Im struggling with understanding when I should be using time series vs a single time aggregated period cross sectional.

Say I have many customers and i want to understand the impact of marketing on sales for a single year (say 2020). I have daily sales for each customer as well as marketing events that have happened throughout the year for each customer. I also have individual non-time varying variables for each customer.

One option is to do something like a two way panel regression looking at monthly sales data per customer as well as monthly marketing. You would likely have to take into account something like adstock (which is common in marketing analysis). Other considerations would also need to account for autocorrelation and the like as well.

Another way is using the total marketing promotion for the whole year compared to the difference in sales for a given customer (2020 vs 2019). This is a single model to see if customers with more marketing events had a higher change in sales compared to the previous year.

Is there a reason why i would want to use a panel regression / time series model over the aggregate for the given year? Is the aggregation method incorrect?

  • $\begingroup$ One thing that i just realized is that the timing of the marketing in this example might matter. For example if most of your marketing for a given customer happened at the beginning of the year versus it spread out throughout the year it might have a difference in impact in terms of sales (due to saturation effects). I suppose that means with the aggregate you assume the distribution of marketing across time is independent of its impact? $\endgroup$ May 25, 2021 at 21:05


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.