Term premia implied by maximum likelihood estimates of affine term structure models are misleading because of small-sample bias. We show that accounting for this bias alters the conclusions about the trend, cycle, and macroeconomic determinants of the term premia estimated in Wright (2011). His term premium estimates are essentially acyclical, and often just parallel the secular trend in long-term interest rates. In contrast, bias-corrected term premia show pronounced countercyclical behavior, consistent with theoretical and empirical arguments about movements in risk premia.