This paper shows that firms with longer rating agency relationships have better credit ratings, conditional on observables. The paper also finds that (1) controlling for observables, firms with longer relationships, while having higher average ratings, do not have lower default rates, (2) relationship benefits are larger among firms with a greater incentive to game their information supplied to agencies or to pressure agencies into giving higher ratings, and (3) investors demand a (price) discount on bonds sold by relationship firms and the correlation between bond yield spreads and ratings is decreasing with relationship length. In sum, the evidence is inconsistent with first-order credit quality explanations but rather supports a “learning-to-gaming” and an “adverse incentives” story.

Theory predicts rating agencies’ incentive conflicts to be stronger in boom periods, leading to biased ratings and a reduced level of rating quality. We investigate this prediction empirically based on three different approaches. First, we show that initial ratings disagree with bond spread levels during boom periods in the way that rating agencies hold a systematically more optimistic view. Second, we reveal that boom bond ratings tend to be more heavily downgraded from an ex post perspective; and, third, we demonstrate that boom ratings are inflated compared with “conflicts-free” benchmark ratings. In several robustness tests we show that the observed “boom bias” does not result from changes in credit-worthiness, adjustments in rating standards, competitive pressure, or market supply, but rather from rating agencies’ incentive conflicts.