We analyze the relationship between conglomerates’ internal capital markets and the efficiency of economy-wide capital allocation, and we identify a novel cost of conglomeration that arises from an equilibrium framework. Because of financial market imperfections engendered by imperfect investor protection, conglomerates that engage in winner-picking (Stein, 1997 [Internal capital markets and the competition for corporate resources. Journal of Finance 52, 111–133]) find it optimal to allocate scarce capital internally to mediocre projects, even when other firms in the economy have higher-productivity projects that are in need of additional capital. This bias for internal capital allocation can decrease allocative efficiency even when conglomerates have efficient internal capital markets, because a substantial presence of conglomerates might make it harder for other firms in the economy to raise capital. We also argue that the negative externality associated with conglomeration is particularly costly for countries that are at intermediary levels of financial development. In such countries, a high degree of conglomeration, generated, for example, by the control of the corporate sector by family business groups, could decrease the efficiency of the capital market. Our theory generates novel empirical predictions that cannot be derived in models that ignore the equilibrium effects of conglomerates. These predictions are consistent with anecdotal evidence that the presence of business groups in developing countries inhibits the growth of new independent firms because of a lack of finance.

We provide a new rationale for pyramidal ownership in family business groups. A pyramid allows a family to access all retained earnings of a firm it already controls to set up a new firm, and to share the new firm's nondiverted payoff with shareholders of the original firm. Our model is consistent with recent evidence of a small separation between ownership and control in some pyramids, and can differentiate between pyramids and dual‐class shares, even when either method can achieve the same deviation from one share–one vote. Other predictions of the model are consistent with both systematic and anecdotal evidence.