What are techniques for forecasting daily sales for a 14 day interval? Note, my historical data is 14 days of daily sales data that may or may not have fallen in the same month (e.g. July 2017) as the interval I am trying to forecast for (e.g. July 2018).
$\begingroup$
$\endgroup$
3
-
$\begingroup$ Uh. What is unclear about this question? I find it clear enough. $\endgroup$– Stephan KolassaCommented Jul 18, 2018 at 16:14
-
$\begingroup$ "daily sales data that may or may not have fallen in the same month" - what does it mean? Your output is 14 daily sales for, e.g., 10-23 July 2018, and your input may be 14 daily sales for, e.g. 1-14 January 2017, or any other 14-day interval. Is this true? If not, what is true? $\endgroup$– user31264Commented Jul 19, 2018 at 1:11
-
$\begingroup$ This means that if I am forecasting in July 2018 then I may not necessarily be using historical data from say July 2017 (the same month as I am forecasting in)..make sense? $\endgroup$– user6866797Commented Jul 20, 2018 at 16:15
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
Daily sales typically exhibit intra-weekly seasonality. However, you have very little data to fit this reliably. You basically have two choices:
- Use the overall historical mean. Note that the overall mean can be surprisingly hard to beat in terms of accuracy.
- To forecast a Tuesday's sales, take the average sales for the two Tuesdays you have in your history, and so on for the other days.
Since you have very little data, the second approach may be unstable, so there is a good chance the first one will be better. If there is any way you could obtain more data, do so.
We have quite a few prior threads on daily forecasting. Consider browsing through these.
answered Jul 17, 2018 at 5:18
-
$\begingroup$ Thank you for your response! Will you clarify "Use the overall historical mean"? Are you saying take the mean of the 14 days of actuals and use that value as the forecasted value for each day in my forecast? Thanks! $\endgroup$ Commented Jul 18, 2018 at 16:11
-
$\begingroup$ Exactly. Just take the average over all historical data points. (Alternatively, use the median for added robustness.) $\endgroup$ Commented Jul 18, 2018 at 16:13