We use a hybrid observational/theoretical approach to study the relation between galaxy kinematics and the derived stellar and halo masses of galaxies up to z = 3 as a function of stellar mass, redshift and morphology. Our observational sample consists of a concatenation of 1125 galaxies with kinematic measurements at 0.4 < z < 3 from long-slit and integral-field studies. We investigate several ways to measure halo masses from observations based on results from semi-analytical models, showing that galaxy halo masses can be retrieved with a scatter of ∼ 0.4 dex by using only stellar masses. We discover a third parameter, relating to the time of the formation of the halo, which reduces the scatter in the relation between the stellar and halo masses, such that systems forming earlier have a higher stellar mass to halo mass ratio, which we also find observationally. We find that this scatter correlates with morphology, such that early-type, or older stellar systems, have higher M * /M halo ratios. We furthermore show using this approach, and through weak lensing and abundance matching, that the ratio of stellar to halo mass does not significantly evolve with redshift at 1 < z < 3. This is evidence for the regulated hierarchical assembly of galaxies such that the ratio of stellar to dark matter mass remains approximately constant since z = 2. We use these results to show that the dark matter accretion rate evolves from dM halo /dt ∼ 4000 M year −1 at z ∼ 2.5, to a few 100 M year −1 by z ∼ 0.5.