The contribution of this paper is twofold. First, we develop a multi-state endogenous growth framework that potentially explains the above puzzles.4 Building on previous work by Acemoglu-Aghion-Zilibotti (2003), we model productivity growth as resulting from both imitation of frontier technology and innovation of technology. We posit that, while imitation mainly requires physical capital and less educated labor, innovation uses highly educated labor intensively. Moreover, workers can migrate, at a cost, towards states that pay higher wages for their skills. Thus, a person who is highly educated by a state that needs mainly to engage in imitation may migrate to a state where his skills will be used in innovation.