Abstract
Analysis of the future behaviour of economic variables can be biased if structural breaks are not considered. When these structural breaks are present, the in-sample fit of a model gives us a poor guide to ex ante forecast performance. This problem is true for both univariate and multivariate analysis and can be extremely important when co-integration relationships are analysed. The main goal of this article is to analyse the impact of structural breaks on forecast accuracy evaluation. We focus on forecasting several interest rates from the Spanish interbank money market. In order to carry out the analysis, we perform two forecasting exercises: (a) without structural breaks and (b) when structural breaks are explicitly considered. We use new sequential methods in order to estimate change-points in an endogenous way. This method allows us to detect structural breaks in all four rates in May 1993. However, the effects of these breaks are not very strong, since we found scarce gains in forecasting accuracy when the structural breaks are included in the models.
Acknowledgements
We thank the participants of the 20th International Symposium on Forecasting for many helpful comments. We also thank Mayte Ledo and Banco Bilbao Vizcaya-Argentaria (BBVA) who kindly provided us with the data. We acknowledge financial support from the Spanish Ministry of Science and Technology (SEJ2006-14354) and Comunidad de Madrid (UCM940063).
Notes
1 We focus here in interest rate forecasting. Garcia and Perron (Citation1996), Dahlquist and Gray (Citation2000) or Bekaert et al. (Citation2001) use regime switching models for interest rates, but in these cases the relevant measure is the in-sample fit of the model.
2 We can find some examples in this line in the section ‘Break Points And Unit Roots’ in Journal of Business and Economic Statistics (Citation1992) where several tests to detect a break point, within the dataset, are presented. Also, it is important to mention the test procedures proposed by Andrews (Citation1993) in a nonlinear context, Bai (Citation1994) and Bai and Perron (1996) for single and multiple change points in the mean of a general process respectively.
3 Continuously compounded rates are computed as r t ≡ (360/N)ln(1 + (N/360)S t ) where S t is 360 days basis simple interest rates corresponding to 1, 3, 6 and 12 months to maturity. N indicates the maturity expressed in number of days.
4 Mean absolute error (MAE), Mean absolute percentage error (MAPE), root mean squared error (RMSE), root mean squared percentage error (RMSPE) and three different versions of the Theil-U coefficient (UTHEIL1, UTHEIL2, UTHEIL3).
5 This test is also valid for loss function that will not be a direct function of this error.