Abstract
How should an economist choose which open question to work on next? I build a directed literature graph from 242,595 published economics-facing papers spanning 1976 to 2026, rank still-open candidate questions using only the graph available at each historical date, and test those rankings against later publication. A learned reranker built on that graph achieves a 17.0% top-100 hit rate for mechanism-thickening questions and 10.7% for direct-closure questions at a five-year horizon, versus 1.7% for both under preferential attachment. The two families differ in what the screening recovers: mechanism-thickening yields denser shortlists, while direct-closure captures a larger share of all eventual realizations. These results establish that local literature structure contains useful upstream screening information, and that economics moves by deepening mechanisms around existing claims at roughly 3 to 12 times the rate at which it closes locally implied direct relations.
Keywords: research direction; economics of science; scientific discovery; question choice; literature graphs
JEL codes: O31, O33, C45, C81