contents.gifindex.gifprev1.gifnext1.gif

Data Smoothing

Data smoothing techniques are used to eliminate "noise" and extract real trends and patterns. Below are some of the available smoothing methods.

Random

This method is best when each period's data has no relationship to the pattern in the previous data. Under this condition, the best prediction for the next value in a series is simply the average of all previous data points.

dph00246.gif

Random Walk

A random walk exists if the next data point is equal to the last data point plus some random deviation. Many financial securities move in this manner. Under this condition, the best prediction for the next value in a series is simply the last value.

dph00247.gif

Moving Average

This method works well if the data contains no trend or cyclic pattern.

dph00248.gif

n is a user-supplied constant greater than zero defining the number of consecutive points to average. Higher values cause greater smoothing.

Simple Exponential Smoothing

This method works well if the data contains no trend or cyclic pattern and the most recent data points are more significant than earlier points.

dph00249.gif

a is the smoothing constant.

Linear Exponential Smoothing (Holt's method)

This method works well if the data contains a trend but no cyclic pattern.

dph00250.gif

a is the level smoothing constant and b is the trend smoothing constant.

Seasonal Exponential Smoothing (Winter's method)

This method works well if the data contains a trend and a cyclic pattern.

dph00251.gif

a is the level smoothing constant, b is the trend smoothing constant, c is the seasonal smoothing constant, and p is the season period.

Vanguard Software Applications

Our collaborative, web-based forecasting applications unite roles, teams and departments in achieving the most accurate forecasts in the world – bar none.