We present a theory of capital structure based on the power of shareholders, bondholders and managers to control the incentive conflicts in large corporations. The manager–owner conflict produces a trade-off between inefficiency in the low state and rents in the high state, and the shareholder–bondholder conflict produces under-investment as in Myers [Journal of Financial Economics 19 (1997) 147]. Since managers and bondholders both prefer more efficient actions in the low state, the two conflicts are interdependent. With risk-less levels of debt, there are no shareholder–bondholder agency costs, but managerial control over the incentive-setting process produces excessive rents. With risky debt, shareholders focus more on returns in the high state so that shareholder–bondholder agency costs increase but managerial rents decrease. Efficient levels of debt holder protection facilitate a reduction in manager–owner agency costs that outweighs shareholder–bondholder agency costs, and are decreasing in firm performance. The results are consistent with the separate empirical results relating control to both compensation and leverage, and suggest how future studies can be integrated.

This paper studies financial policy, investment decisions and the threat of dismissal when managers value control and investments generate manager-specific value. A high probability of investigation focuses the manager on the profitability of replacement and therefore manager-specific value. The probability of an investigation increases when the firm enters bankruptcy. Thus, high debt levels focus the manager on investments that dissuade replacement during bankruptcy procedures. Dividends relax the manager's focus on manager-specific value since there is a lower probability of an investigation following a missed dividend. The ability to make dividend payments, however, is related to ex-post performance and can improve replacement decisions. When managerial quality is commonly known, the expected value of the firm is maximized with a combination of debt and dividend commitments. When managerial quality is hidden information, it is optimal for the combination of debt and dividend commitments to signal quality.