Forecasting interest rates with shifting endpoints

Abstract

SUMMARY We consider forecasting the term structure of interest rates with the assumption that factors driving the yield curve are stationary around a slowly time‐varying mean or ‘shifting endpoint’. The shifting endpoints are captured using either (i) time series methods (exponential smoothing) or (ii) long‐range survey forecasts of either interest rates or inflation and output growth, or (iii) exponentially smoothed realizations of these macro variables. Allowing for shifting endpoints in yield curve factors provides substantial and significant gains in out‐of‐sample predictive accuracy, relative to stationary and random walk benchmarks. Forecast improvements are largest for long‐maturity interest rates and for long‐horizon forecasts. Copyright © 2013 John Wiley & Sons, Ltd.

Forecasting the US Term Structure of Interest Rates Using Nonparametric Functional Data Analysis

“…For example, at the 12-monthahead horizon, the random walk predictions gives a reduction in out-of-sample RMSFE of up to 10%, relative to DNS. This corroborates the findings of Duffee (2011) andvan Dijk et al (2014). On the other hand, DNS presents a good performance for long-term maturities, for example, for forecasting 8-year maturity 6 months ahead and 7-year maturity 12 months ahead.…”

“…The most notable example is the forecast for the short yield (3-18 months) over a 12-month horizon: this results in a mean squared forecast error that is 12% (10 percentage points) smaller than the best-performing basic method, and 23% smaller than the DNS model. In addition, similar to van Dijk et al (2014), over the sample for which we are assessing the forecasts, the random walk forecasts generally do better than the DNS predictions. For example, at the 12-monthahead horizon, the random walk predictions gives a reduction in out-of-sample RMSFE of up to 10%, relative to DNS.…”

Learning From Disagreement in the U.S. Treasury Bond Market

“…Bond yields embody a stochastic trend owing to K P 0t in ( 5) following a random walk. Consequently, long-horizon forecasts of yields may change stochastically over time (rather than converge to constants as in stationary models), and this often improves out-of-sample forecasts from affine term structure models (van Dijk et al (2014) and Bauer and Rudebusch (2020)). In our setting, BL learns about K P 0t in real-time conditional on the history of Z t , whereas K P 0t is often presumed to be known and well proxied by observable economic trend variables.…”

Does Real‐Time Macroeconomic Information Help to Predict Interest Rates?

“…(2017), and to incorporate long‐run inflation expectations as in van Dijk et al. (2014). We also plan to analyze the contribution of specific subsets of SPF forecasts by including individual survey predictions (instead of median survey forecasts) into the real‐time macro‐yields model.…”