Testing the Forecasting Ability of Multi-Factor Models on Non-US Interbank Rates

Abstract

This article examines the forecasting performance of continuous-time multi-factor models, in comparison with other parsimonious models, for the term structure of interbank rates in the UK, Europe, and Japan. The article employs two general dynamic frameworks with different factor structures: the generalized Chan-Karolyi-Longstaff-Sanders family of models and the arbitrage-free dynamic Nelson-Siegel family of models. Applying a battery of accuracy measures and a range of formal tests of forecasting superiority, this research provides evidence that extended multi-factor models demonstrate good out-of-sample forecasting performance for the short segment of the yield curve. However, for the euro and in part for the yen, random walk forecasts consistently pass various tests, indicating a higher level of market efficiency compared to the pound sterling interbank market. Key Findings ▪ For the term structure of interbank rates in the UK, Europe, and Japan, more complex continuous-time models that include more factors are superior in terms of predictive power to models with less factors or discrete-time models. ▪ Based on a battery of accuracy tests and a range of formal tests, extended multi-factor models demonstrate good out-of-sample forecasting performance for the short segment of the yield curve. ▪ For the euro and in part for the yen, random walk forecasts consistently pass various tests, indicating a higher level of market efficiency compared to the pound sterling interbank market.