I got 95 weeks of sales data (i.e., 95 data points) for a retail business, whose plot looks like this:

Sales by Week

Sales are evidently seasonal. Also see plot for Year 1 against Year 2 Sales by Week of Year

Weekly sales by Year

I also got events defined for 8 of the 52 weeks in a year (e.g X-mas, Thanksgiving). Considering there is no other additional data on break-up of sales by Day of Week, stores, products or any other potential regressor, what would be your suggested approach to forecast weekly sales for Year 3?


If you are looking to fit an ARIMA model, first you will have to make sure that the series is stationary. Also check if the variance is stable. If not, I would suggest a ln transformation.

Now important to note, since you have seasonality in your data, your will most like have to take a seasonal difference as well. I would suggest starting with a periodogram, using it to identify the period. From there you can use the ACF and PACF to identify your seasonal part and after that your non-seasonal part of the model. This will give you a SARIMA model. Remember to check that the residuals are white noise before you choose a model.

Forecasting a full year ahead with only two years of data, can prove to be somewhat difficult or rather, it could be inaccurate. Each new forecasted value should be a different week as you stated that each observation is a different week.

Hope this answers your question more or less! You are welcome to ask, give suggestions or make corrections to this.

| cite | improve this answer | |
  • $\begingroup$ Thanks Neil. I am still researching on the approach. The primary challenge is data availability being limited to less than two full seasonal cycles. I will consider your input & update my response once I've worked on them. $\endgroup$ – Nibbles May 12 at 11:42

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.