Speculative Attacks and Financial Architecture: Experimental Analysis of Coordination Games with Public and Private Information

Speculative Attacks are a coordination game with multiple equilibria: If a sufficient number of traders expect devaluation, market pressure forces the central bank to abandon a currency peg that it would have kept otherwise. Thus, expectations are self-fulfilling. Applying the global games approach, Morris and Shin (1998) have shown that there is a unique equilibrium in threshold strategies, if traders have private information on the state of the economy. Uniqueness requires private information to be sufficiently precise when compared to public information. This has triggered a discussion on the optimal policy to release information to financial markets in order to prevent crises with self-fulfilling features. Our experimental design imitates the speculative attacks model by Morris and Shin. We compare sessions with public and private information and test the theory of global games and the payoff dominant equilibrium as a refinement criterion. Independent from the information set-up, most subjects use undominated threshold strategies after the first round. With public information, subjects better coordinate and use significantly lower thresholds than with private information. Dispersion of average threshold across sessions does not depend on the information structure. Reactions to parameter changes follow comparative statics of the private information equilibrium. We conclude that a commitment to provide public information increases prior probability of successful speculative attacks. Furthermore, the destabilizing effect of public information due to self-fulfilling beliefs may be less severe than theory predicts. * We want to thank Hyun Shin for stimulating comments, Isabel Gödde and various colleges at Goethe-Universität Frankfurt for helpful discussions about statistical procedures and tests. All errors are at our responsibility. We are indebted to Urs Fischbacher for his fabulous software package z-Tree that we used for our experiments. Finally, we thank students at Goethe-Universität Frankfurt and Universitat Pompeu Fabra in Barcelona for their participation in our experiments. Financial support by the Volkswagen Foundation and the Interessengemeinschaft Frankfurter Kreditinstitute is gratefully acknowledged. Frank Heinemann acknowledges financial support by the TMR network on “Financial Market Efficiency and Economic Efficiency”. a Goethe-Universität Frankfurt, b Universitat Pompeu Fabra, Barcelona c Goethe-Universität Frankfurt,