Is self-citation the ordinary accumulation of a research program or a form of strategic self-promotion? An author may cite earlier work because it is the natural foundation for what follows, a motive we term Need, or in order to raise the bibliometric indicators that weigh on hiring, promotion, and funding, a motive we term Greed. Because any single self-citation is consistent with either, which of the two governs ordinary practice has long remained unsettled. We take up the question with 2.04 million papers published between 2000 and 2025 by 1.64 million authors, drawn from OpenAlex and classified into 22 business and management fields by the Chartered ABS Academic Journal Guide, with the SJR and ABDC rankings retained for robustness. For every paper we assess how closely its content follows the author's earlier work, measuring the cosine similarity between the paper and the author's prior corpus with two sentence-transformer language models, MPNet (768-dimensional) and MiniLM (384-dimensional), across titles, abstracts, and concept hierarchies. We also separate first-degree self-citations, which an author makes directly, from second-degree ones, in which a co-author cites the team's earlier work. In nested Poisson regressions with author and field-by-year fixed effects, mechanical opportunity accounts for 98.2 percent of the explained variance in self-citation, whereas proximity to the h-index and i10 thresholds contributes less than 0.1 percent and, in all but two small fields, runs against the gaming interpretation. Topical relevance is the strongest behavioral predictor, and senior, highly cited authors, who would gain most from inflated metrics, self-cite the least relative to the standing they already hold. Self-citation is therefore predominantly need-driven, though a modest within-author rise since 2015 hints at a gradually shifting norm. Blanket penalties on self-citation would discourage legitimate scholarship in pursuit of a problem that few researchers have.