This research proposes an economic model for wind farm owners and operators to predict
the amount of electricity to sell to the market to optimize revenue. The model takes as
inputs predictions of imbalance prices as well as probabilistic predictions of wind generation
at the farm. The proposed statistical methodology improves upon current forecasting
by focusing on the accuracy of predictions at extreme events. We decompose the prediction
of prices into a baseline and spike component. The mixture model of a seasonal autoregressive
component for the daily baseline behavior and a autoregressive conditional duration
model for the extreme events, achieves a higher level of accuracy in the prediction and
therefore a higher revenue to the wind farm. Through the improved predictions of price as
well as a probabilistic forecast of wind output, not only can wind farms maximize revenue,
but the independent system operators (IS Os), who control operations of the grid, can also
better account for wind generation in the dispatch process, thus allowing wind to become a
reliable source of power generation.