GCTA estimates exceed biometric models

Hi,

        We have a dataset of about 4500 parent-twin participants with various neurobiological responses and whom we have genotyped for ~520,000 SNPs. The issue is that for some of the responses, the GCTA estimated heritability exceeds that of the biometric model. In fact, we fit all sorts of biometric models (e.g., based only on twins, using twins and parents, examining ACE, ADE, ADCE, ADC’C’’E), and used different GRM cutoffs for GCTA and different subject samples (e.g., using GCTA defaults, selecting our own unrelated set of individuals by picking one twin from each family, using just parents, etc.) and it doesn’t matter – we still get GCTA values exceeding the biometric estimates for additive effects (e.g., GCTA estimated at .6, with biometric heritabilities in the .4-.5 range). What are the circumstances under which this could happen?

Thanks!