Repeated Games Algorithm
This page hosts the source files for the MATLAB implementation of Dilip Abreu and Yuliy Sannikov’s (2011) algorithm for computing equilibrium payoffs of infinitely repeated discounted games with perfect monitoring.
The program has been developed by Dilip Abreu, Ben Brooks, Rohit Lamba, and Yuliy Sannikov since the summer of 2010. It consists of a MATLAB package of routines that implement the algorithm, along with a graphical tool for interactively designing and solving games. The tool can either iterate until convergence, or it can be told to iterate for a fixed number of iterations, depending on whether "solver mode" is set to "converge" or "test," respectively. In test mode, the behavior of the algorithm will be recorded at each action profile and iteration. Here the tool is being used to analyze an example given in Abreu & Sannikov (2011):
The program has been developed by Dilip Abreu, Ben Brooks, Rohit Lamba, and Yuliy Sannikov since the summer of 2010. It consists of a MATLAB package of routines that implement the algorithm, along with a graphical tool for interactively designing and solving games. The tool can either iterate until convergence, or it can be told to iterate for a fixed number of iterations, depending on whether "solver mode" is set to "converge" or "test," respectively. In test mode, the behavior of the algorithm will be recorded at each action profile and iteration. Here the tool is being used to analyze an example given in Abreu & Sannikov (2011):
To use the package, simply download it from the link below, unzip it, open MATLAB, and change the current folder to the folder containing "+rgsolve." To start the graphical interface, enter "rgsolve.rggui" at the command prompt. The zip folder contains a manual with a tutorial.
The following is a demonstration of how the algorithm works, for the example depicted above. With a discount factor of 0.3, the algorithm takes 51 iterations and 0.19 seconds to converge. The image below shows the first four iterations. These would be accessed in test mode. The criterion for convergence is that the extreme points do not move by more than 2^-52.
To download the rgsolve package, click here.
Please contact us via email: