Least-Squares Data Smoothing by Harold A. Kreamer
Superimposed on and often indistinguishable from daily market data trends are seemingly random fluctuations, known as noise. An analyst wants to remove as much noise from data as possible without degrading underlying valid trends. To remove noise from stock, commodity and index data trends as efficiently as possible, I use the least-squares method.
Whereas a lengthy moving average that goes through a sharp peak or trough reduces the extreme value, the least squares method retains most of that value. Triangular and exponential methods purposely weigh data asymmetrically, which introduces some distortion. Least squares minimizes data distortion. In addition, least-squares data smoothing can also be used as a prelude for further data treatments or as a method of analysis on its own.