We combine experimentally elicited preferences with administrative micro data to study actual financial decision-making. Firstly, we estimate risk and (present-biased) time preferences in a reallife context, with horizons up to 10 years, for a large group of pension fund participants. We estimate a present-bias factor of 0.88, an annual discount rate of 3.91% and a CRRA utility curvature of 0.97. Secondly, using a life-cycle framework, we show that the individually estimated preferences explain actual retirement decisions up to 83% of our sample. Freedom of choice creates annual welfare gains up to 4.8%, but realized welfare gains are lower or even negative.