Prediction technique for a new product launch in Pharma


I am working on a forecasting problem for generic drug launch and the only data I have is past sales for branded drugs, unit cost of branded drugs, and other categorical variables such as drug category(chronic, wellness etc.), drug form(tablet, solution etc).

The major problem: Generics are launched once the patent of branded drugs expire. And so, I do not have any past data for generic drugs. I am trying to forecast the demand of generics for inventory optimization.

Secondly, can you also suggest some other significant factors which I should look upon to design the model?

Finally, if the distribution is highly skewed, what would be the most apt data transformation that I can use?


They are working in the field which you are asking about. This website might be helpful to you.


Hi @ankit9285

For sure what you consider is where the generic is launch? In European countries the health system will give its approval and will push for generic to reduce cost, so if for example the generic is to be launch let say in France you should consider that it is a state with public health care system, same in UK.

Transformation well this depend on the model you use, as you mentioned forecasting ARIMA for example is one your list in this case the autocorallration is important and you correct with the lag and the autocorellation factor, if it is a non stationary process then the differentiation all are parameters of the forecast() function of the forecast package if you use this package.
If you go for linear regression your response should be normal and then skewness could be corrected with log, square root (with mirror is necessary) or Boxcox with optimise lambda.

For the independent variables this is based on the business domain and the model if we consider interaction (case of linear model, if you a tree model then the tree will care for you ).

in few words transformations are dependent of the model you will use.
Hope this help