Previous research finds a systematic fall in consumption at retirement, even when these retirements are expected, which implies households do not behave as predicted by the lifecycle/ permanent income hypothesis. However, the worker’s expected date of retirement is typically predicted using an instrument – age – that we show to be correlated with unexpected retirements and will therefore lead to biased estimates. In this paper, we use an alternative instrument for expected retirement: workers’ own subjective beliefs of their expected retirement dates. We find that subjective retirement expectations provide strong predictive power for subsequent retirements above and beyond the impact of age on retirement probabilities. We still find, however, that consumption falls for workers who retire when expected although the estimated impact is 50 percent smaller when using retirement expectations as an instrument instead of age.