[WIP] Arellano bond panel data

[jaimeoliver1]

Hi,

we have implemented the Arellano Bond estimator based on this previous implementation: https://gist.github.com/toobaz/6d01be09bbd56de7038fe8ff26780138.
We have two comments:

• The AB estimator is a panel data model, but it builds on top of an IV model. Therefore, we have thought that importing IVGMM from the linearmodels.iv.model sub-module was the easiest option, but we are open to other options.
• We have not implemented any further tests, given that the model builds on top of already tested models. Any insight on what would the testing strategy be here?

This PR closes #95

[pep8speaks]

Hello @jaimeoliver1! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

• In the file linearmodels/panel/model.py:

Line 3142:57: W605 invalid escape sequence '\e'
Line 3222:100: E501 line too long (122 > 99 characters)
Line 3264:100: E501 line too long (121 > 99 characters)
Line 3292:100: E501 line too long (103 > 99 characters)

Comment last updated at 2022-07-15 10:05:55 UTC

[bashtage]

Can you run black and isort?

[codecov]

Codecov Report

Merging #421 (0e9a902) into main (21d0591) will decrease coverage by 0.47%.
The diff coverage is 18.48%.

@@ Coverage Diff @@ ## main #421 +/- ## ========================================== - Coverage 99.58% 99.11% -0.48%  ========================================== Files 98 98 Lines 16899 20117 +3218 Branches 1987 2968 +981 ========================================== + Hits 16829 19938 +3109  - Misses 30 142 +112  + Partials 40 37 -3 

Impacted Files	Coverage Δ
linearmodels/panel/model.py	93.74% <18.48%> (-5.59%)	⬇️
linearmodels/__init__.py	83.33% <0.00%> (-14.23%)	⬇️
linearmodels/shared/hypotheses.py	98.55% <0.00%> (-1.45%)	⬇️
linearmodels/shared/utility.py	98.68% <0.00%> (-1.32%)	⬇️
linearmodels/panel/data.py	99.38% <0.00%> (-0.62%)	⬇️
linearmodels/panel/covariance.py	99.23% <0.00%> (-0.35%)	⬇️
linearmodels/tests/panel/test_utility.py	99.49% <0.00%> (-0.01%)	⬇️
linearmodels/iv/gmm.py	100.00% <0.00%> (ø)
linearmodels/iv/common.py	100.00% <0.00%> (ø)
... and 27 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 21d0591...0e9a902. Read the comment docs.

[jaimeoliver1]

@bashtage I committed the file with the correct formatting

[bashtage]

I think the best way to test would be to take the Stata code as a reference and test the various outputs and options.

[bashtage]

Overall looks very good.

1. Need to add it to the docs.
2. Can you break these long comments that are making PEP unhappy.
3. Can you use a raw string r"Code with escape like \epsilon" to silence the other.

[jaimeoliver1]

Hi @bashtage,

I've tried adding the stata tests, however I've no access to stata, so I would need a bit of help here.
I think the way to go would be starting by:

1. Add the examples in the documentation here. The dataset used in the docs can be found here

• xtabondnl(0/1).wl(0/2).(kys) yr1980-yr1984year, lags(2)

2. Generate the model results and add them to the results_file

Once the results are in the stata-panel-simulated-results.txt, I'll be able to implement some tests here

[jaimeoliver1]

Also, we fixed the linting error and silence the "\epsilon" issues.
We also added documentation in the docstring. However, doc build is failing (even in master...)

[bashtage]

Thanks for following up. I can run Stata and generate results.

[bashtage]

Going to close and reopen to trigger a rebuild.