Genetic correlations (GC) are relevant parameters in genetic evaluations, particularly when a breeding program aims to achieve genetic progress for multiple traits altogether. These correlations are typically estimated from a base population and included among the many parameters that define the distribution used to predict breeding values (BV) for the selection candidates. In such a fashion, GC are assumed to be identical for all selection candidates. However, on a preliminary study on the output predicted BV of sires with more than 500 daughters from a dairy population, we observed that the GC among daughters from different sires may differ substantially, i.e., different sires may express different GC between traits through their daughters. Thus, if GC are specific values inherent to each individual, they could be considered as a phenotype. Moreover, negative GC may simply be a consequence of a concealed regulatory trait, responsible to balance the trade-off between observable traits. Considering the classic antagonism between production and fertility traits, it is reasonable to believe that individuals on the extremes of the trade-off distribution (i.e. individuals who present a very high breeding value for one trait but a very low breeding value for the other trait) are likely to present a low breeding value for this concealed regulatory trait. However, due to our inability to directly measure this potential regulatory trait, it can be considered a latent phenotype, making it difficult to be included in the unified index for the selection candidates. It is possible though, to use simulations to assess the genetic progress of such regulatory trait under different scenarios of selection in a breeding program, and understand the medium to long-term consequences of these scenarios on the regulatory trait and on the observed genetic correlation between the measurable traits. Under the assumption that the genetic correlation between production (PROD) and fertility (FERT) traits is an observable consequence of a latent regulatory (RGLT) trait, we have simulated RGLT with h2=0.1 on a base population, and then simulated PROD and FERT with h2=0.3 and h2=0.04 respectively, such that they had an average concave parabolic relationship with RGLT, with parameters fine-tuned to achieve an observed genetic correlation of -0.2 between the simulated BV for PROD and FERT. A fourth generic (GENR) trait with h2=0.3 was simulated with a positive genetic correlation of 0.8 with RGLT, and very low resulting observed genetic correlations between GENR and PROD and between GENR and FERT (approx. -0.02 and 0.03 respectively). GENR represented any trait also controlled by RGLT, however with no apparent relationship with PROD or FERT, such that GENR is not yet considered in a genetic evaluation, but may be of interest in the future. We evolved this base population over 50 generations under three different scenarios of selection: (1) uniquely for PROD; (2) for both PROD and FERT with weights 0.8 and 0.2 respectively; and (3) for PROD, FERT, and RGLT with weights 0.8, 0.15, and 0.05 respectively, assuming that we could obtain BV for RGLT. Simulations were replicated 500 times in order to asses the genetic progress of PROD, FERT, RGLT, and GENR, as well as their genetic correlations. At the end of the 50 generations, genetic progress of the traits was best/intermediate/worst for scenarios 1/3/2 for PROD, 2/3/1 for FERT, 3/2/1 for RGLT, and 1/3/2 for GENR. A final result was that under the assumption that the genetic correlation between PROD and FERT is a consequence of RGLT, rather than a parameter modulating the bivariate distribution of PROD and FERT, the selection for both PROD and FERT did not result in a great intensification of the negative genetic correlation between these two traits. This result contrasted our previous findings for dual selection of antagonistic traits, in which negative correlations were greatly intensified, a result mathematically and statistically sound, but biologically debatable. Therefore, the findings of our simulation study provide support to our hypothesis that GC are an observed consequence of a latent regulatory phenotype. Last but not least, although our study focused on the antagonism between PROD and FERT, the concept that GC may be the consequence of a latent phenotype can be extended to many other traits. This project has received funding from the European Union’s Horizon 2020 Programme for Research & Innovation under grant agreement n°101000226.