Working Paper: CEPR ID: DP7796
Authors: Andrea Carriero; George Kapetanios; Massimiliano Marcellino
Abstract: We propose a new approach to forecasting the term structure of interest rates, which allows to efficiently extract the information contained in a large panel of yields. In particular, we use a large Bayesian Vector Autoregression (BVAR) with an optimal amount of shrinkage towards univariate AR models. Focusing on the U.S., we provide an extensive study on the forecasting performance of our proposed model relative to most of the existing alternative specifications. While most of the existing evidence focuses on statistical measures of forecast accuracy, we also evaluate the performance of the alternative forecasts when used within trading schemes or as a basis for portfolio allocation. We extensively check the robustness of our results via subsample analysis and via a data based Monte Carlo simulation. We find that: i) our proposed BVAR approach produces forecasts systematically more accurate than the random walk forecasts, though the gains are small; ii) some models beat the BVAR for a few selected maturities and forecast horizons, but they perform much worse than the BVAR in the remaining cases; iii) predictive gains with respect to the random walk have decreased over time; iv) different loss functions (i.e., "statistical" vs "economic") lead to different ranking of specific models; v) modelling time variation in term premia is important and useful for forecasting.
Keywords: Bayesian methods; Forecasting; Term structure
JEL Codes: C11; C53; E43; E47
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| BVAR model (C32) | improved forecast accuracy (C53) |
| BVAR model (C32) | changing market conditions (D49) |
| evaluation criteria (C52) | perceived effectiveness of the forecasting models (C53) |
| modeling time variation in term premia (C22) | improved forecast accuracy (C53) |