Taking the special nonlinear characteristics of the
domestic and international gold price into account, this paper
systematically analyzed its nonlinearity by the methods of BDS
test, R/S analysis and improved largest Lyapunov exponent. We find
three main results: (1) ARMA-GARCH model could adequately explain
the linear and nonlinear dependence of gold price series; (2)
long-memory does not exist anymore in price series explained by
ARMA-GARCH model; (3) chaos phenomenon which is sensitive to the
initial value does not exist either in the residuals of regression
model. Therefore, we believe that the nonlinearity of gold price
is mainly characterized in conditional heteroscedasticity rather
than chaos.