It is well established that returns to spending on research and development are positive and accrue over several years — that is, firms benefit from higher levels of cumulative R&D spending. The authors study how a given amount of such spending is best allocated, over time, to optimize R&D performance. Under a persistent policy, allocations remain nearly constant irrespective of circumstances; under a dynamic policy, R&D spending increases (resp., declines) when opportunities arise (resp., fail to materialize). The authors use a sample of 3,711 publicly listed companies, observed for seven years (on average) between 1982 and 2003, to compare the outcomes of these R&D allocation policies. They find that a dynamic allocation strategy is associated with worse R&D performance in terms of patent quantity and quality. The results indicate that the originality of an invention, and also the firm’s familiarity with an invention’s technological basis, are factors that can mitigate or amplify the harm caused by variability. Finally, the authors establish that R&D performance suffers from the unpredictable part of dynamic spending; the predictable part has either no effect or a positive one.