I develop a theoretical framework that explains the increasing and convex pattern of skill premia through diminishing aversion towards the ambiguous possibility of skill obsolescence. High-income workers are shown to invest more in education as their concern with forgone income is progressively lower than their less credentialed counterparts. As a result, high-skill (low-skill) individuals invest in their stock of human capital beyond (below) what is optimal if the true obsolescence frequency was known to them. This learning glut (deficit) subsequently pays large dividends (losses) during unexpected episodes that exhibit increased ambiguity. A calibration of the model is able to match the skill premium curve in the US economy.
Host: Christian Traeger