The failure of consumption-based asset pricing models to match thestochastic properties of the equity premium and the risk-free ratehas been attributed by some authors to frictions, transaction costs,or durability. However, such frictions primarily would affect thehigher-frequency data components: Consumption-based pricing modelsthat concentrate on long-horizon returns should be more successful.We consider two consumption-based models: time-separable utility, andthe habit model of Constantinides. We estimate a vector ARCH modelthat includes the pricing kernel and the equity return, and use thefitted model to assess the model's implications for the equitypremium and for the risk-free rate. Neither model performs well at aquarterly horizon, but at longer horizons the Constantinides modelcan match the mean and the variance of the observed equity premium,captures time variation of the equity premium, and can better matchthe observed risk-free rate. We conclude that the equity-premium andrisk-free-rate puzzles are primarily problems for shorter-horizonreturns.