Credit rating agencies face a difficult trade-off between delivering both accurate and stable ratings. In particular, its users have consistently expressed a preference for rating stability, driven by the transactions costs induced by trading when ratings change frequently. Rating agencies generally assign ratings on a through-the-cycle basis whereas banks' internal valuations are often based on a point-in-time performance, that is they are related to the current value of the rated entity's or instrument's underlying assets. This paper compares the two approaches and assesses their impact on rating stability and accuracy. We find that while through-the-cycle ratings are initially more stable, they are prone to rating cliff effects and also suffer from inferior performance in predicting future defaults. This is because they are typically smooth and delay rating changes. Using a through-the-crisis methodology that uses a more stringent stress test goes halfway toward mitigating cliff effects, but is still prone to discretionary rating change delays.