We analyze 40 cosmological re-simulations of individual massive galaxies with present-day stellar masses of M * > 6.3 X 1010 M in order to investigate the physical origin of the observed strong increase in galaxy sizes and the decrease of the stellar velocity dispersions since redshift z 2. At present 25 out of 40 galaxies are quiescent with structural parameters (sizes and velocity dispersions) in agreement with local early-type galaxies. At z = 2 all simulated galaxies with M * 1011 M (11 out of 40) at z = 2 are compact with projected half-mass radii of 0.77 (?0.24) kpc and line-of-sight velocity dispersions within the projected half-mass radius of 262 (?28) km s--1 (3 out of 11 are already quiescent). Similar to observed compact early-type galaxies at high redshift, the simulated galaxies are clearly offset from the local mass-size and mass-velocity dispersion relations. Toward redshift zero the sizes increase by a factor of ~5-6, following R 1/2(1 + z) Delta *a with Delta *a = --1.44 for quiescent galaxies ( Delta *a = --1.12 for all galaxies). The velocity dispersions drop by about one-third since z 2, following Delta *s1/2(1 + z) Delta *b with Delta *b = 0.44 for the quiescent galaxies ( Delta *b = 0.37 for all galaxies). The simulated size and dispersion evolution is in good agreement with observations and results from the subsequent accretion and merging of stellar systems at z 2, which is a natural consequence of the hierarchical structure formation. A significant number of the simulated massive galaxies (7 out of 40) experience no merger more massive than 1:4 (usually considered as major mergers). On average, the dominant accretion mode is stellar minor mergers with a mass-weighted mass ratio of 1:5. We therefore conclude that the evolution of massive early-type galaxies since z 2 and their present-day properties are predominantly determined by frequent 'minor' mergers of moderate mass ratios and not by major mergers alone.