Author(s)

S. Fink & J. Walde

Abstract

Due to the importance of the risk-free capital market interest rate, nearly all large economic and financial institutions deal with the analysis of its future development. Although sometimes advanced econometric methods (VAR, ECM) are used instead of or alongside the standard OLS regression approach, almost all of the work in this field deals with the basic assumption of (multi-variate, multi-equation) linear relationships between the variables. In our paper we try to find out whether nonlinearity can really be neglected. We apply artificial neural networks as a nonlinear modelling tool. Using monthly data from 1960–2005, we forecast the interest rate by means of multi-layer perceptrons (MLP). As a benchmark method we use vector autoregression models dealing with the identical dataset. The obtained results give evidence of the underlying nonlinearity of the problem. The MLP outperform the classical tools with regard to different error measures and especially in capturing the turning points of the interest curve. Keywords: error correction model, multi-layer perceptrons, interest rate, nonlinear modelling. 1 Introduction Due to the importance of the risk-free capital market rate, nearly all large economic and financial institutions deal with the analysis of its determinants and future development. In most cases the German 10-year Government Bond, the so-called \“Bund”, is used as a benchmark for the European capital market. As the futures markets are highly liquid, the most widely used forecasting technique

Keywords

error correction model, multi-layer perceptrons, interest rate, nonlinear modelling.