0
$\begingroup$

I don't really know much about stats so if the following sounds completely absurd, i'm sorry.

I've got 18 months of retail data Jan 2014 to now. It's an entirely online business, and I've got daily, weekly, monthly and annual sales and website traffic data. I've been asked to work out to what extent offline advertising campaigns are driving traffic to our website.

Firstly, there it seems like there is a rough quadratic trend for daily data with peak traffic midweek on Wednesday (although we send out a newsletter on wednesday) and trough on the weekend on Saturday. There then seems to be an exponential trend for monthly traffic with Jan-September remaining relatively constant (almost horizontal linear line) then a sharp curved slope from October up to Christmas, then a vertical drop at the start of jan. Unfortunately only have one cycle of seasonal data (1 christmas period) as the business is relatively new.

It's also clear that there is significant year-on-year growth.

So I am trying to build some sort of model around the above.

a) Firstly, Is it possible? If not I'll stop wasting my (and your) time b) If so, where do I start.

Thanks a lot in advance.

$\endgroup$
0
$\begingroup$

It's definitely possible. Look into seasonality-adjusted time series models, such as Holt-Winters. These can deconstruct univariate time series into their base components: level, trend and seasonality. There is a two-year minimum for initialization period for HW so you may need to consider ARIMA models, which would require creating a host of dummy variables for the various date components such as day-of-week, month, and year. Good luck.

$\endgroup$

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.