Content area
Full Text
(ProQuest: ... denotes formulae omitted.)
Campbell and Shiller [1991], Cochrane and Piazzesi [2005], Diebold and Li [2006] and many others have shown that today's yield curve possesses significant information about the dynamics of future yields. Vector autoregression (VAR) models can forecast interest rates with different maturities, but these forecasts can contain arbitrage opportunities. To avoid arbitrage it is important to use affine term structure models.' However, many affine models focus only on in-sample fit as opposed to out-of-sample forecasting (de Jong [2000] and Dai and Singleton [2000]). Moreover, those that do focus on outof-sample forecasting, notably Duffee [2002], conclude that the models forecast poorly. In this article, we review some models that are used to investigate the expectations of professional economic forecasters2 for the purpose of out-of-sample forecasting. The results suggest that survey data from professional forecasters generate significant improvements in interest rate forecasts up to one year ahead.
There is a vast literature on predictability of riskless interest rates. Fama and Bliss [1987] , Campbell and Shiller [1991], Cochrane and Piazzesi [2005], Diebold and Li [2006] and many others have shown that the current yield curve possesses significant information about the dynamics of future yields. Many yield curve models ignore the information from macroeconomic variables. However, Ang and Piazzesi [2003], Ang, Dong, and Piazzesi [2007], Ludvigson and Ng [2005], Aruoba, Diebold, and Rudebusch [2006] and others have also considered models that directly link these latent factors to macroeconomic variables.
In a widely recognized paper, Ang and Piazzesi [2003] augmented a standard threefactor term structure model with two macroeconomic variables that enter the model through a short-term interest rate equation. They reported that the macroeconomic variables account for a large share of the variation in the short-term and middle section of the yield curve. Other works, such as Aruoba, Diebold, and Rudebusch [2006], Ludvigson and Ng [2005], and Dewachter and Lyrio [2006], also consistently found the usefulness of macro variables in explaining and/or predicting government bond yields, but all of their work exploited small macroeconomic information sets, and each model could also work with different information sets. Stock and Watson [2002] published an extensive menu of alternative sources of additional information when they analyzed 215 different macroeconomic variables. Even after the highly correlated time series...