Testing for GARCH effects: a one-sided approach