Charles Goodhart and Luca Benati have a new paper on the yield curve where they "have investigated why the yield curve has appeared to have had such predictive power for future output growth at times in the past, but also why this may now have largely disappeared. [They] examine the relationship... for the US and the UK since the Gold Standard era, and for the Eurozone, Canada and Australia in the Post-WWII period... [their] results suggest that, historically, the additional predictive power of the spread for future output growth –over and above that already encoded in other macroeconomic variables – often appeared during periods of uncertainty about the underlying monetary regime."
So the yield curve's predictive power is contingent on the uncertainty of the underlying monetary regime. This finding is consistent with what Michael Bordo and Joseph Haubrich found in their paper, The Yield Curve, Recessions and the Credibility of the Monetary Regime: Long Run Evidence 1875-1997 (with Joseph G Haubrich) NBER Working Paper No10431, 2004. Here is the abstract:
"This paper brings historical evidence to bear on the stylized fact that the yield curve predicts future growth. The spread between corporate bonds and commercial paper reliably predicts future growth over the period 1875-1997. This predictability varies over time, however, particularly across different monetary regimes. In accord with our proposed theory, regimes with low credibility (high persistence of inflation) tend to have better predictability."