Parameterize IRS rounding rules for indexed parameter values

[jdebacker]

Tax-Calculator should parameterize the rounding rules of indexed parameters in policy_current_law.json. There will probably need to be an update to the ParamTools package to handle rounding in the extrapolation of parameters.

The need for rounding has been highlighted in PR #2652, where a new JSON file was created to help with the round-trip-reform test. With out that, all JSON files of past legislative reforms need to be updated to reflect rounding rules in years beyond those affected by the legislative change.

Advantages of building in rounding rules:

1. Reduce time and errors in maintenance currently required to update rounded parameter values each year
2. Reduces the size of policy_current_law.json -- it would no longer need to include all values from the first to last known values (currently 2013-2022), but rather just values for any legislative changes
3. Allows for JSON files representing enacted laws to keep their integrity (no additional year values need to be added for rounded values going beyond the years affected by the legislation -- JSON files represent only the legislative changes)
4. Improves accuracy of calculations in years beyond the last known year

cc @MattHJensen @hdoupe

[hdoupe]

Hey @jdebacker , this sounds like a nice enhancement. I think you may be able to implement the new rounding rules on the taxcalc.Parameters class by overriding the extend_func method: https://github.com/PSLmodels/ParamTools/blob/0107eceb4a732e3c0eea32dd83b56ae253d009cb/paramtools/parameters.py#L973-L1020

By default, it rounds to 2 places. But, this method has access to the parameter's name, so you could round it however you'd like.

There's a simpler example in the tests that might make it a little easier to see how the method works: https://github.com/PSLmodels/ParamTools/blob/0107eceb4a732e3c0eea32dd83b56ae253d009cb/paramtools/tests/test_parameters.py#L2229-L2283

[jdebacker]

Cool - thanks for the pointers @hdoupe!

[jdebacker]

Relevant issue over in the ParamTools repo: PSLmodels/ParamTools#130

[jdebacker]

Note - will need to record the base year for indexing for each parameter. This can probably be recorded in the parameter metadata in policy_current_law.json

@Peter-Metz do you recall doing any research on the base years for each parameter that can be used for indexing?

[jdebacker]

Another note for this issue: in addition to recording the base year for each parameter in it's metadata, there would also need to be some data on what price index is used for indexing and over what years it's used. For example, one could use the unchained CPI-U until 2018, but the unchained CPI-U after.

[maxkossek]

How should we handle hard-coded values obtained from IRS in update_pcl.py? Based on some preliminary testing, I'm unable to recreate the values in policy_current_law.json from rounded and inflation-indexed parameters. How were the values in policy_current_law.json generated before 2022 and after 2022 (with update_pcl.py?)? Were these calculated using the growth rates in the repo, or manually entered by referencing official IRS, SSA, etc values?

@martinholmer What is your opinion on this issue?

@jdebacker Do you expect this change to be implemented in such a way as to not require any changes in the expected test results? Based on some local testing, we might have to make changes to the expected values for these tests:

Tax-Calculator/taxcalc/tests/test_reforms.py

Line 89 in d96f456

def test_round_trip_reforms(fyear, tests_path):

Tax-Calculator/taxcalc/tests/test_reforms.py

Line 172 in d96f456

def test_reform_json_and_output(reform_file, tax_year, tests_path):

Another note for this issue: in addition to recording the base year for each parameter in it's metadata, there would also need to be some data on what price index is used for indexing and over what years it's used. For example, one could use the unchained CPI-U until 2018, but the unchained CPI-U after.

Originally posted by @jdebacker in #2654

This is a good idea. We could also an add an index option such as "none" so that a parameter can change from indexed to non-indexed, and vice versa.