Although almost half of the world's population lives under nondemocratic regimes, the questions of how policy decisions are made and how power changes hands in nondemocracies have received relatively little attention in the political economy literature. Gordon Tullock (1987) suggested that because there are no strong institutions ensuring consensus and regulating the election and succession of leaders, non-democratic regimes rapidly degenerate into personal rule, where a single dictator dominates every aspect of decision-making. In this paper, we draw on our work on dynamic coalition formation and investigate Tullock's conjecture formally. Our game-theoretic analysis leads to the opposite of Tullock's conjecture: provided that players are sufficiently forward-looking, juntas do not dynamically converge to personal rule. On the contrary, relatively large juntas may emerge and persist as ruling coalitions for a very simple and intuitive reason: the absence of strong institutions not only enables some junta members to eliminate others, but also implies that current members cannot make credible commitments and in particular cannot refrain from engaging in further rounds of elimination.