Forecast engine parameters

This table lists the input parameters that are available on the Parameters tab of the Forecast Engines screen:

Display name Description Algorithms used Accepted range

Engine

default
Boundary Constant

This represents the number of standard deviations of the historical value from the forecasted value.

Recommended values:

  • 100: If outlier detection is not required.
  • 3.5: If outlier detection is required.
 Best, H-W Positive decimal 100
Confidence Limit The confidence limit at which harmonic significance is tested and confidence intervals are calculated. Each sine and cosine of each harmonic is tested by creating a confidence interval using this limit. If the interval includes zero for either sine or cosine, the harmonic is removed from the model. BATS 0, 90, 95, 97.5, 99, or 99.5 95
Decision Criterion

This determines the decision criterion type used for testing the model which is "best". This is used in Best and BATS algorithms, and for Holt-Winters parameter optimization.

Note: AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion) are typically better than SD (Standard Deviation).

Best,

BATS, H-W

(Optimize = TRUE)

 AIC, BIC, SD AIC
Forecast Test Magnitude This represents the magnitude of forecast versus history and provides a check to assess the accuracy of the future forecast. All Positive decimal 0
History Periods Used

The maximum number of periods that is passed to the forecast engine. The default periods for monthly forecast engines is 36; The default periods for weekly forecast engines is 104.

Note: When the history periods available in the System settings, or in the individual Planning Entity is less than this value, then the number of available history periods is used.
 All Positive Integer 36. The default is monthly.
History Trend Damping Factor This represents the history trend damping factor for the Holt-Winters algorithm. This is used to dampen the trend during the model fitting process. This factor is also used to continue dampening the trend into the forecast. If there is a strong trend in the data, the trend damping factor must be close to 1. Therefore, a value of 0.95 is considered reasonable. If trend damping is not required, a value of 1 can be used. Best, H-W 0 < x <= 1 1
Initialization Type

This determines the initialization method to be used when the Algorithm field is set to H-W (Holt-Winters).

Note: This parameter is not applicable if the Algorithm field is set to Best.
H-W NOINITIALISATION, MEDIANS, AVERAGING, DECOMPOSITION, LEASTSQUARE, BACKCASTING MEDIANS
Intervention Regression Discount

This adjusts the regression coefficients of the forecast. This parameter is used in place of the routine version in specific situations, such as detecting an outlier or a step-change.

Note: 
  • This parameter value should typically be less than the standard discount factor.
  • This is a mandatory parameter.
 BATS 0 =< x < 1 0
Intervention Seasonal Discount

This adjusts the seasonal coefficients of the forecast. This parameter is used in place of the routine version in specific situations, such as detecting an outlier or a step-change.

Note: This parameter value should typically be less than the standard discount factor.
 BATS 0 =< x < 1 0
Intervention Trend Discount

This adjusts the trend elements of the forecast (level and growth). This parameter is used in place of the routine version in specific situations, such as detecting an outlier or a step-change.

Note: This parameter value should typically be less than the standard discount factor.
 BATS 0 =< x < 1 0
Intervention Variance Discount

This adjusts the variance estimate of the forecast. This parameter is used in place of the routine version in specific situations, such as detecting an outlier or a step-change.

Note: This parameter value should typically be less than the standard discount factor.
 BATS 0 =< x < 1 0
Exponentially weighted moving average Lambda

This represents the weighting factor used by the exponential moving average algorithm. This parameter is used to increase or decrease the influence of the new observation on the moving average value.

Note: 
  • If the coefficient value is 0, the new observation value does not impact the moving average value.
  • If the coefficient value is 1, the new observation value becomes the moving average value.
  • The ideal parameter value range is 0.05 <= x <= 0.25, or 0.2 <= x <= 0.3.
 BEST, EWMA 0 =< x <= 1 0.2
Least Squares Outlier Standard Deviation

The minimum number of standard deviations required for a data point to be reported as an outlier.

Note: 
  • This is a monitoring parameter.
  • If the value is 0, the outlier detection is disabled.
  • It is recommended to specify lower values such as 2, as outliers are expected to have significantly large errors. This is due to the Least Squares model including outlier observations in the initial calculation and is therefore relatively insensitive to outliers.
 BEST, LS Integer 2
Level Inflation Factor The level inflation factor used in BATS monitoring. This parameter is applied to the current standard deviation estimate, to create two alternative models (one higher and one lower) to compare with the current model. BATS 0 to 99.9 3.5
Level inflation Threshold The level inflation threshold used in BATS monitoring. This parameter indicates the minimum value of the Bayes factor for level shift before a tracking signal is activated. An outlier is declared if the Bayes factor value is less than this parameter value in an individual period. If the Bayes factor value is less than this parameter value in cumulative periods, then the tracking signal and process are activated. BATS 0 < x <= 1 1
Level Smoothing Coefficient

This represents the level smoothing coefficient for the Holt-Winters algorithm. This parameter is used to increase or decrease the influence (such as weighting and rate of decay) of the new observation on the level component of the Holt-Winters model.

Note: 
  • If the coefficient value is 0, the new observation does not have impact on the level of the model.
  • If the coefficient value is 1, the new observation becomes the level (after removing seasonal influence where applicable).
  • If the Optimize Parameters are enabled, the engine applies various values based on optimization parameters and use the values that produce the model with the lowest decision criteria.
 Best, H-W (Optimization = FALSE) 0 =< x <= 1 0
Maximum Model Terms

This is the maximum number of terms in the generated model.

Note: 
  • If Period Level is set to Months, it includes level, trend and 6 harmonics (6 pairs of a sine and cosine).
  • If Period Level is set to Weeks, it includes level, trend and 26 harmonics (26 pairs of a sine and cosine).
  • This parameter does not include regressors.
 BATS 1 to 54 14
Minimum Events Length This controls the minimum length of history that does not have a historical event held against them to process Events. The non-event history does not have to be contiguous. This minimum length is required so that there is enough baseline (non-event) history for fitting a baseline statistical model, using the LeastSquares method. This parameter is applicable when the Event Modeling Method field is set to LeastSquares. If the minimum non-event history periods requirement is not met, events for that item-location combination are not considered. The forecasts are generated by assuming the Event Modeling Method field value is set to None, and a warning is displayed. All positive integer (>0) 3
Model Form

This determines the type of model to use such as constant (level only), linear (level and trend), constant seasonal (level and seasonality), seasonal (level, trend and seasonality).

This parameter should work in conjunction with the Seasonal type selected.

Note: 
  • This is a mandatory parameter for all algorithms except Events Only.
  • Constant seasonal is only applicable to the BATS algorithm.
 All, except EVO NOTSET, CONSTANT, LINEAR, CONSTANT_SEASONAL, SEASONAL NOTSET
Moving Average Outlier Standard Deviation

This is the minimum number of standard deviations required for a data point to be detected as an outlier.

Note: 
  • This is a monitoring parameter.
  • If this parameter value is 0, the outlier detection is disabled.
  • For outlier testing, you should specify low values such as 2, as outliers are expected to have significantly large errors. This is as a result of Moving Average model incorporating outlier observations in the initial calculation and therefore being relatively insensitive to outliers.
BEST, MA, EWMA Integer 2
Moving Average Point Value This is the number of data points used in each moving point calculation. This is a simple moving average and is not a centered moving average. EWMA, MA Positive integer 4
Optimization Increment

This is the increment of the relevant Holt-Winters smoothing coefficient during an iteration of the optimization algorithm.

Note: 
  • This is a Late Acceptance Hill Climbing (LAHC) optimization heuristic parameter.
  • If this parameter value is reduced towards 0, the number of iterations increases, which will reduce performance.
 BEST, BATS, H-W 0 =< x <= 1 0.1
Optimization Percentage

This determines the number of candidate solutions that must be calculated consecutively where the solution is not saved in order to end the optimization process. The value is the percentage of the total number of runs that can occur based on the optimization increment and the model form.

Note: 
  • This is a Late Acceptance Hill Climbing (LAHC) optimization heuristic parameter.
  • If this parameter value is 1, the run performs the total number of runs that can occur based on the optimization increment and the model form.
  • If this parameter value is 0, a single run is performed which uses the optimization starting values for the relevant smoothing coefficients.
  • As this parameter value approaches 1, the length of the run increases and this reduces performance. However, the chances of reaching the global minimum of the MLR run, that is, the best smoothing parameters increases.
BEST, BATS, H-W 0 =< x <= 1 0.1
Optimization Starting Value Alpha

This is the starting value for the level smoothing coefficient for the Holt-Winters algorithm.

Note: This is a Late Acceptance Hill Climbing (LAHC) optimization heuristic parameter.
 BEST, BATS, H-W 0 =< x <= 1 0
Optimization Starting Value Beta

This is the starting value for the trend smoothing coefficient for the Holt-Winters algorithm.

Note: This is a Late Acceptance Hill Climbing (LAHC) optimization heuristic parameter.
 BEST, BATS, H-W 0 =< x <= 1 0
Optimization Starting Value Gamma

This is the starting value for the seasonal smoothing coefficient for the Holt-Winters algorithm.

Note: This is a Late Acceptance Hill Climbing (LAHC) optimization heuristic parameter.
 BEST, BATS, H-W 0 =< x <= 1 0
Outlier Exception Horizon
This determines the history horizon in the Outlier Exception test at the period level of the calculated forecast. For example, the number of months if the Period Level field is set to Months. Outliers are still set for the full history, however, an outlier can be reported as a static alert (at pconst), if there is an outlier within the most recent periods covering the exception horizon. If not specified, there is no limit.
Note: This is a monitoring parameter.
 BATS, BEST, EWMA, LS, MA Positive Integer Null
Perform Outlier Testing

This determines whether checks are performed for outliers.

Note: This is a monitoring parameter.
 BATS, BEST, EWMA, H-W, LS, MA True or False True
Perform Step Changes

This determines whether checks are performed for step changes.

Note: This is a monitoring parameter.
 BATS, BEST, H-W True or False True
Perform Tracking

This determines whether checks are performed for tracking signals.

Note: This is a monitoring parameter.
 BATS, BEST, EWMA, H-W, LS, MA True or False True
Periodicity This is the periodicity of the passed history data and mask information. This defines the recurring pattern of the data required for predicting seasonal influences. For example, Months is set to 12, or 13, Weeks is set to 52, and Quarters is set to 4. Best, ARIMA, BATS, H-W, NA

ARIMA, Holt-Winters, Naive (seasonal) algorithms: 4 (quarterly data); 12, 13 (monthly data), 52 (weekly data); 7, 365 (daily data). BATS algorithm: 4, 7, 12, 13, 52.

For all other algorithms 0 is valid.

 0
Routine Regression Discount

This is used to increase or decrease the influence, such as, weighting, rate of decay of the new observation on the regression component of the BATS model.

Note: 
  • As the discount factor value tends to 1, the model is more dynamic.
  • As the discount factor value tends to 0, the model is less dynamic.
 BATS 0 =< x < 1 0
Routine Seasonal Discount

This is used to increase or decrease the influence such as weighting, rate of decay, of the new observation on the seasonal component of the BATS model.

Note: 
  • As the discount factor value tends to 1, the model is more dynamic.
  • As the discount factor value tends to 0, the model is less dynamic.
 BATS 0 =< x < 1 0
Routine Trend Discount

This is used to increase or decrease the influence, such as weighting, rate of decay, of the new observation on the trend component of the BATS model.

Note: 
  • As the discount factor value tends to 1, the model is more dynamic.
  • As the discount factor value tends to 0, the model is less dynamic.
BATS 0 =< x < 1 0
Routine Variance Discount

This is used to increase or decrease the influence, such as weighting, rate of decay, of the new observation on the variance component of the BATS model.

Note: 
  • As the discount factor value tends to 1, the model is more dynamic.
  • As the discount factor value tends to 0, the model is less dynamic.
 BATS 0 =< x < 1 0
Step Change Exception Horizon This determines the history horizon in the Step Change Exception test at the period level of the calculated forecast. For example, the number of months if Period Level is set to Months. Step Changes are still set for the full history, however only if there is a step change within the most recent periods covering the exception horizon will it be flagged as a static alert (at pconst). If not specified, there is no limit.
Note: This is a monitoring parameter.
 Best, BATS, H-W Positive Integer Null
Step Change Maximum Run Length

This determines the maximum number of continuous outliers required to trigger a step change in the Holt-Winters algorithm. The outliers must all be in the same direction.

Note: This is a monitoring parameter.
 Best, H-W Integer 3
Tracking Control Limit

This determines the control limit for tracking where tracking is enabled. If this parameter value is set to 0.99, the tracking signal detection is disabled

Note: This is a monitoring parameter.
 BEST, EWMA, H-W, LS, MA 0.8, 0.9, 0.95, 0.96, 0.97, 0.98, 0.99 0.99
Tracking Exception Horizon This determines the history horizon in the Tracking Exception test at the period level of the calculated forecast. For example, the number of months if the Period Level field is set to Months. Tracking Signals are still set for the full history, however only if there is a tracking signal within the most recent periods covering the exception horizon will it be flagged as a static alert (at pconst).
Note: This is a monitoring parameter.
 Best, BATS, EWMA, H-W, LS, MA Positive Integer Null
Tracking Type

This determines the tracking signal strategy to be used if tracking is enabled.

Note: This parameter is a monitoring parameter.
 Best, EWMA, H-W, LS, MA NOTRACKING, SIMPLETRACKINGSIGNAL, BROWNSCUSUMTRACKINGSIGNAL, TRIGGSSMOOTHEDTRACKINGSIGNAL, AUTOCORRELATIONTRACKINGSIGNAL NOTRACKING
Recent History This indicates the number of recent history periods to verify for sales activity. If the items have no sales history for the specified number of periods specified in the parameter, the engine does not generate a forecast for those periods and a recent History Exception is reported by the engine. All Integer >= 0, based on Periodicity Default settings are not specified in the engine.
Run Length Limit

This is the run length limit used in BATS monitoring. It defines the number of periods monitored for significant shifts in the level and/or variance. This is the maximum length a tracking signal can run before being reported if this is not reported earlier. When a tracking signal is detected, the item or location combination is recorded in Tracking Signal Exception, where mapped.

Note: This is a monitoring parameter.
 Best, BATS Positive integer 4
Scale Inflation Factor

This is the scale inflation factor used in BATS monitoring. This factor is used to develop an alternative model to monitor the increase in the variance estimate.

Note: This is a monitoring parameter.
 Best, BATS 0.001 to 100 20
Scale inflation threshold

This is the scale inflation threshold used in BATS monitoring. This is a minimum value that the variance Bayes factor can be before a tracking signal is activated. If the Bayes factor falls below this threshold, the tracking signal process is activated.

Note: This is a monitoring parameter.
 Best, BATS 0 < x <= 1 0.01
Scale Smoothing Coefficient

This is the scale smoothing coefficient for the Robust Holt-Winters algorithm. This is used to increase or decrease the influence of the forecast error on the scale component of the Robust Holt-Winters model.

Note: 
  • If the coefficient value is 0, the forecast error does not impact the scale of the model.
  • If the coefficient value is 1, the forecast error has a significant impact on the scale of the model, making it more unstable.
  • We recommend setting the parameter value to 0.1 and not modifying it.
Best, H-W 0 < x <= 1 0.1
Seasonal Smoothing Coefficient

This is the seasonal smoothing coefficient for the Holt-Winters algorithm. This is used to increase or decrease the influence of the new observation on the relevant seasonal index of the Holt-Winters model.

Note: 
  • If the coefficient value is 0, the new observation does not impact the relevant seasonal index of the model.
  • If the coefficient value is 1, the new observation becomes the relevant seasonal index (after the influence of the level has been removed).
  • If the Optimize Parameters switch is enabled, the engine applies several values based on optimization parameters and uses the values that produce the model with the lowest decision criteria.
 Best, H-W (Optimize = FALSE) 0 =< x < 1 0
Seasonal Type

This determines the type of seasonality to use for Holt-Winters and BATS models. For example, additive, or multiplicative.

Note: This is a mandatory parameter.
Best, BATS, H-W NONE, ADDITIVE, MULTIPLICATIVE NONE
Short History

This determines the minimum length of the history required for model fit to be performed. The history length must be equal to or greater than the Short History parameter specified.

Note: 
  • This is a mandatory parameter.
  • User defined masks are removed before the test.
 All, except EVO Integer > 0 0
Minimum periods for SMP

This determines the minimum number of non-zero history values for an item to be considered as a non-SMP (slow moving product, product of intermittent, or sparse demand).

Note: User defined masks are removed before the test.
 All, except EVO Integer >= 0 0
SMP Gap Discount Factor This is the discount factor used to smooth the level in BATS specifically for SMPs. This is used to smooth the estimate of the current length of time between sales events. BATS 0 =< x < 1 0.2
Trend Smoothing Coefficient

This is the trend smoothing coefficient for the Holt-Winters algorithm. This is used to increase or decrease the influence of the level on the trend component of the Holt-Winters model.

Note: 
  • If the coefficient value is 0, the difference between the new and old levels does not impact the trend of the model.
  • If the coefficient value is 1, the difference between the new and old levels becomes the trend.
  • If the Optimize Parameters switch is enabled, the engine applies several values based on optimization parameters and use the values that produce the model with the lowest decision criteria.
 Best, H-W (Optimize = FALSE) 0 < x <= 1 0
Trim Factor

This is the number of extreme values to remove when running the Trimmed Mean technique of Forecast Combining. This is a one-sided value, which means, when removing the first and last values, x = 1.

Note: This parameter is applicable, only if the Forecast Combining switch is enabled.
Best where Classic Combining = True Positive Integer 1
Winsorizing Factor

This determines the number of extreme values to replace and is used in the Winsorized Mean technique. This is a one-sided value, which means, when removing the first and last values, x = 1.

Note: This parameter is applicable only if the Forecast Combining switch is enabled.
 Best where Classic Combining = True Positive Integer 1
Weighted AIC Maximum

This is the maximum difference from the lowest AIC value for a technique to be included in the weighting calculation. This is used in the Weighted AIC technique.

Note: This parameter is applicable only if the Forecast Combining switch is enabled.
 Best where Classic Combining = True Positive Decimal 4
Learning Rate

This determines the number of errors (online forecast – observation) rebuilt into the model, in the calculation of the model coefficients.

Note: This parameter is applicable only if the Forecast Combining switch is enabled.
 Best where any of ML Train-Test, ML Train-Test randomization, ML Cross-Validation, ML full dataset or ML full dataset randomization = True Positive Decimal 0.001
Epoch

This is the number of loops of the whole data set that the model performs, to calculate the model coefficients. The higher the value for Epoch, the more accurate the model but with decreased performance.

Note: This parameter is applicable only if the Forecast Combining switch is enabled.
 Best where any of ML Train-Test, ML Train-Test randomization, ML Cross-Validation, ML full dataset or ML full dataset randomization = True Positive Integer 50
Folds

This is the number of folds used in setting up cross-validation data sets.

Note: This parameter is applicable only if the Forecast Combining switch is enabled.
 Best where Machine Learning Cross-Validation = True Positive Integer 5
Random Number Seed

This determines whether to perform randomization.

Note: 
  • If this parameter value is < 0, randomization is not performed.
  • If this parameter value is = 0, non-reproducible randomization is performed. The random order is different every time when randomization is performed.
  • If this parameter value is > 0, reproducible randomization is performed. If the number is the same, the random order is the same, therefore, results are reproducible.
  • This parameter is applicable, only if the Forecast Combining switch is enabled.
 Best where one of ML Train-Test randomization, ML Cross-Validation or ML full dataset randomization = True Integer -1
MLR Trim Factor

This deletes the highest “x” and lowest “x” forecast in each time period prior to the combining process. This is a one-sided value, that means, when removing the first and last values, x = 1.

Note: This parameter is applicable only if the Forecast Combining switch is enabled.
 Best where any of ML Train-Test, ML Train-Test randomization, ML Cross-Validation, ML full dataset or ML full dataset randomization = True Positive Integer 0
Allow Negative Forecast

This determines whether the Forecast Engine generates negative values as Forecast.

Note: 
  • If this parameter value is set to 1, the forecast engine can generate negative forecast values.
  • If this parameter value is set to 0, the forecast engine generates 0 for the values calculated to be negative.
 All 0 or 1 1