Distribution type-When to use G or N ?

[former_member696146]

Hello Experts,

In what scenarios, do we use Gamma distribution and when do we use Normal distribution ? My understanding is that Gamma distribution is more robust when compared to Normal distribution,but how to decide during implementation phase as what model would work better for the business.Do we have some pointers which would help us on finalizing the approach.Your inputs are highly appreciated.

BR,

Praveen

[lev_degtyarov · Accepted solution]

Hello Praveen,

You can describe demand distribution over exposure period (LT+PBR) by different types but I would recommend you to use Gamma distribution in most of cases. Gamma distribution is wider class of distributions than Normal, see below. I know only one use case when Normal distribution can be used, but this is not for real use case modelling.

Here are some key advantages to use Gamma:

(1) Special case of Normal distribution. As the shape parameter α in a Gamma distribution grows larger, the distribution becomes more symmetric and more like a normal distribution. Note, even this looks something like a normal distribution, though it’s not quite symmetric.

(2) Special case of Exponential distribution. When the shape parameter α = 1, Gamma distribution simplifies to exponential distribution. The exponential distributions can be used as a good approximation for items with lumpy demand, items with intermittent demand, or slow-moving items.

(3) The demand described by the Gamma distribution is always positive and continuous.

That is the good because real life demand is always positive. If you look at Normal distribution it has negative part, but demand cannot be negative.

Best Regards

Lev Degtyarov