Dynamics of Interest Rate Curve by Functional Auto-regression

The paper applies methods of functional data analysis – functional auto-regression, principal components and canonical correlations – to the study of the dynamics of interest rate curve. In addition, it introduces a novel statistical tool based on the singular value decomposition of the functional cross-covariance operator. This tool is better suited for prediction purposes as opposed to either principal components or canonical correlations. Based on this tool, the paper provides a consistent method for estimating the functional auto-regression of interest rate curve. The theory is applied to estimating dynamics of Eurodollar futures rates. The results suggest that future movements of interest rates are predictable only at very short and very long horizons