A major issue in integrating renewable energy into power grids is short-term forecasting. If some share of electric power is derived from renewable sources, gaps between demand and supply must be made up by other forms of generation. Because of the uniquely short-lived nature of electricity, utilities need to be able to forecast over horizons of a few hours. Up to now, studies of wave energy have relied primarily on the flux, due to the unavailability of data on power flows. This study analyzes the power output using simulations. Five types of converters are simulated. The Pelamis P2 is currently in operation. The other four are currently under development: the two-body heaving converter, the heave buoy array, the three-body oscillating flap device, and the oscillating water column. Fluxes and power output series are calculated for three sites in the Pacific Northwest. Two sets of forecasting experiments are run for the fluxes and the power series. The first uses time series models—regressions and neural networks. The second uses a large-scale physics model, WAVEWATCH III. The time series models forecast more accurately over short horizons, but the error increases rapidly as the horizon extends. The physics model generates similar degrees of accuracy over a range of horizons. The convergence point, at which time series and physics models yield similar degrees of accuracy, is in the range of 8–11 h. A third set of forecasting experiments is run for the gap between the demand for power and the supply from waves. The optimal means of predicting the gap is to run separate forecasts for demand and supply, and take the difference. The forecast error is lower than when the gap is forecasted directly.