This study compares the size and power of autoregressive conditional heteroskedasticity (ARCH) tests that are robust to the presence of a misspecified conditional mean. The approaches employed are based on two nonparametric regressions for the conditional mean: an ARCH test with a...