Forecasting Stories 2: The Power of a Seasonality Index
Read this second entry in a series on time series analysis and seasonality, and see how, through 2 simple use cases, the power of a seasonality index is uncovered.
By Rajneet Kaur, Data Scientist
The attainment %(actuals/target) of April was 78%. We had missed by a fifth.
The strange fact was we had performed poorly on all weeks. Following are the weekly attainment figures: Week 1: 75% Week 2: 77% Week 3: 79% Week 4: 81%. What was amiss? What is the reallife forecasting story we uncovered? Let's find out:
CASE 1 : Forecast overindexed in April
Here is what happened: As we can see from the image, April forecast seasonality was over indexed by 11%, i.e. at 89% of yearly average while actuals were trending towards 78%.
What does a seasonality index mean?
It simply refers to all variables normalized to a range close to 1, so that all variables are comparable. As explained in the image, we divide each number by their yearly average to calculate the index. This way, the average of all values in the entire variable column is always 1.
Hence interpreting the April seasonality, April being holiday is low performing month for this product. The forecast does partially take this into account, with 89% target compared to average of the year. However, actual performance across years can be seen ~80%. Hence the targets or forecast need to be even lower to be realistic.
Let us now move on to the 2nd story.
CASE 2 : Optimizing the marketing dollar
For President's day, your sales seasonality index is 134% while your marketing spends index is 51%!
And hence we should reallocate the spends from July Black Friday day to President's week. Alex, my client pondered over what I explained and said 'Makes sense, thanks'.
Comparing 2 seasonality indexes can give some powerpacked insights. Numbers after all, are good or bad only when they are compared against another.
The rest of the story is based on events, specially holidays in the US calendar. Not all of the following are holidays, and different events impact sales in different ways. Also, some events are more important than others. Have a look at the spends and sales seasonality indices:
What conclusions can you draw? My first observation was that we are spending a much higher proportion of marketing budget on July Black Friday, whereas President's day week results in higher sales. Alex appreciated the recommendation, since we would end up with a higher ROI without adding a single penny to the budget.
Now, as a consultant, it is our role to create such intelligent variables and highlight key differences. Many surprising and actionable insights lay in the hidden world of seasonality index. Hope you uncover them soon!
Also hope you enjoyed these stories. I have also written Forecasting Stories 1 and Analytics Frameworks, would appreciate any feedback.
Bio: Rajneet Kaur is a passionate and resultdriven marketing and data science executive with 6 years of business and technology experience.
Original. Reposted with permission.
Related:
 Forecasting Stories: Is it seasonality or not?
 How do we Better Solve Analytics Problems?
 How (not) to use Machine Learning for time series forecasting: The sequel
Top Stories Past 30 Days

