In this week’s What-if Wednesday post, Arkieva Supply Chain Optimization Consultant – Abhishek Shah – shares the results of a what-if demand forecast simulation using the gamma multiplicative forecast parameter of the Winters forecasting method.

forecasting methods gamma multiplicative Winters

In the previous blog, we experimented with Beta – the trend parameter of Holt-Winters Multiplicative method – to see how the forecast gets affected as the weight changes from 0.05 to 0.8.

This week we are going to run similar tests with Gamma, the seasonal parameter. We will use the same historical data set for our analysis. Similar to Alpha and Beta, Gamma values range from 0 to 1; and larger Gamma means more weight on recent historical seasonality.

For our analysis, we will keep the alpha and beta values at 0.05 each and will vary the Gamma values from 0.05 to 0.8 to evaluate how the forecast changes. You can see the numbers at the end of the post.

The first chart below shows historically observed values from the last three years and the subsequent graphs represent forecasts using a Gamma value of 0.05, 0.1, 0.2, 0.4, and 0.8.

Statistical Forecasting Method: Holt-Winters Multiplicative Gamma Parameter Simulation Results

The overall behavior of the forecast with changing gamma is similar to what we saw with the additive method in the previous blog. There is little change in the seasonality of the forecast when Gamma values range from 0.05 to 0.4 but we notice that there is an uptick in the month of Mar’17 and Aug’17 as Gamma goes up to 0.8 which clearly indicates the influence of last one-year history (Mar’16). Also, just like the alpha and beta values in the multiplicative method, the seasonal changes, as we change gamma parameter from 0.05 to 0.8, are proportional to the level and not constant. You can compare the forecast from the previous method here.

multiplicative forecast parameter simulation gamma

[ Read Also: What-if Wednesday: Forecast Parameter Simulation – Beta Multiplicative Method ]

 

What-if Scenario Forecast Numbers

Month History   Month ForecastG.05 ForecastG.1 ForecastG.2 ForecastG.4 ForecastG.8
Dec-13 62,719   Dec-16 41,248 41,326 41,462 41,789 43,721
Jan-14 60,312   Jan-17 39,915 40,318 41,014 42,065 43,612
Feb-14 38,119   Feb-17 37,090 36,554 35,774 35,195 36,506
Mar-14 51,926   Mar-17 41,412 41,927 43,009 45,346 50,492
Apr-14 64,856   Apr-17 43,968 44,038 44,132 44,227 44,827
May-14 67,589   May-17 43,173 43,292 43,325 42,689 39,633
Jun-14 57,151   Jun-17 41,235 41,089 40,847 40,557 40,791
Jul-14 53,502   Jul-17 41,233 41,417 41,747 42,192 41,778
Aug-14 47,696   Aug-17 41,134 41,181 41,425 42,389 45,259
Sep-14 49,189   Sep-17 42,343 42,458 42,842 44,109 47,811
Oct-14 62,103   Oct-17 43,111 43,498 44,186 45,268 46,822
Nov-14 45,289   Nov-17 40,362 40,228 40,077 40,108 40,530
Dec-14 38,105   Dec-17 39,785 39,872 40,033 40,414 42,397
Jan-15 45,149   Jan-18 38,784 39,164 39,830 40,865 42,477
Feb-15 38,361   Feb-18 36,398 35,922 35,233 34,749 36,061
Mar-15 46,655   Mar-18 40,473 40,953 41,967 44,174 49,117
Apr-15 44,090   Apr-18 42,916 42,994 43,112 43,270 43,992
May-15 51,350   May-18 42,303 42,424 42,484 41,968 39,291
Jun-15 40,574   Jun-18 40,641 40,519 40,323 40,110 40,441
Jul-15 52,236   Jul-18 40,717 40,894 41,217 41,675 41,420
Aug-15 44,306   Aug-18 40,695 40,747 40,989 41,917 44,685
Sep-15 43,866   Sep-18 41,842 41,957 42,328 43,535 47,084
Oct-15 45,607   Oct-18 42,581 42,941 43,589 44,633 46,212
Nov-15 45,457   Nov-18 40,144 40,031 39,912 39,980 40,456
Dec-15 47,582              
Jan-16 46,313              
Feb-16 39,110              
Mar-16 56,107              
Apr-16 48,553              
May-16 39,483              
Jun-16 43,245              
Jul-16 41,954              
Aug-16 48,323              
Sep-16 52,052              
Oct-16 49,837              
Nov-16 41,619              

 

So, in businesses where there have been quite some changes within the last one year that impacts seasonality – for example: acquired new customer or added another product line – which would continue in the subsequent years, then we might want to use higher values of Gamma so that the seasonality from the recent year is captured in the forecast moving forward.

Enjoyed this post? Subscribe or follow Arkieva on LinkedinTwitter, and Facebook for blog updates