Multi Level nested logit. about pyblp HOT 6 CLOSED

For example, I would like to have a rho for "mushy" and "nonmushy" and second rho for "sweet" and "unsweetened"? Is this possible?

from pyblp.

You're right that pyblp doesn't currently support the three-level nested logit you're describing (choices over the nest, subneset, and alternatives). I haven't implemented variations on the nested logit model (except for including random coefficients) because my focus has been on RC-based variants.

There are a lot of nested logit variants and implementing them isn't high up on my to-do list. Expressions for derivatives get messy and my understanding is that literature has to some extent moved more towards RCs. But I'm happy to accept contributions!

I'll keep this issue open as a possible future enhancement to the package.

from pyblp.

I think most people would prefer to estimate a joint distribution of random coefficients for "sweet" and "mushy" in this case. Otherwise you have to take a stand on whether consumers choose "sweet" first and then "mushy" or the other way around.

In principle I think this would be straightforward because the resulting estimator is linear in parameters.

There are other models of partially overlapping nests that are a bit more involved.

Is there a particular example with aggregate data you were thinking of? There seem to be a lot of variants out there and nothing I would call the "standard" version.

from pyblp.

Dear Jeff and Chris:
Thanks for the quick reply, I'm currently trying to do an RCNL on the smartphone market. I thought about to divide the smartphone into 3 groups(low, mid, high ends), and a subgroup of (iOS, Android).

If there is a strong in-group preference among iOS users (or Android), then it is important to implementing RCNL in that case.

So far, I have run an RCNL base on just one layer(iOS, Android), and it is significantly different from the RC model with the random coefficient on iOS.

I was wondering, what would the substitution pattern looks like if I implemented multilevel RCNL.

Thanks for your response. You guys are awesome.

from pyblp.

from pyblp.

Cleaning up issues, multi-level nested logit isn't something we've had much demand for, and it'd be pretty difficult to implement. So I'm closing this for now.

from pyblp.