Decomposition pulls out predicted mean, trend, and seasonality to understand variability. You must understand seasonality and trend to identify variability/randomness.
Before you start
The page provides two fields that enable you to select the context as a combination of customer and product data. Both fields give options from both hierarchies at all levels. You can either select the required options from the lists displayed or manually enter the required value using the search function.
Decomposition pulls out predicted mean, trend, and seasonality to understand variability. You must understand seasonality and trend to identify variability/randomness.
Parameters
The Summary Statistics table displays key summary statistics for the product and customer combination selected:
| Parameter | Value |
| Mean | Displays the mean value of pure history. |
| Standard Deviation | Displays the dispersed data value in relation to the mean. |
| C-Var (Mean) | Displays the C-Var value defined as standard deviation divided by the mean of pure history. |
| C-Var (Trend) | Displays the C-Var value defined as standard error divided by the mean of the trendline. |
| C-Var (Residual) | Displays the difference between the standard deviation and predicted values. |
| Slope/Mean | Displays the slope value divided by the mean value. The slope of a line is the change in the y variable over the change in the x variable. |
Steps
The C-Var Comparison chart allows you to compare the C-Var measurements:
- C-Var (Mean): The C-Var defined as standard deviation divided by the mean of pure history.
- C-Var (Trend): The C-Var defined as standard error divided by the mean of the trendline.
- C-Var (Residual): The difference in standard deviation and predicted values.
The History Validation card indicates if there is enough history to perform decomposition analysis.
The Selected History chart displays the period from which the application can determine seasonality. It displays the Effective Adjusted History overlaid with Seasonality Initialization Range for each financial year (CFY: Current Financial Year) to provide a year-on-year view of demand.
- Seasonality Initialization Range: Displays the seasonality initialization range. The analysis considers the product effectivity date and displays the point from which the history is used to determine seasonality.
Note: A minimum of 24 months history is required to detect seasonal trend.
- Effective Adjusted History: Displays the demand history derived from D1120 History Correction.
The Year on Year Selected History chart provides a year-on-year comparison of the chained history (demand history). You can hover over the legend to view specific yearly views across a 52-week period. This helps you to review seasonality patterns of demand.
The Seasonal Indices chart measures how a season throughout a cycle compares with the average season.
The Trend chart displays the trend, which is what's left once the seasonality index is removed from history.
Residual
The Residual chart displays what is left in a time series model after fitting a model. For many, but not all, time series models, the residuals are equal to the difference between the observations and the corresponding fitted values.
Residuals are useful in checking whether a model has adequately captured the information in the data. A good forecasting method yields residuals with the properties below:
- Uncorralated residuals: If there are correlations between residuals, there's information left in the residuals that should be used in computing forecasts.
- Residuals have zero mean: If the residuals have a mean other than zero, the forecasts are biased.
