Vertrauensschutz in Rente für Arbeitslose

[TeBackh]

Special age thresholds for people with legitimate expectations during the phase-in of tightened eligibility requirements. Lots of additional improvements that popped up along the way.

[codecov[bot]]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Comparison is base (922aae7) 91.02% compared to head (f6986bb) 91.31%.

Additional details and impacted files

```diff @@ Coverage Diff @@ ## main #645 +/- ## ========================================== + Coverage 91.02% 91.31% +0.29% ========================================== Files 48 48 Lines 3143 3225 +82 ========================================== + Hits 2861 2945 +84 + Misses 282 280 -2 ```

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[MImmesberger]

When working on the Vertrauensschutz, I noticed that some rules regarding the statutory retirement ages of Arbeitslose are not implemented yet.

Currently, we mostly use piecewise linear functions. Now I have a reform where a piecewise linear function would have more than 40 segments, so it is both cumbersome to write and difficult to understand in the end. Currently I use the implementation below. But this has the downside that we need a new policy function even though only parameters changed. Any advice? I am leaning towards the piecewise linear function.

  1992-01-01:
    reference: Rentenreformgesetz 1992. BGBl. I S. 2261 1989 § 41
    note: Increase of full retirement age from 60 to 65 for birth cohort 1941-1952.
    [.....]
    1951:
      0: 64.083333
      2: 64.166667
      4: 64.25
      6: 64.333333
      8: 64.416667
      10: 64.5
    1952:
      0: 64.583333
      2: 64.666667
      4: 64.75
      6: 64.833333
      8: 64.916667
      10: 65
    1953:
      scalar: 65
  1997-01-01:
    reference: Wachstums- und Beschäftigungsförderungsgesetz 1996. BGBl. I S. 1461 1996
    note: Increase of full retirement age from 60 to 65 for birth cohort 1937-1941.
    0:
      lower_threshold: -inf
      upper_threshold: 1936.916666666
      rate_linear: 0
      intercept_at_lower_threshold: 60
    1:
      upper_threshold: 1941.916666666
      rate_linear: 1
    2:
      upper_threshold: inf
      rate_linear: 0

[hmgaudecker]

I don't quite follow. Why do we need a new policy function? In general, writing out the parameters is much better IMHO, even if it becomes lengthy. piecewise_linear is really made for individual-level stuff like the tax code, not for converting aggregate parameters. Of course it can be used for that purpose, too -- but it typically becomes very hard to read.

If the answer to Why do we need a new policy function? is Because of the second part for the 1937-1941 cohorts, better convert that to direct parameters, too.

[MImmesberger]

If the answer to Why do we need a new policy function? is Because of the second part for the 1937-1941 cohorts, better convert that to direct parameters, too.

Yes, exactly!

[MImmesberger]

In the last commit I rewrote the functions and parameter inputs for the Rente wg. Arbeitslosigkeit and implemented the Vertrauensschutz rules for 1997 and 2006.

When working on it, I noticed that the former implementation was not quite correct because increases in statutory retirement ages after the abolishment of the Rente wg. Arbeitslosigkeit were also applied there. I guess this is an artifact of an older version of GETTSIM.

This is why tests are currently failing. For example, in test_data/renten_alter/2010/hh_id_769.yaml a retirement age of 67 for the Rente wg. Arbeitslosigkeit is the expected output. AFAIK, the full retirement age was 65 when the Rente wg. Arbeitslosigkeit was abolished.

Long story short: We probably have to update the test data.

Edit: The Rente für Frauen also needs a major overhaul. Same problems as with Rente wg. Arbeitslosigkeit and also some Vertrauensschutz rules. But implementation should be very similar to this one here.

[MImmesberger]

ToDo:

• [x] Fix the wrong retirement age tests
• [x] Set an end date for functions that use the Vertrauensschutz variable as an input. End date = Year in which there can be no additional retirement entries through this Vertrauensschutz rule.

[MImmesberger]

In case you are confused by the recent changes: Apparently they first accelerated the FRA increase (in July 1996) while the ERA increased as before (in 1992). Then a few months later (September 1996), they abolished the increase in the ERA but kept the acceleration in the FRA from July 1996.

[MImmesberger]

I added some tests for the Rente wg. Arbeitslosigkeit. Those tests are made up using the texts of the relevant laws.

I have one problem that I'm not able to solve: Some tests fail because of a pandas error: ValueError: If using all scalar values, you must pass an index.

All of the failed tests seem to address functions that just return a parameter from the params dictionary, i.e.

def func(params):
  return params["some_param"]

Other than that, the PR is ready to be reviewed.

[hmgaudecker]

Thanks!!!

I have one problem that I'm not able to solve: Some tests fail because of a pandas error: ValueError: If using all scalar values, you must pass an index.

All of the failed tests seem to address functions that just return a parameter from the params dictionary, i.e.

def func(params):
  return params["some_param"]

So the problem is that for whatever reason, this implies that instead of something like this:

{'entgeltp_update': array([31.0625])}

the following arrives in line 738 of interface.py:

{'ges_rente_arbeitsl_vorzeitig': array(60.)}

I don't quite see why that happens as the outputs of functions look the same (just a scalar). @lars-reimann, would you mind having a quick look?

In any case, we'll just need some workaround and truly fix it when looking seriously at the interface.

[lars-reimann]

Hmm, the tests fail for functions that only depend on ges_rente_params:

@dates_active(end="1991-12-31", change_name="ges_rente_arbeitsl_vorzeitig")
def _ges_rente_arbeitsl_vorzeitig_ohne_staffelung(
    ges_rente_params: dict
):
   ...

Adding any other dependency resolves the issue:

@dates_active(end="1991-12-31", change_name="ges_rente_arbeitsl_vorzeitig")
def _ges_rente_arbeitsl_vorzeitig_ohne_staffelung(
        ges_rente_params: dict,
        geburtsjahr: int,
):
   ...

I didn't figure out why that happens yet, though.

[hmgaudecker]

Ahh, sure, the params are partialed in beforehand, so something weird must be happening when doing the calculations via the DAG.

Just add a fake dependency, corresponding # noqa: XXX statements, and an issue so as not to forget about it when re-doing the interface.

[MImmesberger]

Ah, thanks @lars-reimann @hmgaudecker. It worked!

[hmgaudecker]

I added some tests for the Rente wg. Arbeitslosigkeit.

Cool. Only thing is that for new tests, can you please call the file for what is being tested instead of test_hh_id_[x].yaml? That was a leftover of the bulk conversion of csv-formatted tests to individuals / households.

E.g., I am just looking at the 2007 change. So there could be tests (making up the year here):

• test_renten_alter / 2015 / cohort_1951_wegen_arbeitsl.yaml
• test_renten_alter / 2015 / cohort_1952.yaml

[MImmesberger]

Thank you for the comments! Will keep those in mind for future PRs. I implemented the changes or replied to your concerns.