From CFAI 

Pt,s=Et[˜Pt+1,s−1]/(1+lt,1)+covt[˜Pt+1,s−1,˜mt,1] pg 388

“ For example, the correlation between 2-year bond returns in Canada and Canadian economic growth is −30.95%, but the equivalent correlation for 30-year Canadian government bonds is −12.50%. These facts indicate that short-dated bonds have been more reliable hedges against bad economic times than long-dated bonds, which, in turn, means that the bond risk premium should be higher for long-dated government bonds than for their short-dated equivalents.” Pg  420  

If covariance is more for short term bond than long term bond,then wouldn’t Pt,s be higher for longer term bond and hence the risk, which is a dividing factor will be a lower term. What am I missing?