0

I am trying to run fixed effects and random effects regressions on data which is not pure panel data but rather independent cross-sections over several years. In addition, the yearly cross-sections are of different sizes. The dataset consists of rows of syndicated loans with loan-level variables and borrower country-level variables. I'd like to run country fixed and random effects models while also controlling for year and industry fixed effects. In the end I aim to compare their results together and with results from a pooled OLS model.

The question is how can I apply pdata.frame() so that I can regress using plm() when my data has multiple observations (loans) per country-year pair? If I try to use year and country I get the following error:

Warning in pdata.frame(df, index = c("borrower_country", "year")) :
duplicate couples (id-time) in resulting pdata.frame
to find out which, use, e.g., table(index(your_pdataframe), useNA = "ifany")

I know that it is caused by duplicate id-time pairs, but since it's a central feature of the data I don't know how to fix the issue. I know that panel data methods have been used in studies with a similar data structure, but I don't understand how to apply the methods to my data. I've tried to find an answer but it seems as if there is a gap in this area. I've searched online, stackexchange, stackoverflow, textbooks I've listed below etc. but I can't seem to find a solution.

Any help would be much appreciated.

  • Woolridge (2012). Introductory Econometrics: A Modern Approach.
  • Woolridge (2010). Econometric Analysis of Cross Section and Panel Data.
  • Baltagi (2005). Econometric Analysis of Panel Data.
  • Tsionas (2019). Data Econometrics Empirical Applications.

PS. Is this an issue that would be easier to solve with Stata? I'm used to working with R but I am ready to try to do it with Stata if it would help. Although, based on previous questions I've read I suspect I'd face the same issue there.

Moz
  • 1
  • 1
  • Try to use lm4 like in this answer https://stackoverflow.com/a/49168979/2824732 – Robert Sep 07 '21 at 23:19
  • I'm not sure if that is directly applicable to my issue since it uses maximum likelihood and not GLS, [link](https://stats.stackexchange.com/questions/446361/plm-in-fixed-effects-model-doesnt-work-with-id-and-time). I'm not experienced with ML methods and studies on similar datasets have used GLS so ideally I'd like to do the same. Multiple studies say they use country random/fixed effects in their models but don't expand on the topic, e.g. "Creditor Rights, Enforcement, and Bank Loans" by Bae & Goyal (2009). It makes my issue seem trivial but I can't find information on how to solve it. – Moz Sep 08 '21 at 08:51
  • (extended) crosspost https://stackoverflow.com/questions/69087503/data-structure-issue-with-plm-multiple-observations-with-the-same-year-country – Helix123 Sep 09 '21 at 07:14

0 Answers0