Forecasting the term structure of interest rates for Turkey: a factor analysis approach

Abstract

We perform factor analysis on monthly yield curves estimated by Nelson-Siegel model using the Turkish secondary government securities market data. Monthly yield curves are characterized by three factors which are estimated using nominal volume-weighted average monthly zero-coupon yields. According to the loadings of each factor, we label the factors as level, slope and curvature. Next, we forecast yield curves using AR-GARCH and random walk processes and compare their relative performance. Our results indicate that the three factor model has high explanatory power and that the AR-GARCH specification has superior forecasting power for Turkey.

Acknowledgements

We would like to thank, without implicating, Atila Basaran, Burak Saltoglu and Hakan Kalkan for their suggestions in the process of preparing the paper. We also thank an anonymous referee and Oya Pinar Ardic for helpful comments. Additionally, we thank Information Systems, Research, Training and Publications department of Istanbul Stock Exchange for providing data. An earlier version of the paper was presented at session 4 in the Computational Management Science Conference, 2–5 April 2004, Neuchâtel, Switzerland; we also thank the participants for helpful comments. Alper acknowledges financial support from TUBA – GEBIP (Turkish Academy of Sciences – Young Scientists Scholarship Program).

Notes

¹ Among others, see McCulloch (1971), Vasicek and Fong (1981) and Nelson and Siegel (1987) on the estimation of the term structure; Litterman and Scheinkman (1991) and Knez et al . (1994) on the factor analysis of interest rates; Duffee (2000) and Diebold and Li (2003) on the forecasting of the term premia.

² Among others, see Knez et al . (1994), Bliss (1997b) and Diebold and Li (2003) on the explanatory power of the three factor model.

³ Among others, see Alper (2001), Berument and Günay (2003), Denizer et al . (2000), Özatay (2000) and Ertu[gtilde]rul and Selçuk (2001) on the Turkish economy during 1990s.

⁴ Factors estimated using correlation matrix or covariance matrix is essentially the same. Loadings are different however transformation of loadings is possible, see Johnson and Wichern (2002) for details. See Brooks and Skinner (2000) for an alternative interpretation of fundamental factors.

⁵ See Alper et al . 2004 for further details.

⁶ Diebold and Li (2003) find the parameters of Nelson–Siegel model, which they interpret as three factors, to follow an AR(1) process.