This paper discusses a resampling procedure in estimation of optimal
portfolios when the returns are the class of nonstationary ARCH models with timevarying
parameters. The asymptotic properties of weighted Gaussian quasi maximum
likelihood estimators ? ?GQML of time-varying ARCH(p) processes are studied, including
asymptotic normality. In particular, the extra bias due to nonstationarity of the
process is investigated. We consider bias adjusted estimators ??GQML by use of resampling.
In this paper we assume that the optimal portfolio weight g depends on the
ARCH parameter ?, i.e., g = g(?). Then the asymptotic distribution of the optimal
portfolio estimator g(??GQML) is derived. We numerically evaluate the magnitude
of g( ? ?GQML) and g(??GQML) for actual financial data, which shows eventually the
effect of bias adjustment