GCTA-COJO: Imputed vs Hard-called

Hi,

Thanks for developing and maintaining GCTA. Much appreciated.

I am running GCTA-COJO --cojo-slct to select independently associated SNPs.

My data is imputed to HRC and is in VCF format.  Currently, I have converted it to PLINK binary format, which I assume hardcalls the data as mentioned here: www.cog-genomics.org/plink2/input#oxford.

My questions are:

1. It is worthwhile to use genotype probabilities as opposed to hard-called genotypes?
2. If is it worthwhile to use genotype probabilities, how to I convert from VCF to MACH format? From what I can see existing tools do not do this, such as QCTOOL, BCFTools, and PLINK.

Thanks,

Vince