•Strong behavioral responses to a pension reform lowering the FRA by up to 2 years.•Bunching response at the FRA exceeds the control group’s bunching by 83%.•A 1.0-year decrease in the FRA reduces the pension claiming age by 0.3-0.4 years.•Fiscal costs of the reform higher than expected.•Responses driven by reference point effect of new FRA and by financial incentives. This paper analyzes behavioral responses to a 2014 reform in the German public pension system that lowered the full retirement age (FRA) of individuals with a long contribution history by up to two years and framed the new FRA as reference age for retirement. Using administrative data from public pension insurance accounts, we first document a substantial bunching response at the FRA exceeding the control group’s bunching by 83%. Second, we show in a difference-in-difference setting that a 1.0 year decrease in the FRA leads to a reduction in the average pension claiming age by 0.3–0.4 years. Treated individuals neither have poorer health nor are more likely to be liquidity-constrained than individuals in the control group. Our results suggest that the strong responses to the reform are driven both by the new FRA serving as a reference point and by financial incentives. Estimated fiscal costs of the reform are at the upper end of the range of previous back-of-the-envelope calculations.