Panel regression-continuous dependent variable with lots of zeros

[Catherine Zhang]

Dear All,

Thank you a lot for looking through this post. I am doing a monthly panel regression with fixed effect. My dependent variable is continuous, which is the first difference of fund expense. My aim is to explain fund expense change by some predictors. However, the summary stats give that about 95% of my DV equals to zero, which means that the majority of my sample funds did not change their expense monthly. So my question is: if this is a problem for conducting standard panel regression? I know that zero-inflated poisson model is suitable for count data with lots of zero, but it is not said to fit continuous variables. As zero change in my study is meaningful, so I do not think that I should drop all the observations with zero DV.

Would really appreciate if anyone could help with this!

[William Lisowski]

This is beyond my current expertise, but in the past I have been involved in analyses that fit your sort of data as what was then called a "two-part model". These days that is possibly what is meant by a "hurdle model", but there may also be other alternatives.

In your case, this type of model would be appropriate if you think that the decision about the fund expense is made by the fund managers as a set of two decisions:

• do we change the fund expense this month?
• if so, by how much do we change it?

Given that fund expense changes about once every two years on average, this seems a reasonable assumption. Setting fund expense charges is similar to many other problems in the pricing of consumer goods: sellers prefer a stable price to a volatile one.

If this interests you, you can find more about it in the documentation for the churdle command in the Stata Base Reference Manual PDF included in your Stata installation and accessible from Stata's Help menu.

[Catherine Zhang]

Thank you very much for your detailed explanation and suggestions. This helps me a lot! Will look into the hurdle model for my study!

[Phil Bromiley]

Another approach might be to use level of expenses as the dv and control for past expenses (accounting for endogenity)

[Catherine Zhang]

Another approach might be to use level of expenses as the dv and control for past expenses (accounting for endogenity)

Thank you a lot for the advice.Will look into this as well!