V.7:10 (326-330): A better way to smooth data by J.S. Payne, Ph.D.

A better way to smooth data by J.S. Payne, Ph.D.

What every trader desires is indicators that give strong signals with no misleading period-to period jitter, or noise. We all know what noise is to a trader, but let's restate it. Noise is something in your data that happens so quickly you can't take advantage of it. Noise is "high frequency" data not indicative of price movement sustained through several time periods. For a day trader, this could be a price spike shown in a one-minute bar chart. For an overnight trader, noise could be a price spike that occurred as a false breakout on a single day.

Many methods are used to separate the noise from the more meaningful price trends. The most popular remains moving averages, which combine the data from a set of periods and lower the significance of any one piece of data. If we can measure noise quantitatively then we can directly compare the "noise removal efficiency" of different types of moving averages. Traders have used moving averages for many years to reduce noise in their data, but with two penalties. The first and obvious penalty is that moving averages always lag the original data. A less obvious penalty is that not all noise is removed. In effect, we accept the lag which goes with moving averages to obtain imperfect noise reduction.

Early users of simple moving averages noted what was termed "the drop-off" effect. The drop-off effect is obvious when a spike in the data is suddenly dropped out of the calculation because it is now older than the averaging time period. There also is a less obvious "jump on" effect which occurs when a new piece of data is included in the average for the first time. The "jump on" and "drop-off" effects occur with all types of moving averages and are the mechanisms by which noise is transmitted from your data through the calculations to your moving average.