my question is the following:
I have to forecast the proportion of the sales of a product. For example:
For product 17503 there are four types of categories A, B, C and D. The total demand of the product is forecast in another way that does not have to do with statistics, but what this proportion means is: if someones forecast that the sales demand of product 17503 is going to be 1,000 units then this proportions means this:
Demand of Product 17503:
A= 1000*0.389= 389 units of product 17503 in category A.
B= 1000*0.389= 389 units of product 17503 in category B.
C= 1000*0.167= 167 units of product 17503 in category C.
D= 1000*0.056= 56 units of product 17503 in category D.
At present, what we do to estimate this proportions is given the sales of the day we estimate the proportions and then we reevaluate each day by the cumulative sales. As it is seen in the previous table.
One of the problems with this approach is the following:
-When a new category is sent to the store, for example D. It plays with total disadvantage to the other categories.
I have tried to forecast these with ARIMA models, ETS, Hierarchical Forecasting etc. but the problem is that the series is highly intermittent with a lot of 0's in some sales days.
Has somebody solve a similar problem that can recommend me some literature about this or give me a general idea?