Forecast Method

Forecast method used to generate current forecasted data.

Name

The name of the forecast parameter scheme.

Forecast Stream

The forecast stream to which these parameters will apply.

Seasonal Profile

The seasonal profile assigned to the forecast parameter.

Seasonal profiles contain multipliers for each time slice of the year. Each slice's final forecast value is calculated by multiplying the base forecast value by the slice's seasonal multiplier.

A seasonal profile must be established for this to be populated.

# of Slices for Forecast Error Calculation

The number of slices of demand history that will be used to calculate forecast error. Forecast error is based on the difference between demand history and archived forecast and is calculated and used for MAPE, MAD, and RMSE on the Demand page. These are for display only and have no effect on Best Fit calculations.

Std Deviation - Use forecast error

Yes indicates that Servigistics will use forecast errors in Forecast Error Standard Deviation calculations, and then Forecast Error Standard Deviation will be used in the Safety Stock calculations.

No indicates that Forecast Error Standard Deviation will not be calculated. Servigistics will use demand history in History Average/History Standard Deviation calculations.

For trending and highly seasonal pairs, using Forecast Error Standard Deviation (instead of History Standard Deviation) in Safety Stock calculations generates an optimal Safety Stock value.

Std Deviation - Use Scaled Percentage

Yes indicates that Forecast Standard Deviation will be used in calculating levels. Forecast Standard Deviation will be computed using Coefficient of Variation:

ForecastSD = (HistorySD/HistoryAvg) * ForecastAvg

where ForecastAvg is the average of forecasts across forecast horizon (for example, the forecast horizon in stream configuration detail will be used in the ForecastAvg calculation).

Std Deviation - Use de-seasonalized and de-trended history

Yes indicates that history data will be de-seasonalized and de-trended when calculating the standard deviation.

For pairs with seasonal or trended demand this will usually result in a lower standard deviation. Lower standard deviation can lead to lower Safety Stock.

This parameter must be set to No if you are using Outlier Detection (when you set a value for Outlier - Definition).

Enable Tracking Signal

A measurement that indicates whether the forecast average is keeping pace with any genuine upward or downward changes in demand. It monitors the forecast to see if it is biased high or low. As used in forecasting, tracking signal is the number of mean absolute deviations that the forecast value is above or below the actual occurrence.

Ignore demand before first non-zero slice

New parts or new locations may have zero demand values in older demand history slices. This setting determines whether the system will ignore these zeros and base the forecast for each SKU on a smaller demand horizon that starts with the SKU's first slice with a demand value greater than zero; or will assume the zeros are valid demand values and use the entire demand horizon.

Yes indicates that the system will only use history slices forward from the first non-zero value. No indicates that the system will use all values over the demand horizon.

For example, if set to Yes, and the demand horizon is set to 36 slices, with the first non-zero demand occurring 3 slices ago, then only 3 slices will be used for the forecast. If set to No, then all 36 slices will be used.

Note that there are global settings that can control whether the system uses the first non-null value or assumes null values are zeros.

Use Phase In Date

Phase In Date is set for parts when creating or modifying a part or when modifying a SKU. It indicates when the part will become active.

When Use Phase In Date is set to Yes, forecasting will respect the part's Phase In Date when generating forecasts for new parts.

Use Phase Out Date

Phase Out Date is set for parts when creating or modifying a part or when modifying a SKU. It indicates when the part will become inactive.

When Use Phase Out Date is set to Yes, forecasting will respect the part's Phase Out Date when generating forecasts for new parts.

Note that forecasting will not generate forecasts for end-of-life parts with valid Phase Out Dates.

# of History Slices

Number of history slices to compare against historical forecasted data.

# of Horizon Slices

The number of periods of forecasting that are projected.

Maximum # of History Slices

Servigistics will use the maximum of this number or the stream configuration demand history horizon as the maximum number of slices of demand history data that can be used by the Best Fit algorithms. (Certain global settings may further restrict the number of available demand history slices.)

These slices are used as both the demand values from which the forecasts are calculated; and as the demand values that are compared to the best fit forecasts to calculate error for the candidate forecast methods.

Note that Winters Multiplicative and Crostons forecasts require at least one year of demand history data.

Outlier Management

Demand outliers occur when there is an anomaly in a slice of demand history (where anomaly is defined by Outlier - Definition). Outlier - Management indicates what actions to take when an outlier is detected where:

See Outlier Adjustments for more information.

Outlier Detection Limit

Defines a demand outlier, which is an anomaly in a demand slice. The threshold in numbers of history data standard deviations (+ or -) where a data point is flagged as an outlier. Can be any value between 1.0 and 5.0 inclusive.

Outlier Correction Limit

A factor applied to the standard deviation which defines the limit to which you want the Outlier Adjustment to correct the demand history point. A higher value results in a smaller Outlier Adjustment value. The value entered can be between 0 and 5.0.

Outlier Max Occurrence

The maximum number of consecutive data points that can be considered as outliers, rather than legitimate data. This number represents the point at which you believe adjustments should be stopped because a new, valid trend for demand may have begun.

Null means that there is no maximum, so Servigistics will always create adjustments or notifications for outliers.

Zero is not allowed, because it has no meaning.

EXAMPLE: A value of 3 means that after the 3rd consecutive instance of an outlier (the 4th time one is found), Servigistics will remove the three previously created adjustments and not create an adjustment or notification for the fourth outlier candidate.

Adjustment Lower Limit

Yes indicates that Servigistics will automatically adjust the demand history outlier if the demand value is less than the lower control limit (LCL). No indicates that Servigistics will not adjust the demand history outlier if the demand value is less than LCL.

LCL = History Average - (Outlier Detection Limit * Standard Deviation)

Use Confidence Level for Outlier Adjustments

The method for calculating outlier adjustments. Valid options are:

Upper Control Limit Confidence Level

The percentage of the time that actual demand is observed to be less than the forecasted demand upper limit. The default is set to 99.73%, with the ability to store up to 3 decimal places.

Correction Confidence Level

The percentage of the time that the actual demand is observed to be less than the forecasted demand. The default is set to 99.73%, with the ability to store up to 3 decimal places.

Comments

Comments or notes about the forecasting parameter scheme.

Max Best Fit History

Servigistics will use the maximum of this number or the stream configuration demand history horizon as the maximum number of slices of demand history data that can be used by the Best Fit algorithms. (Certain global settings may further restrict the number of available demand history slices.)

These slices are used as both the demand values from which the forecasts are calculated; and as the demand values that are compared to the best fit forecasts to calculate error for the candidate forecast methods.

Note that Winters Multiplicative and Crostons forecasts require at least 12 slices of demand history data. Assuming a typical best fit forecast window of 12 slices, Maximum # of History Slices needs to be at least two years of data for these two methods to be considered.

Minimum # of History Slices

When populated, this sets the number of slices of demand history data that will be used by the Best Fit algorithms (other than Winters, Crostons, and Double Exponential) to calculate forecasts.

When left blank, Servigistics will use 12 as the default. If less than 12 history slices are available then Servigistics will use all the slices available and generate the error "Slides<1."

Holdout Window

A sub-window in the forecast window that sets the number of slices across which demand will be compared to forecast values to calculate forecast error (MAPE, MAD, RMSE). This should be set to a period of time in which you are most interested in an accurate forecast.

The Hold Out Window equals lead time setting can be used to over ride this value to be equal to each SKU's lead time.

Best Fit Selection Criteria

Sets the forecast error measure that will be used to select the forecast method that best fits actual demand.

Options include Composite Error, Best 2 Out Of 3, MAPE (Mean Absolute Percentage Error), MAD (Mean Absolute Deviation), and RMSE (Root Mean Square Error).

Tie-breaker Error Type

Used when Best 2 Out Of 3 Best Fit selection criteria is used. It determines which forecast error measure takes precedence in the event of a tie.

Best Fit Methods

The forecast methods to be used in best fit analysis. Servigistics will use the priority of the methods to select a method if there is a tie.

Check for Intermittency

Yes indicates that Servigistics will run two tests that will check for intermittency. The tests are based on the value entered as Minimum Demand Interval and periods between demand must be > 1.25.

If intermittency is detected, Servigistics will select Crostons as the best fit method regardless of error results. If intermittency is not detected, Crostons is no longer a candidate.

Periods Between Demand

The number of periods between demands that is used to calculate intermittency. This value is calculated as slices of zero / (slices of non-zero – 1)

The value of the Periods Between Demand would be 6 / (5 – 1) = 1.5

Intermittency Test - Minimum Demand Interval MVR

The ratio to apply to the intermittency test. The ratio compares the number of periods with no demand to the number of periods with demand. The default value should normally be used.

Apply personal seasonal profile to forecast

When set to Yes, the user-specified seasonal profile will be used in Best Fit Forecasting. All methods excluding Winters Multiplicative will use deseasonalized/detrended demand in forecast error calculations.

Holdout Window equals lead time

Overrides the Holdout Window value. If the Holdout Window value is blank, this value will be adjusted automatically to equal the SKU's lead time.

Maximum number of rolling forecasts

Limits the number of rolling forecasts that will be created for each forecast method. This means it also limits the number of forecast error values that will be calculated and averaged for each method.

Setting this value between 6 and 12 may significantly reduce processing time when the forecast window is large and the hold out window is small.

Use Autocorrelation Analysis to consider Winters

Yes indicates that autocorrelation analysis will be used to consider Winters Multiplicative forecast method.

If the SKU is determined to be seasonal Winters is selected regardless of error results. If auto correlation determines the SKU is not seasonal, Winters is no longer considered as a candidate.

Use Autocorrelation Analysis to detect trend

Yes indicates that autocorrelation analysis will be used to detect trending. If a trend is detected all of the flat line forecasting methods are no longer considered as candidates.

Minimum # of History Slices to Consider Crostons

Sets the number of demand history slices that will be used to calculate the Crostons best fit forecast.

Maximum of this value and Minimum # of History Slices is compared with Stream Configuration History Slices and the minimum of the two is chosen for Best Fit Forecasting.

When left blank, Servigistics will use Minimum # of History Slices as the default.

Minimum # of History Slices to Consider Winters

Sets the number of demand history slices that will be used to calculate the Winters Multiplicative best fit forecast.

Winters Multiplicative is not considered as a candidate if this minimum is not met with the available demand history.

When left blank, Servigistics will use Minimum # of History Slices as the default.

Min # of History Slices to Consider Double Exponential Smoothing

Sets the number of demand history slices that will be used to calculate the Double Exponential best fit forecast.

Double Exponential is not considered as a candidate if this minimum is not met with the available demand history.

When left blank, Servigistics will use Minimum # of History Slices as the default.

Determine Seasonality

If Apply personal seasonal profile to forecast is NOT selected on the Best Fit tab, this field indicates whether or not to generate a seasonal profile for each SKU assigned to this parameter assuming the SKU's demand pattern is determined to be seasonal.

Note that a seasonal profile created by this setting will be overridden if a seasonal profile is assigned on the Forecasting tab.

Consider as Group?

Creates a seasonal profile based on a group of pairs.

Yes indicates that demand history for all pairs in the attached segment will be used to create a single, group seasonal profile that will be used in forecasting the pairs associated with this parameter.

Setting this to Yes implies that the pairs in the segment assigned to this parameter likely share similar seasonality demand patterns.

Output Seasonal Profile?

Yes indicates that Servigistics writes the seasonal profile data (that was generated based on the above settings) to the seasonal index and seasonal profile tables for offline analysis purposes.

This setting has no impact on the forecast itself and can negatively impact processing performance.

# of History Slices for History Analysis

The number of history slices to use to make the seasonal profile.

Output Adjusted History?

Yes indicates that the seasonal profile for each SKU will be used to de-seasonalize the SKU's demand history which is then written to the adjusted history table for offline analysis purposes.

This setting has no impact on demand used to create the forecast and can negatively impact processing performance.

Year-over-Year Trend Determination Threshold

Computation of trend is from previous year-over-year averages. This is the weighting given to last year compared to the prior year.

This value that takes the up or down slope out of seasonality.

Year-over-Year Trend - Starting Weight

Establishes the initial value of the first year's demand history.

Year-over-Year Trend - Weight Decrement

The value to decrement the coefficient by each previous year. For example, the default values of 2 and 1 give weights to the prior year-over-year at 2/3 and the previous one at 1/3. To weight them equally, use a Starting Weight of 1 and a Decrement of 0. To rank the last 3 sets of years as 1/2, 1/3, 1/6 (3/6, 2/6, 1/6), use a Starting Weight of 3 and a Decrement of 1. Finally, to rank the last three years as 5/9, 3/9, 1/9, use a Starting Weight of 5 and a Decrement of 2.

Seasonal Indices - Starting Weight

Computation of seasonal profiles is from previous year-over-year changes. This is the weighting given to last year.

Seasonal Indices - Weight Decrement

The value to decrement the coefficient by each previous year. For example, the default values of 5 and 2 give weights to the prior year of 5/8 and the previous one at 3/8. With three years of data, the values would be 5/9, 3/9, 1/9. To weight them equally, use a Starting Weight of 1 and a Decrement of 0. To rank the last 3 sets of years as 1/2, 1/3, 1/6 (3/6, 2/6, 1/6), use a Starting Weight of 3 and a Decrement of 1.