resolve a 2000 years old problem. Robert Aumann and Michael Maschler, in a paper published in 1985, showed that a conundrum from the Babylonian Talmud, which defied scholars’ attempts at comprehension over two millennia, was naturally resolved when applying the concept of the nucleolus.[1]
Schmeidler also pioneered the study of non-atomic strategic games,[2] in which each player has negligible impact on the play of the game, as well as the related concept of “congestion games”, where a player's payoff only depends on the distribution of the other players’ strategic choices (and not on individual choices).
Schmeidler has made many other contributions, ranging from conceptual issues in implementation theory, to mathematical results in measure theory. But his most influential contribution is probably in decision theory. Schmeidler was the first to propose a general-purpose, axiomatically-based decision theoretic model that deviated from the Bayesian dictum, according to which any uncertainty can and should be quantified by probabilities. He suggested and axiomatized Choquet Expected Utility,[3][4] according to which uncertainty is modeled by a capacity (not-necessarily-additive set function) and expectation is computed by the Choquet integral.
While this approach can be used to explain commonly observed behavior in Ellsberg's experiments, Schmeidler's motivation was not to explain psychological findings. Rather, along the lines attributed to Frank Knight and John Maynard Keynes, the argument is normative, suggesting that it is not necessarily more rational to be Bayesian than not.[5] While in the experiments, drawing balls from urns, one may adopt a probabilistic belief, in real life one often couldn't find a natural candidate for one's beliefs.[6]
