Selasa, 27 Mei 2014

Tulisan Bahasa Inggris (Decomposition Approach to Forecasting)



by Thomas Metcalf, Demand Media
Decomposition forecasting is a proven way to develop more accurate forecasts.
Decomposition is a forecasting technique that separates or decomposes historical data into different components and uses them to create a forecast that is more accurate than a simple trend line. By forecasting each component separately before combining them, you can assess the importance of each and emphasize or discount them according to changing market or economic conditions.
Forecasting With Trend Line
The easiest way to forecast any variable is to simply extend a trend line based on historical data. Whether you accomplish this manually with regression analysis or by using a spreadsheet such as Excel, you can establish a trend and extend it into the future. The shortcoming of this method is that it fails to take into account predictable fluctuations around the trend. For example, you might forecast a retail sales projection of 8 percent for next year based on historical information, but if you are looking at fourth-quarter sales, when most of your business occurs, you would be missing the mark if you did not account for the seasonal variation.
Decomposition Approach
The decomposition approach to forecasting recognizes that a forecast cannot be completed unless you include all components of historical data. Although the components may vary, depending on what variable you are forecasting, you might include a long-term underlying trend line, a cyclical variation such as a business cycle, which would fluctuate around the trend, and a seasonal variable, which could be based on weather or holiday consumer activity. Depending on the variable you are attempting to forecast, you could even include a weekly variable.
Decomposing Historical Data
To illustrate how decomposition forecasting works, consider projecting retail sales as an example. For simplification, assume the only variable applied to the long-term trend is a seasonal component. You can create the trend line using regression analysis. To determine the seasonal component, using your historical data, divide the actual value of sales by the trend value at that point. After you complete this for all of your historical data sets, you can compute an average for each of the four seasons to derive seasonal factors. To project sales for the fourth quarter, multiply the projected trend value for that future quarter by the seasonal factor. The projection you compute with this method is more accurate than using the trend line alone.
Expanding the Model
The formula for forecasting sales is R = ST, in which "R" equals sales revenue, "S" equals the seasonal component and "T" is the underlying trend line. The model can be expanded to include other components, such as a cyclical component. Obviously, the more components, the more difficult the computations, and that is when a program such as Excel comes in handy. As with all forecasting models, it is up to you to interpret and explain the significance of the data you use.

Source : http://smallbusiness.chron.com/decomposition-approach-forecasting-81131.html

Tidak ada komentar:

Posting Komentar