# Philipp Strack, Clark Medalist 2024

*American Economic Association Honors and Awards Committee**April 2024*

Philipp Strack** **is a deeply creative and prolific microeconomic theorist whose research has enriched our understanding of economics on several fronts. Among his many accomplishments are (1) contributions to studies of decision-making and behavioral economics, particularly in dynamic context settings that involve learning and information transmission; (2) the formalization of ideas, such as information cost functions or notions of privacy, so that these concepts can be more fruitfully applied; and (3) the development of a new analytical approach to mechanism design. His work on (1) has challenged conventional wisdom and significantly pushed the boundaries of theories of decision-making and behavioral economics outward, and his work on (2) and (3) represent a new wave of the economics of information.

*Studies of decision-making and behavioral economics*

Strack has advanced the next generation of behavioral economics by examining the limits of behavioral theories and by showing the care needed to interpret and measure empirical facts. For example, Strack (with Heidhues, *American Economic Review *2021) shows that time-*consistent *agents may anticipate future choices that are different from what they ultimately choose. Similarly, convergence of beliefs is a classic test for learning about a true state. Strack (with Heidhues and Kőszegi, *Econometrica *2018) shows that when an agent overestimates their own ability, learning can lead beliefs to diverge further from the truth, and that this effect is exacerbated when the loss from suboptimal actions is large. The argument can be seen in a team context, in which overconfidence leads to the blaming of others for suboptimal outcomes. This core idea is extended in unpublished work with Heidhues and Kőszegi, in which Strack explores how overconfidence and learning can generate discrimination and prejudice, suggesting that the disparate notions of overconfidence and prejudice should be approached jointly in many contexts.

Strack’s rich extension of the canonical drift-diffusion model (with Fudenberg and Strzalecki, *American Economic Review *2018) establishes a new benchmark in economics, psychology, and neuroscience for exploring the timing of choices. The extension incorporates uncertainty about payoffs and accounts for the selection of observed outcomes. This deeper exploration also gives rise to new statistical tests of the drift-diffusion model (with Fudenberg, Newey, and Strzalecki, *Proceedings of the National Academy of Sciences *2020). These, and his other contributions, have gone far in building bridges between theory and behavioral economics as well as between theory and empirical work.

*New frameworks for modeling information and related ideas*

Strack’s approach to formalizing ideas is beautifully illustrated by his axiomatic work on informational cost functions. Arrow (1985) observed that “there are costs of information, and it’s an important and incompletely explored part of decision theory in general to formulate reasonable cost functions for information structures.” Strack (with Pomatto and Tamuz, *American Economic Review *2023) provides an axiomatization of cost structures that capture the notion of a “constant marginal cost of information.” The underlying axioms for such a setting are natural: that more precise information is more costly, and that the cost of collecting *n *random samples is linear in *n, *as in Wald (1945) and Arrow, Blackwell, and Girshick (1949). With these conditions, the paper provides a cost function that is an alternative to the entropy-based cost used in the rational inattention literature (Sims 2003). Relating information costs to economic properties is essential to study information choice and for those who want to understand the burgeoning market for data. This recent paper is poised to have a major impact on theory in information economics, as well as more applied areas such as monetary economics and finance, where researchers have long relied upon entropy-based cost formulations, despite their well-known limitations.

His work on privacy is a second example of his approach to formalizing loosely held concepts. We would often like to be sure that certain policy or business decisions “preserve privacy,” which could mean that the collection of certain information is prohibited or that those decisions do not discriminate across individuals based on protected characteristics such as race, gender, or age, or even potentially complex combinations of such characteristics. There is now much discussion on the potential for algorithms and AI to amplify or reduce “bias” whatever that might precisely mean. Strack (with Yang, *Econometrica *R&R 2024) proposes and characterizes a notion of privacy-preservation for information structures, asking that certain aspects of the state of the world must be kept private, or that decisions must be taken independent of certain aspects of that state. The idea turns out to be captured by the requirement that *no posterior update *must occur on the protected sets. Strack characterizes information structures that comply with this requirement and shows how a decision maker can pre-process data to make it independent of protected characteristics so that any algorithm applied to this data must produce predictions free from racial or gender bias.

*New analytical approaches in mechanism design*

Strack’s pathbreaking work on extreme points and majorization (with Kleiner and Moldovanu, *Econometrica *2021) unlocks new doors for the theory of mechanism design. The literature following Myerson’s 1981 Nobel Prize-winning paper has largely applied his seminal ideas, leaving genuinely new frontiers relatively unexplored. Strack shows that many existing design problems and several new problems can be profitably viewed as an optimization problem over a set of monotone functions, where each element of the set *majorizes *a given function (or is majorized by some given function).

Majorization is a comparison of how “variable” one function is relative to another, which originally found its way into economics via the study of economic inequality or decision-making under risk. Notions such as mean-preserving spreads, Lorenz domination, or second-order stochastic dominance are all closely related. To explicitly view majorization as a *constraint *on an optimization problem is a brilliant idea that seems natural after the fact but leads to new insights. For instance, the problem of Bayesian persuasion can be framed as a choice of posterior distributions (via signal structures) that the sender can impose on a receiver, but each such distribution must be a garbling of the receiver’s prior, or in other words, must majorize the prior. The sender then optimizes some objective function defined on posterior mean, so this is an instance of the problem Strack studies. But there are many others: among them, equivalence results in mechanism design, Myerson’s original design problem, optimal delegation, or design problems for agents with non-expected utility. Strack and co-authors provide general theorems that characterize the extreme points of feasible sets constrained in this way. The reason this is so useful is that optimization problems on such sets that involve linear or convex objectives will find their solutions among those extreme points.

These techniques apply more generally to new design questions. For instance, Strack (with Gershkov, Moldovanu, and Zhang, *Journal of Political Economy, *2020) studies optimal auctions when bidders can make pre-auction investments that could affect their value. This is a new problem in which bidder valuations are effectively nonlinear in win probabilities, thereby precipitating a non-expected utility structure on reduced-form payoffs and rendering a direct application of Myerson’s techniques impossible. Strack’s majorization approach provides the right degree of generalization to attack this problem successfully.

The non-expected utility structure of the previous example, in turn, has intrinsic value. Consider the design of optimal insurance policies. It is well known that expected utility requires uncomfortably large parameters of risk aversion to rationalize widely observed individual choices, such as paying relatively large premia to avoid relatively small deductibles. Theories predicated on non-linear transformations of probability weighting promise to do a better job of reconciling such observations with real-life behavior. Strack (with Gershkov, Moldovanu, and Zhang, *American Economic Review*, 2023) solves the optimal insurance design problem for agents with non-expected utility agents. Along the way, the theory generates a set of new predictions (such as capped coverage) that are not obtained in standard models.

Philipp Strack's mix of creativity, technical skills, and incredible productivity has enabled him to solve long-standing open questions in economic theory while building bridges to a wide range of adjacent disciplines, such as psychology, neuroscience, and computer science. This citation only seeks to illustrate part of the depth and range of his contributions, which bring to mind the period of the 1970s and 1980s when economic theorists sought to definitively explore the new and exciting implications of game-theoretic reasoning. The 2024 John Bates Clark Medal is awarded to Philipp Strack in recognition of these impressive achievements.