I have to predict money amounts, which are always greater than 0. The distribution is very tailed (i.e. there are many small values but also many large data).

Just wondering would a count data model (e.g. GLM + Poisson) be suitable for this kind of scenario?



  1. you shouldn't expect to get different results by changing from dollars to cents or from working in thousands to working in millions - but that's what happens if you use a Poisson model.

  2. With money amounts you would generally expect spread to be proportional to mean rather than its square root (this would apply also to quasi-Poisson models). Indeed if one considers the effect of changes in inflation/interest under a Poisson or quasi-Poisson model (even if it's just constant either side of one change), we're left with inconsistent assumptions about variance.

Among GLMs "spread proportional to mean" would tend to suggest a gamma model.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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