Pré-publication, Document de travail: Retirement is commonly described as a pure labor supply decision, despite the potential importance of the demand side channel. This is partly due to the fact that both dimensions are often di?cult to disentangle. In this paper, I manage to overcome this di?culty by using a unique natural experiment, the progressive ban of mandatory retirement in France in the 2000s. Drawing on an extensive administrative dataset, I use inter-industry reform-induced variations in mandatory retirement legislation, thereby insulating this factor from other determinants of retirement, such as ?nancial incentives. I ?nd that demand-side determinants through mandatory retirement do a?ect retirement patterns: exit rates from employment are estimated to be 6% higher when mandatory retirement is possible. Secondly, as the mandatory retirement age coincides with the full rate age, I exhibit a previously uncovered determinant of the large bunching in retirement distribution at this age. Mandatory retirement is estimated to explain 10% of the observed spike at full rate.