Can consumption-based mechanisms generate positive and time-varying real term premia as we see in the data? I show that only models with time-varying risk aversion or models with high consumption risk can independently produce these patterns. The latter explanation has not been analysed before...