Among distributed energy resources, wind power has recently showed great potential and it is being promoted in many countries. The wind power penetration increase and the trend of the wind farms to enter the market, makes necessary the development of new prediction tools. Prediction tools have been fully verified and used for demand forecast and, more recently, to predict the market prices. Among the different prediction methods proposed, the initial and more deeply verified were the statistical ones. In the present work, time series statistical methods are used to explore and assess the achievable improvement over a persistent model. Both autoregressive (AR) models that consider wind power time series and antoregressive with exogenous variable (ARX) models which include wind speed time series are used to carry out the prediction. An analysis of the appropriate parameters for the models is carried out and shows, for different wind farms, comparative results for persistent, AR, and ARX models with 6, 12, and 24 It forecast which can be useful for the day-ahead and hour-ahead electricity markets. The study also includes a comparative analysis of the different wind farms considered both independently and in an aggregated manner. The forecast improvement due to the use of ARX models and to the wind farm aggregation is fully assessed.