This paper analyzes the endogeneity of licensing arrangements in cost-reducing cumulative innovation. There exists the following results. First, for the first-generation patentee, ex post licensing matters for rent extraction while ex ante licensing matters for efficiency. Second, if the second-generation innovator does not exit, then the firms  profits as well as social welfare are all irrelevant to whether ex ante licensing is allowed. Third, costly litigation can occur on the equilibrium path and its occurrence is also irrelevant to ex ante licensing. Interestingly, the conditional probability of the first-generation patentee winning litigation first decreases and then increases in patent breadth. Fourth, optimal patent breadth depends on the tradeoff between litigation costs and antitrust effect.