The Ap index is a daily average of the range of magnetic disturbance in each three-hour UT interval. It is the only global magnetic index routinely forecast. Predictions of Ap are subjective estimates made on the basis of past behavior of Ap and events at the Sun. The predicted variance in present forecasts is only 30%. Here we develop a set of linear filters for Ap forecasting. We show that auto regressive filters with constant coefficients do as well as human forecasters. Filters with time varying coefficients predict 37% of the variance. If we assume Ap is driven by solar wind velocity through an auto regressive moving average we predict 47% of the variance. However, this filter requires today's average solar wind velocity to predict today's Ap. Only with remote sensing of the solar wind a day in advance would it be possible to implement this model. We apply our filters to data for January 1997 determining whether effects of the January 6--11 magnetic cloud can be predicted. The filters do not predict CME related activity, but they do predict activity later in January obtaining a monthly average prediction efficiency of 59% that should be compared SEC results of 40%. ¿ 1998 American Geophysical Union