Building Non-Lagging Stream Filters with Multiple Candidate Averages

Summary

The article proposes cluster filters for smoothing non-stationary data as it arrives. Rather than applying one conventional filter to a complete historical series, the method runs several filters in parallel and selects among their outputs using a model of how the underlying sequence behaves. Its simple example assumes price characteristics change over time, treats each close as noisy, places the latent value between consecutive closes, and expects direction to persist. It then chooses between short simple and exponential moving averages according to the prior filter direction and their positions relative to the previous value.

The author presents visual comparisons in which the basic cluster filter appears more responsive than individual averages and roughly comparable to a tuned JJMA. A more complex version is illustrated smoothing momentum and preserving a unit impulse without visible delay or distortion. These are demonstrations, not evidence of predictive performance or trading profitability. The article acknowledges that the apparent ideal response is a special case for a nonlinear filter and says broader study is needed before drawing general conclusions; its final claim is that fully featured non-lagging indicators may be possible.

Key ideas

• Cluster filters select among multiple filter outputs to estimate incoming stream data.
• The example combines short simple and exponential averages under assumptions about noise and trend persistence.
• Filter selection depends on the previous direction and the candidate averages' positions.
• The article's visual tests suggest responsive smoothing, but do not demonstrate a trading edge.
• The impulse example is a special case and does not prove that a perfect general filter exists.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.