Coca cola demand forecast (models)

[karakilamaravilla]

One small addition to Gareth's points. I think the "day of the month" variable" is just a sequence number within the month. In other words, "1" might sometimes be the first observation but the second day of the month. Since you know the day of the week, you could probably manually insert dates, although I imagine it would take an hour or so.

It's already the date considered look at the answer I gave Gareth

regards

[startz]

It's nice of you to share all this information.

Can you explain the word "quasivariation?" I'm not familiar with it.

[karakilamaravilla]

It's nice of you to share all this information.

Can you explain the word "quasivariation?" I'm not familiar with it.

http://www.uv.es/webgid/Descriptiva/Imagen17.jpg

A measure that minimizes the square of the error minus the average error.

Remember that the error in my model is given by

Error! M = _dayspiso - Matrix! M

Where _díaspiso is the real (for the period from 01/01/2017 to 04/30/2017)
And matrix! M is given by each model! M.

Thank you

[startz]

Ah, thanks.

[karakilamaravilla]

Before delving too deeply into this, I'll point out a couple of things.

First you might want to structure your workfile as a dated workfile. Going on the information you gave (it starts in 2014, the series "mes" is month, and "salida_del_mes" is day of the month), you can use the following code to turn your workfile dated:

Code: Select all

wfopen "para pregunta del foro.wf1"
series year = 2014
smpl @first+1 @last
year = @recode(mes<mes(-1), year(-1)+1, year(-1))
smpl @all
series date = @dateval(@str(year)+"/"+@str(mes)+"/"+@str(salida_del_mes), "YYYY/MM/DD")
pagestruct @date(date)


Note that this then contradicts your first post (where you said the data ends 4/30/2017 - it actually ends 5/17/2017).

It also creates an irregular workfile, since, I presume, some days were holidays etc...


You might want to check the data in the workfile though. A quick look at, say, SKU14 is weird. There are lots of 1s and 0s, and also lots of negative numbers (negative demand?). The data don't appear to match up with the data in the later Excel file you posted.

Gareth,

Any recommendations to improve my models?

Best regards,

[EViews Gareth]

Unfortunately, I do not give econometric advice.

[startz]

One suggestion I made earlier was to include month dummies. But I suspect the most important advice is to look carefully about Gareth's comment on something being wrong with the data, both the large number of 1s for some skus and the data not matching the Excel file.

[karakilamaravilla]

Unfortunately, I do not give econometric advice.

Thank you anyway!

[karakilamaravilla]

One suggestion I made earlier was to include month dummies. But I suspect the most important advice is to look carefully about Gareth's comment on something being wrong with the data, both the large number of 1s for some skus and the data not matching the Excel file.

Startz,

These are particular SKUs as they correspond to sales promotions.

"They are at the customers' request"
In fact the demand is zero, but to solve the problem of division by zero, add one to each demand.

[startz]

One suggestion I made earlier was to include month dummies. But I suspect the most important advice is to look carefully about Gareth's comment on something being wrong with the data, both the large number of 1s for some skus and the data not matching the Excel file.

Startz,

These are particular SKUs as they correspond to sales promotions.

"They are at the customers' request"
In fact the demand is zero, but to solve the problem of division by zero, add one to each demand.

Got it.
Presumably, these should be excluded from the forecasting then.