In Lewandowski model, we can add data driven events (DDEs) to the model to indicate the periods of unusual history and calculate the estimated impact of the events.
The purpose of a DDE is to explain a portion of the model error. When the system calculates the model, it determines the average amount of error. If the error for a
particular time period varies from the average error, the difference is referred to as an unexplained error. This unexplained error becomes the percent of demand
attributed to the DDE.
Data driven events can be applied to history or forecast.
When DDE is applied to future period the the starting percents are set equal to the optimized percent which is derived by running Calculate Model process.
When DDEs occur in the past, the net history quantity remains the same, but the distribution of base history and DDE type change.
Lets understand it with an example:
Say the total history was 100 and there were no events so the Base History would be 100 in this case.
If DDE (20 units) is added to history point then the Total History remains the same i.e. 100 but 20 units may be allocated to DDE and this reduces the Base History to 80.
When DDE is added in the History period, Type 3 record will be generated and if DDE is added to forecast period, Type 8 record will be generated.
If we Map Lower Level DFU to Higher Level DFU which is having DDE at lower level then after running the MapDFU process the type 3 records will be converted to type 6
If DDE is in history period and if DDE is in Forecast period then after running MapDFU process the type 8 record will be converted into type 6 records.
The issue is explained with the example:
Adding DDE to Future period:
The Lewandowski algorithm doesn’t optimize the value of DDE when it’s added to forecast because system doesn’t have actual sales data in that period.
Hence algorithm uses simple logic to calculate the quantity of DDE to be added in particular time period and in this here Impact Parameter doesn’t have any significance.
Say:
Base History is 1500
Starting Percentage: 1
Optimized Percentage: 1(Calculated after running Calc Model Process)
Impact Parameter: 1
Now DDE for period will be calculated as [Base History * Optimized Percentage] i.e. (1500)*(1/100) = 15
Example :
Impact Parameter =1
Base History = 1500
1. Starting Percentage : 0.5
Optimized Percentage: 0.5
DDE : 7.5
2. Starting Percentage : 1.0
Optimized Percentage: 1.0
DDE : 15
3. Starting Percentage : 10
Optimized Percentage: 10
DDE : 150
Adding DDE to History
The Lewandowski algorithm optimizes the value of DDE after running the Calc Model Process.
Algorithm uses complex statistical calculation to derive Optimized percentage and based on that final DDE quantity is calculated.
Scenario 1: By increasing the value of Impact parameter, considering Starting percentage as constant, the Optimized percentage will be increased so as DDE quantity calculated by running Calc Model process.
Starting Percentage : 1
1 Impact Parameter : 0.5
Optimized Percentage: 26.93
DDE : 2223.28
Total Error : -18%
2 Impact Parameter: 1.0
Optimized Percentage: 45.19
DDE : 3654.47
Total Error : -8%
3 Impact Parameter: 9.99
Optimized Percentage: 477.85
DDE : 41620
Total Error : 295%
Scenario 2 : When Starting percentage is varied the optimized percentage doesn’t have any particular pattern.
For Ex.
Impact Parameter : 1
Total History : 12375.00
1 Starting Percentage : 0.5
Optimized Percentage: -10.74
DDE : -1715.74
Total Error : 15%
2 Starting Percentage : 1.0
Optimized Percentage: -18.5
DDE : -3007.27
Total Error : 7%
3 Starting Percentage : 10
Optimized Percentage: -5.25
DDE : -825.76
Total Error : 20%
The calculation is done keeping all other parameter constant.
Here’s the calculation for DDE’s:
When we add a DDE on History, Calc Model goes through several iterations on history to find optimal impacts.
The calculation would base on the following equation:
DDEt (optimized) = DDEt + (?DDE) {(Et) / [(Mt) (St) (LEFt)]}
where:
DDE (optimized) = Optimized DDE profile
DDE = Original DDE profile
?DDE = DDE impact coefficient
M = Dynamic mean
S = Seasonality profile
LEF = Linear external factor coefficient
E = Model error
t = Time period
As Paula said, the optimization goes iteratively and, it’s really tough to calculate on an Excel or paper.
When DDE’s added on Forecast side, the calculation is (Base FCST Qty* Starting Percent)
~~~
|
