Abstract
<jats:p>We develop a theory of growth in which firms forage in idea space. A firm exploits a patch of related ideas, gradually exhausting opportunities for quality improvement, and then searches for a new patch. We cast this explore-exploit tradeoff as a tractable optimal-stopping problem and embed it in an endogenous-growth model. The composition of innovation — improving existing ideas versus discovering new ground — emerges as an equilibrium object. To construct an empirical representation of the idea space, we apply natural language processing to patent text data. The data support the theory’s central premises: returns to local exploitation diminish; firms stay longer on richer patches; and entry into new patches yields more and better patents. We calibrate the model to U.S. data and establish two results, on the composition of growth and on its pace. First, at a twenty-year horizon, patenting in new clusters accounts for over half of growth from quality improvements: sustained growth rests on firms continually entering new territory. Second, the model sign-identifies the origins of the productivity slowdown of the last four decades: exploitation spells have not shortened, weighing against worsening exploitation and tentatively pointing to harder exploration.</jats:p>