erc/metu
INTERNATIONAL CONFERENCE IN ECONOMICS  IV
September 13-16, 2000, Ankara

Testing for Nonlinear Dependencies and Chaos in the Turkish Daily Gold Returns

Murat Çinko (Marmara University)
Burak Saltoğlu (Marmara University)

Abstract

The return predictability in financial markets has been an old and debated areas over a long time. One of the recent challenges in this context is to test the existence of non-linear dependencies and chaotic structures. This is because the seemingly random nature of the return pattern might be due to a deterministic chaotic structure. Therefore determining whether a return series do show random pattern or has a deterministic nonlinear structure is of vital importance. This is also important from the market efficiency point of view. In this paper we aimed to test whether the Turkish daily gold return series do have a nonlinear pattern by certain nonlinearity tests. To this end we have selected to apply the five of most popular nonlinearity tests which can be seen in Barnett et al (1997) on the daily Turkish Gold returns. Contributive aspects of this paper is twofold. First, very little has been done in testing non-linearity of gold returns (both in domestic and international markets, except Stengos (1996) who has applied the BDS test). The relative success of each of these non-linearity tests can be best assessed by applying them on various areas and data. The second contribution of this paper is to thoroughly investigate the time series properties of the Turkish gold returns. As can be seen in the literature very little has been done in modelling the gold returns (both linear and nonlinear aspects) in Turkey ever since it has been established in 1995. To conduct the anlysis, the following five nonlinearity tests (as discussed in Barnett et al (1997)) developed by Hinich (1982) , Brock et al (1996), Bispecturum test, the Lyapunov exponent estimator of Nychka et al( 1992), Lee et al (1993) and finally Kaplan (1994) were applied on the daily Turkish gold returns. The daily data spans between August 1 1995 of and January 29 of 2000 with 1120 observations. The data has been prefiltered to remove the potential seasonality. The results we have obtained in this paper indicate that in almost all cases we accepted the existence of nonlinear return predictability.Furthermore, the market efficieny implications of these findings were discussed.

Economic Research Center
Middle East Technical University
06531 Ankara Turkey
Phone: + 90 312 210 3044, 210 2003
Fax: +90 312 210 1244
e-mail: metuerc@metu.edu.tr