In order to extract as much energy as possible from ocean waves, an optimal control must be implemented in a wave energy converter (WEC), which requires the knowledge of the future incident waves (η). One of the most used methods to predict the future η, is to use a linear combination of past η values. Several models can be found in the literature, but only two of these models are compared in this paper, the autoregressive (AR) and autoregressive moving average (ARMA) models. Real wave data from different locations is used to determine which model is the best and in which scenario. This comparison addresses the discrepancies between [1], where the ARMA model is discarded for showing no improvement against the AR, and [2], which states that the ARMA model does improve the AR. The present paper shows that the two models achieve a similar performance for all the different conditions analysed. Thus, due to the simplicity and the lower computational requirement, the AR model is chosen as the best model for prediction.