Im estimating the carhart 4 factor model. Im testing for heteroskedasticity to see whether i need to use adjusted standard errors, but i am finding conflicted results. All but one test (ARCH) are unable to reject the null of homoskedasticity (p values between .4 and .8). For some reason both ARCH and GARCH do find heteroskedasticity, and im not sure why.
Does anyone know the reasons behind this? Should I use robust standard errors based on the ARCH model or follow the other models?