0
$\begingroup$

I'm trying to forecast sales in one country (US) for a given period (let's say June and July 2019) based on previous sales in US and how sales in US correlate to sales in Canada, for which I have the June and July 2019 data.

So, my data looks somewhat like this:

country | date | sales
US 2019-03 1000
US 2019-04 2000
US 2019-05  500
CA 2019-03  700
CA 2019-04 1500
CA 2019-05  200
CA 2019-06  700
CA 2019-07 1600

Any suggestions on how to proceed?

$\endgroup$
4
  • $\begingroup$ So what have you tried? You can use a linear model or ARIMA for ex. with CA as a variable. $\endgroup$ Commented Aug 20, 2019 at 12:33
  • $\begingroup$ @user2974951 I have tried a linear regression so far, but I'm wondering whether there is a better way to go. $\endgroup$
    – barbara
    Commented Aug 20, 2019 at 12:36
  • 2
    $\begingroup$ ARIMA(X) would be the next step then. $\endgroup$ Commented Aug 20, 2019 at 12:38
  • $\begingroup$ @user2974951 great $\endgroup$
    – Fr1
    Commented Aug 20, 2019 at 12:46

1 Answer 1

1
$\begingroup$

My suggestion since you are dealing with time series and your errors are likely to be not iid, is to use a ARIMAX model like this , look also at this if you use Matlab)so that you can take into account:

  • the effect of other external regressors on the values of $y_{t}$

  • the effect on y of lagged values of y

  • the possible serial correlation in the error term (if correlation is not in the residuals but in the squared residuals then my suggestion is to model the error term accordingly, for example a solution may be to look at a GARCH regression, which from the coding point of view Matlab considers a special case of the previous one linked, you can find some discussion online by googling Matlab regression with Garch errors like this)

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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