Weather data - Seasonality and Inverse seasonality with sales

The purpose of the exercise if to identify cases where temperature and sales are inversely correlated despite having seasonal patterns and to show that our model predict the sales for such scenarios where we allow the temperature to always have positive impact on overall revenue. We also explore alternate modeling changes to see if the predicted sales fit better with the smoothed ones.

Input data of low correaltion of average Max temp with sales

1) Inverse seasonal WBLs for average Max temp with sales Inverse seasonal Inverse seasonal Inverse seasonal 2) Seasonal WBLs for average Max temp with sales Seasonal

Weather impact, predicted sales & smoothed sales

1) As is code Inverse seasonal Seasonal 2) Model weather such that its does not always have a positive impact on sales
Inverse seasonal Seasonal 3) Changing the behavior of model to allow different regression weights for each WBL for weather & periodic layer. Inverse seasonal Seasonal
4) Remove periodic layer & model weather such that its does not always have a positive impact on sales Inverse seasonal Seasonal 5) Model with higher Desired STDs of periodic layer to allow for diverse representations of periodicity across WBLs Inverse seasonal Seasonal