Based on narrative identification, we construct a novel comprehensive dataset of pension reform measures in OECD countries from 1970 to 2017. We then study the timing of these measures. Our main and new result is that business cycle indicators are important for their timing: a worsening makes contractionary measures more likely and expansionary measures less likely. The demography matters only in the sense that the OECD-wide demography explains the general reform trend for a country. We find no evidence that country-specific or short-run demographic developments matter. We discuss a conceptual framework with adjustment costs of changing pension generosity that can account for both the reform responsiveness to the business cycle and the lack of responsiveness to changes in demographic forecasts. We also discuss potential policy implications of our findings.