This paper develops an equilibrium model of the annuities market where agents have private information about their mortality, and where the predictive value of this information decays over time. The paper shows that in this case, insurance companies will observe a duration-related trend in the mortality of annuitants under certain conditions. This effect is tested for using a Cox proportional hazards methodology and data from the South African annuities market, which since the early 1990’s has permitted phased withdrawals of retirement savings instead of mandating pure annuitisation. Evidence is equivocal: substantial differences are found between the duration-related mortality trends of different insurance companies, data problems seem to have some effect, and factors outside the model which might change the results cannot be excluded. However, the presence of a strong duration-related trend cannot be decisively rejected. The observed trend indicates that mortality at earlier policy durations is better than at later durations by the equivalent of about 6 years of age, although data factors cannot be precluded as a cause of this trend.