Philipp Külpmann

Assistant Professor, Institute of Economics, Corvinus University of Budapest and
Affiliated Member, Vienna Center for Experimental Economics, University of Vienna

philcontactipp.websitekulpofficemanncontact@uniwebsite-corofficevinucontacts.hu

CV

My research interest lies primarily in Experimental Economics and Game Theory.
Currently, I am organizing the Budapest Experimental and Behavioral Economics (BEBE) Seminar.

Publications

Working papers

  • Aggregate Nash Attainability and Boundary Equilibria under Behavioral Heterogeneity, with Wieland Müller and Alexander K. Wagner.
    Abstract
    We study which Nash equilibria of a finite normal-form game can still arise as aggregate play when each player role is represented by a population containing both behavioral and optimizing agents. In each role, a fixed fraction of agents follows an exogenous action distribution, while the remaining agents best respond to aggregate play. A Nash equilibrium survives if and only if, for every role and every action, the behavioral mass assigned to that action does not exceed its Nash equilibrium probability. When this condition holds, optimizing agents choose the residual distribution and aggregate play reproduces the original Nash equilibrium. When it fails, behavioral types assign too much probability mass to some action, and exact aggregate recovery is impossible. We further show that any additional aggregate equilibria generated by behavioral heterogeneity must lie on the boundary of the aggregate-feasible set and are sustained by behavioral types playing non-optimal actions. Behavioral heterogeneity may thus prune some Nash equilibria and create genuinely new ones, and we characterize both.
    Details & materials
  • When Decoys Fail: Evidence from Video Game Edition Choice, with Fábián Garas and Botond Ujvári and Andreas Orland.
    Abstract
    We study whether classic decoy effects generalize to a consequential digital marketplace setting. In a preregistered, incentivized online field experiment designed to resemble the Steam storefront, 435 participants repeatedly chose between a cheaper base video game edition and a more expensive target edition, with or without a third option constructed as an attraction or compromise decoy. Contrary to standard decoy predictions, neither decoy increased target choice in the full sample. Decoys were themselves chosen at nontrivial rates, and descriptive, non-inferiority, and regression analyses indicate small, null, or opposite-direction effects when decoy choosers are excluded. The results suggest that familiar, information-rich, and consequential platform settings can substantially weaken classical decoy effects. More broadly, the paper shows that choice architecture in digital markets should be tested in the environments in which it is intended to operate.
    Details & materials
  • Playing Prisoner's Dilemma Games with a Large Language Model, with Andreas Orland and Kazuhiro Takemoto.
    Abstract
    Large Language Models (LLMs) are increasingly participating in strategic environments. We provide a behavioral characterization of ChatGPT-3.5-turbo in a one-shot Prisoner’s Dilemma game. Using a full factorial experimental design, we independently vary payoff parameters, the number of interaction partners, the strategy space, and the elicitation order, while eliciting both decisions and stated beliefs. We document three central patterns. First, the model exhibits high baseline cooperation despite pessimistic expectations about others. Second, cooperation is largely insensitive to payoff variation but systematically responds to structural and procedural manipulations such as group size, action space, and elicitation order. Third, the gap between beliefs and actions is condition-dependent. We interpret these findings as a behavioral profile of an aligned LLM operating in a strategic environment. The results suggest that interaction protocols and institutional structure may shape outcomes more strongly than marginal incentive adjustments when such agents are involved.
    Details & materials
  • Coordination without Common Knowledge: Separating First- and Higher-Order Belief Failures, with Davit Khantadze.
    Abstract

    We study coordination when the rules and payoffs of the game are common knowledge, but recognition of the efficient focal action is not. In such environments, a subject who recognizes the focal action may still fail to coordinate on it either because she doubts that her opponent also recognizes it, or because she doubts that this recognition is sufficiently shared. We adapt the five-sector-disc environment of Blume and Gneezy (2010) and add a Prediction task to a Self task and a Coordination task. The self task identifies recognition types. The prediction task identifies whether a recognizer’s first-order belief about the opponent’s type crosses the relevant threshold. The coordination game then reveals whether coordination failure remains once this first-order threshold is crossed.

    The design provides a choice-based identification strategy for separating coordination failure due to low first-order beliefs from failure due to higher-order beliefs, without directly eliciting beliefs. In the data, 75 of 130 subjects are classified as high-recognition types. Among them, 57 choose the distinct sector throughout, 7 fail due to first-order beliefs, and another 7 fail only due to higher-order beliefs. Thus efficient coordination is common even without common knowledge of recognition types, but half of the failures cannot be reduced to first-order pessimism. The results show that higher-order doubt affects coordination behavior in a simple laboratory environment without explicit incentives for recursive reasoning.

    Details & materials

In progress

Contact

Office

Building E, room 239 (2nd floor)

Postal Address

Philipp Külpmann
Institute of Economics
Fővám tér 8
1093 Budapest, Hungary