Forecasting output growth by the yield curve: the role of structural breaks

This paper proposes a new structural-break vector autoregressive (VAR) model for predicting real output growth by the nominal yield curve information. We allow for the possibility of both in-sample and out-of-sample breaks in parameter values and use information in historical regimes to make inference on out-of-sample breaks. A Bayesian estimation and forecasting procedure is provided which accounts for the uncertainty of structural breaks and model parameters. We discuss dynamic consistency when forecasting recursively with structural break models, which has been ignored in the existing literature, and provide a solution. Applied to monthly US data from 1964 to 2006, we find strong evidence of structural breaks in the predictive relation between the yield curve and output growth in late 1979 and early 1983. The short rate has more predictive power for output growth than the term spread before 1979 while the term spread becomes more significant since the breakof 1983. Incorporating the possibility of structural breaks improves out-of-sample forecasts of output growth from 1 to 12 months ahead.