Unit root tests with conditional heteroskedasticity

摘要: Abstract This paper considers the finite-sample accuracy (size) of Dickey-Fullerunit root tests when errors are conditionally heteroskedastic. We consider specific case that error variance follows a GARCH (1,1) model. The Dickey-Fuller tend to overreject in presence errors, but problem is not very serious except process nearly degenerate (in sense ratio intercept initial near zero) and volatility parameter large.