K Configuring the Forecast150 and CausalEstimate

This appendix describes how to configure the Forecast150 procedure that generates the forecast and estimates the promotion effects.

About the Forecast and the CausalEstimate Procedures

Previously RDF used only the Forecast procedure to generate both the forecast and to estimate the promotion effects. Now the Forecast procedure is divided into two procedures:

Forecast150 procedure — generates the forecast for all forecast methods

CausalEstimate procedure — calculates the promotion effects at both final or source level.

The Forecast150 Procedure Syntax section contains the specifications and syntax for configuring the Forecast150 procedure.

Forecast150 Procedure

The following sections detail the Forecast150 procedure for generating the forecast for all forecast methods.

Forecast150 Requirements

The following libraries must be registered in any domains that will use the Forecast150 solution extension:

AppFunctions
RdfFunctions

Forecast150 Parameter/Model Dependencies

The following models require that the stated measure is to be provided.

Bayesian model — Plan measure required
Profile model — Profile measure required
Causal Model — The estimated promotion effects and causal baseline

Using the Forecast150 Procedure

The following notes are intended to serve as a guide for configuring the Forecast150 procedure within the RPAS Configuration Tools:

Note:

The forecast output from the Forecast150 clears out the elapsed weeks results. In other words, after running this special expression, the forecast measure is only populated for the periods from forecast start to forecast end date. You must customize the RPAS rule to save the forecast value.

Refer to the appropriate input parameters and output measures when using the Forecast150 procedure.
The resultant measure (frcstout) should be at the same intersection as your history measure (pos). This will be the base intersection of the final level.
The Forecast150 procedure is a multi-result procedure, meaning that it can return multiple results with one procedure call within a rule. In order to get multiple results, the resultant measures must be configured in the Measure Tool and the specific measure label must be used on the left-hand-side (LHS) of the procedure call. The resultant measure parameters must be comma-separated in the procedural call.
The startdatemeas that specifies the forecast start date needs to be periodically updated (every week or so) by configuring rules.
The forecast methods are specified using the mask measure. This is an int measure. Refer to Table K-3, "Forecast Method/Mode List" for the expected values of this measure for each forecast method.

Forecast150 Procedure Syntax

The syntax for using the Forecast150 procedure is shown in Example K-1.

Example K-1 Generic Example

FORECAST: FORMEAS [, INT: INTMEAS, CUMINT:CUMINTMEAS, PEAKS:PEAKSMEAS, CHMETHOD:METHMEAS, CHLEVEL:LVLMEAS, CHTREND:TRENDMEAS, ALERTS:ALERTSMEAS] <-FORECAST(MASK:MEASKMEAS, FORCSTSTART:STARTDATE, FORECASTLENGTH:FORECASTLENGTH, HISTORY:HISTORYMEAS, PERIOD:PERIOD [
[,{PROMO_0:PROMO0,PROMOEFF_0:PROMOEFF0,PROMOEFFTYPE_0:PROMOEFFTYPE0, } … {,PROMO_N:PROMON,PROMOEFF_N:PROMOEFFN,PROMOEFFTYPE_N:PROMOEFFTYPE_N, PROMOEFFTYPE_N: PROMOEFFTYPEN, } POVLPEFF:OVERLAPEFFS,POVLPCOM:OVERLAPCOMBIN, PROMOOVERLAPMTHD:OVERLAPMETHOD,PROMOPVALUE:OVERLAPPVALUE,PEFFMETHOD:PROMOEFFMETHOD],
HISTSTART: HISTSTARTMEAS,MINWINTERS:MINWINTERSMEAS, MINHOLT: MINHOLTMEAS, MINCROSTON:MINCROSTON, MAXALPHA:MAXALPHA, MAXWINTERSALPHA:MAXWINALPHA, MAXPROFILEALPHA:MAXPROFILEALPHA, BAYESALPHA:BAYESALPHA, TRENDDAMP:TRENDDAMP, {VALID_DD:VALID_DD, DDPROFILE:DDPROFILE }, PLAN:PLAN, PROFILE:PROFILE, AGGPROFILE:AGGPROF, SPREADPROFILE:SPREADPROF, BAYESIAN_HORIZ,BAYESIAN_HORIZ,  CAPS:CAPSMEAS, CAPRATIOS:CAPRATIOSMEAS, USECAPPING:USECAPPING, MINCAPHIST:MINCAPHIST, PLANINT:PLANINTMEAS, PLANCUMINT:PLANCUMINTMEAS, CAPINTERVALS:CAPINTERVALS]

Configuration Parameters and Rules

The Table K-1, "Input Parameters for the Forecast150 Procedure" and Table K-2, "Output Parameters for the Forecast 150 Procedure" explain the specific usage of the parameters names used in the procedure. Table K-3, "Forecast Method/Mode List" provides the expected values of the mask measure for each forecast method.

Table K-1 Input Parameters for the Forecast150 Procedure

Parameter Name	Description
aggprofile	The celendar aggregation profile. Data Type: real Multiple Allowed: No Required: No
baycapratio	The Bayesian Cap Ratio Data Type: Real Multiple Allowed: No Required: No
bayesalpha	The maximum Bayesian alpha value. Data Type: Real Multiple Allowed: No Required: No
bayesian_horiz	The horizon to which the Bayesian adjust is applied. Data Type: Integer Multiple Allowed: No Required: No
capintervals	Indicator whether cap the interval Data Type: Real Multiple Allowed: No Required: No
capratios	Cap ratio for each time series. Data Type: Boolean Multiple Allowed: No Required: No
caps	Caps for each time series. Data Type: Real Multiple Allowed: No Required: No
causalextbaseline	Baseline for Causal Forecast Data Type: Real Multiple Allowed: No Required: No
ddprofile	De-seasonalized demand measure used only for profile-based forecasting. Data Type: Real Multiple Allowed: No Required: No
extraweekmth	Extra Week Process Method Data Type: Integer Multiple Allowed: No Required: No
extraweeksrc	Extra Week Data Source Data Type: Boolean Multiple Allowed: No Required: No
fallbackmth	Fall Back Forecast Method Data Type: Integer Multiple Allowed: No Required: No
forecastlength	The length of the forecast. Data Type: Integer Multiple Allowed: No Required: Yes
frcststart	The forecast start date. Data Type: Datetime or String Multiple Allowed: No Required: No
history	The input measure the forecast is based on. Data Type: Real Multiple Allowed: No Required: Yes
histstart	The historical start date Index. Data Type: Integer Multiple Allowed: No Required: No
intcapratiolower	Lower Interval Cap Ratio Data Type: Real Multiple Allowed: No Required: Yes
intcapratioupper	Upper Interval Cap Ratio Data Type: Real Multiple Allowed: No Required: Yes
mask	Array that identifies what forecast method is used for each time series. Refer to Table K-4, "Numeric Values Assigned to the Forecast Model/Model List". Data Type: Integer Multiple Allowed: No Required: Yes
maxalpha	The maximum alpha value. Data Type: Real Multiple Allowed: No Required: No
maxholtgamma	The maximum gamma value for Holt Method. Data Type: Real Multiple Allowed: No Required: No
maxprofilealha	The maximum Alpha value for Profile Method. Data Type: Real Multiple Allowed: No Required: No
maxwintersalpha	The maximum Alpha value for Winter Method. Data Type: Real Multiple Allowed: No Required: No
maxwintersgamma	The maximum Gamma value for Winter Method. Data Type: Real Multiple Allowed: No Required: No
mincaphist	The minimum number of weeks before capping can be used. Data Type: Real Multiple Allowed: No Required: No
mincroston	The minimum Croston history. Data Type: Integer Multiple Allowed: No Required: No
minholt	The minimum Holt history. Data Type: Integer Multiple Allowed: No Required: No
minwinters	The minimum Winters history. Data Type: Integer Multiple Allowed: No Required: No
movingaveragewindowlength	The moving average window of Average method. Data Type: Integer Multiple Allowed: No Required: No
peffmethod	The promotion effect method Data Type: Integer Multiple Allowed: No Required: No
period	The forecasting period for calculating seasonal coefficients. Data Type: Integer Multiple Allowed: No Required: Yes
plan	The Plan measure. This measure's intersection may not be higher than the intersection of a corresponding forecast source level. Data Type: Real Multiple Allowed: No Required: No
plancumint	The cumulative Interval of the plan associated with the plan (PARAMETER forecast); Bayesian only. Data Type: Real Multiple Allowed: No Required: No
planint	The interval of the plan associated with the plan (PARAMETER forecast); Bayesian only. Data Type: Real Multiple Allowed: No Required: No
povlpcom	The overlapping promotion combination Data Type: String Multiple Allowed: No Required: No
povlpeff	The overlapping promotion combined effects Data Type: Real Multiple Allowed: No Required: No
profile	The Seasonal Profile measure. Data Type: Real Multiple Allowed: No Required: No
promo_	The Promo variable measure (one for each promotion). Data Type: Integer Multiple Allowed: Yes Required: No
promoeff_	The calculated promotional effects (one per promotion). Data Type: Real Multiple Allowed: Yes Required: No
promoefftype_	The calculated promotional effects type (Linear or Exponential) Data Type: Integer Multiple Allowed: Yes Required: No
promooverlapmthd	The overlapping promotion method Data Type: Integer Multiple Allowed: No Required: No
promopvalue	The overlapping adjust factor Data Type: Real Multiple Allowed: No Required: No
seasonalindexsmooth	Seasonal Index Smooth factor in AWinter Data Type: Real Multiple Allowed: No Required: No
spreadprofile	The profile to spread to final forecast level. Data Type: Real Multiple Allowed: No Required: No
trenddamp	The trend damping parameter. Data Type: Real Multiple Allowed: No Required: No
usecapping	A Boolean measure that indicates whether capping is applied. Data Type: Boolean Multiple Allowed: No Required: No
valid_dd	The maximum non-zero history to use de-seasonalized demand value for seasonal profile based forecasting. Data Type: Integer Multiple Allowed: No Required: No
wintersmode	Winter Mode Data Type: Integer Multiple Allowed: No Required: No

Table K-2 Output Parameters for the Forecast 150 Procedure

Parameter Name	Description
forecast	Forecast output. Data Type: Real Multiple Allowed: No Required: Yes
ape	Forecast APE output. Data Type: Real Multiple Allowed: No Required: Yes
std	Forecast STD output. Data Type: Real Multiple Allowed: No Required: Yes
peaks	Peaks, which are used for calculating baseline of the forecast. Data Type: Real Multiple Allowed: No Required: No
chmethod	Selected method. Refer to Table K-4, "Numeric Values Assigned to the Forecast Model/Model List". Data Type: Integer Multiple Allowed: No Required: No
chlevel	ES level. Data Type: Integer Multiple Allowed: No Required: No
chtrend	ES trend. Data Type: Real Multiple Allowed: No Required: No
chalpha	ES alpha. Data Type: Real Multiple Allowed: No Required: No
chgamma	ES Gamma. Data Type: Real Multiple Allowed: No Required: No
alerts	A high-level forecast alert generated by the forecast engine. Data Type: Boolean Multiple Allowed: No Required: No
povlpflag	The overlapping indicator in Forecast Data Type: Boolean Multiple Allowed: No Required: No

Table K-3 Forecast Method/Mode List

Model	Numeric Value
AUTO ES	1
SIMPLE	2
HOLT	3
WINTERS	4
CASUAL	5
AVERAGE	6
NO FORECAST	7
COPY	8
CROSTON	9
M. WINTERS	10
A. WINTERS	11
SIMPLE CROSTON	12
BAYESIAN	13
LOADPLAN	14
PROFILE	15
MOVING AVERAGE	17
COMPONENTS	19

Forecast Method/Model List

Table K-4 provides the numeric value assigned to the forecast model/model list.

Table K-4 Numeric Values Assigned to the Forecast Model/Model List

Model	Numeric Value
AUTO ES	1
SIMPLE	2
HOLT	3
WINTERS	4
CASUAL	5
AVERAGE	6
NO FORECAST	7
COPY	8
CROSTON	9
M. WINTERS	10
A. WINTERS	11
SIMPLE CROSTON	12
BAYESIAN	13
LOADPLAN	14
PROFILE	15
MOVING AVERAGE	17

CausalEstimate Procedure

The following sections detail the CausalEstimate procedure which estimates the promotion effects at both final level and source level. The final level is usually item/store/calendar, the source level is higher than the final level. While estimate the promotion effects at source level, RDF pools all the data points at the final level, then estimate the promotion effects.

CausalEstimation Procedure Syntax

The syntax for using the CausalEstimation procedure is shown in Example K-2.

Example K-2 CausalEstimation Procedure Syntax

POVLPCOM:POVLPCOM,POVLPEFF:POVLPEFF,PEFFMETHUSED:PEFFMETHUSED,PROMOEFF_0:PROMOEFF_0,{PROMOEFF_N:PROMOEFF_N},APE:APE,STD:STD<-CausalEstsimation(HISTORY:HISTORY,HISTSTART:HISTSTART,HISTEND:HISTEND,MASK:MASK,POOLINGELIGIBLEMASK:POOLINGELIGIBLEMASK,MAXB:MAXB,MINB:MINB,KEEPCLAMPEDMAXB:KEEPCLAMPEDMAXB,AGGPROF:AGGPROF,PROMOOVERLAPMTHD:PROMOOVERLAPMTHD,ITEMGROUP:ITEMGROUP,NUMOFEXTRAPOINTS:NUMOFEXTRAPOINTS,CAUSALDATASRCTHR:CAUSALDATASRCTHR,PROMO_0:PROMO_0,PROMOTYPE_0:PROMOTYPE_0,PROMOEFFTYPE_0:PROMOEFFTYPE_0,PROMONEG_0:PROMONEG_0,PROMOOVER_0:PROMOOVER_0,{PROMO_N:PROMO_N,PROMOTYPE_N:PROMOTYPE_N,PROMOEFFTYPE_N:PROMOEFFTYPE_N,PROMONEG_N:PROMONEG_N,PROMOOVER_N:PROMOOVER_N})

Configuration Parameters and Rules

The Table K-5, "Input Parameters for CausalEstimation" and Table K-6, "Output Parameters for CausalEstimation" explain the specific usage of the parameters names used in the CausalEstimation procedure.

Table K-5 Input Parameters for CausalEstimation

Parameter Name	Description
history	The input measure the forecast is based on. Data Type: Real Multiple Allowed: No Required: Yes
histstart	The historical start date. Data Type: DateTime Multiple Allowed: No Required: Yes
histend	The historical enddate. Data Type: DateTime Multiple Allowed: No Required: Yes
mask	Run Mask Indicator Data Type: Boolean Multiple Allowed: No Required: Yes
poolingeligiblemask	If item/store eligible mask for pooling Data Type: Boolean Multiple Allowed: No Required: Yes
maxb	The maximum ratio between beta and baseline. Data Type: Real Multiple Allowed: No Required: No
minb	The minimum ratio between beta and baseline. Data Type: Real Multiple Allowed: No Required: No
keepclampedmaxb	Determines whether variables exceeding minb are clamped or values are dropped and regression is re-run. Data Type: Real Multiple Allowed: No Required: No
aggprof	Aggregation profile from the low calendar to higher lelvel. Data Type: Real Multiple Allowed: No Required: No
promooverlapmthd	The Promotion Overlapping method to deal with overlapping promotion. Data Type: Integer Multiple Allowed: No Required: No
itemgroup	The item/store to group mapping Data Type: boolean Multiple Allowed: No Required: No
numofextrapoints	The number of extra data points before/after promotion periods Data Type: Integer Multiple Allowed: No Required: No
causaldatasrcthr	The data dales value threshold for promotion variabble. Data Type: Real Multiple Allowed: No Required: No
promo_	The Promo variable measure (one for each promotion). Data Type: Integer Multiple Allowed: Yes Required: No
promotype_	The promo type measure (one for each promotion). Data Type: Integer Multiple Allowed: Yes Required: No
promoefftype_	The calculated promotional effects type (Linear or Exponential) Data Type: Integer Multiple Allowed: Yes Required: No
promoneg_	The indicator whether the negative promo effects allow. Data Type: Boolean Multiple Allowed: Yes Required: No
promoover_	The promo effect override measure (one for each promotion). Data Type: Boolean Multiple Allowed: Yes Required: No

Table K-6 Output Parameters for CausalEstimation

Parameter Name	Description
povlpcom	The overlapping promotion combination Data Type: String Multiple Allowed: No Required: No
povlpeff	The overlapping promotion combined effects Data Type: Real Multiple Allowed: No Required: No
peffmethused	The promotion effect method Data Type: Integer Multiple Allowed: No Required: No
promoeff_	The calculated promotional effects (one per promotion). Data Type: Real Multiple Allowed: Yes Required: No
ape	Forecast APE output. Data Type: Real Multiple Allowed: No Required: Yes
std	Forecast STD output. Data Type: Real Multiple Allowed: No Required: Yes