Phenotype-from-genotype prediction models promote adoption of precision medicine by identifying individuals genetically susceptible to a certain disease, which may lead to lifestyle changes, regular screening and earlier intervention. The training data for these models is decentralized and very sensitive, which makes pooling all data together impossible. As a result, researchers commonly train their models on a single dataset stored at a single location. Federated learning is a novel yet rapidly developing machine learning paradigm that trains a model in multiple data silos, increasing accuracy and reducing bias. The successful applications of federated learning are rapidly emerging across all healthcare domains. Our manuscript contributes to the adoption of federated models in healthcare by providing extensive evidence of their promise and analyzing their behavior. To the best of our knowledge, this is the first research on the performance of federated models on genomic data.