Moving Average

Moving Average is a linear data smoothing technique. By reading a column of moving averages, you can discover trends in averages over time. The average of the adjacent observations replaces each observation. Using the moving averages of a data series reduces the amount of variation in the data.

Moving averages are often used to eliminate unwanted fluctuations, for example, to smooth the data series. For example, you can remove cyclical, seasonal, and irregular patterns and leave only the trend. A quadratic equation is calculated to fit a curve to the averaged data. This equation is used to reconstruct smoothed data at the ends of the series. The trend and forecast report uses a three-step moving average by default.

From the Moving average points drop-down list, select the number of moving average points. The Trend and Forecast report uses a three-step moving average by default. From the Forecast Points drop-down list, select the number of data points to forecast.

Single exponential smoothing

Single exponential smoothing is a weighted-averaging technique, where the weights decrease with the age of the data. Use single exponential smoothing when there is no trend in your data. The data is smoothed according to the following equation:

SIt = aXt + (1+ a)SIt-1

The equation has the following parts:

Double exponential smoothing

Double exponential smoothing performs two weighted-averaging calculations, using the results from the previous calculation as input for the next calculation. Use double exponential smoothing when your data has a linear trend. The data is smoothed according to the following equations:

SIt = aXt + (1+ a)SIt-1

SIIt = aSIt + (1+ a)SIIt-1

Triple exponential smoothing

Triple exponential smoothing performs three weighted-averaging calculations, using the results from the previous calculation as input for the next calculation. Use triple exponential smoothing when your data has a nonlinear trend. The data is smoothed according to the following equations:

SIt = aXt + (1+ a)SIt-1

SIIt = aSIt + (1+ a)SIIt-1

SIIIt = aSIIt + (1+ a)SIIIt-1

Adaptive response smoothing

Adaptive response smoothing is similar to single exponential smoothing, except that the value of the smoothing constant is calculated as a function of the data and previous errors. Thus the value of the smoothing constant is higher for larger errors and lower for smaller errors.

You can use adaptive response smoothing when your data does not show a trend. Use adaptive response smoothing for forecasting from a series without a trend because adaptive response smoothing is responsive to changes in the pattern of the data.

Linear forecasting

Linear forecasting uses historical data to fit to a linear curve.

Exponential forecasting

Exponential forecasting fits your base data points and forecast data points to an exponential curve.

Geometric forecasting

Geometric forecasting fits your historical data of base data points and forecast data points to a geometric curve.

Hyperbola forecasting

Hyperbola forecasting fits your historical data consisting of base data points and forecast data points to a hyperbolic curve.

Modified hyperbola forecasting

Modified hyperbola forecasting fits your historical data to a modified hyperbolic curve.

Quadratic forecasting

Quadratic forecasting applies the least-squares principle to quadratic curves that cannot be transformed directly into a linear equivalent. It fits your base data points and forecast data points to a quadratic curve.

Log quadratic forecasting

Log quadratic forecasting fits your base data points and forecast data points to a log quadratic curve.