Decomposition of Japanese Yen Interest Rate Data Through Local Regression

Seven different Japanese Yen interest rates recorded on a daily basisfor the period from 1986 to 1992 are simultaneously analyzed. Byintroducing a new concept of short term trend, we decomposeeach interest rate series into three components, long termtrend, short term trend and irregular. It is obtained by atwo step lowess smoothing technique. After that, amultivariate autoregressive model (MAR) is fitted to the vectorvalued time series obtained by combining those seven irregularcomponents. The decomposition and MAR model fitting were quitesatisfactory. It enables us to understand well various aspects ofinterest rate series from the trends, the MAR (2) coefficientsand its residuals. The result is compared with the decompositionthrough sabl and the advantages of our procedure will bedemonstrated in relations to other parametric model fitting likeARCH or GARCH. Based on the decomposition we can have betterdaily prediction and more stable long term forecasting.