We introduce a simple game-theoretical model that captures the main aspects of the repeated interaction between an issuer and a credit-rating agency. It involves up-front payments of issuer-fees and direct publication of requested ratings. Due to pecuniary injuries for untruthful ratings, the credit-rating agency should always report truthfully in the subgame perfect equilibrium. Knowing this, the issuer should never request a rating. Conducting laboratory experiments, we find that behavior significantly deviates from the equilibrium prediction in favor of a cooperative solution: issuers frequently do request ratings, which is often reciprocated with untruthful good ratings.