Function to modify a reform

[MaxGhenis]

I'm working on analyses that involve modifying reforms, for example removing deductions, and adding UBI to existing reforms. The UBI amount will be calculated in my script (to be revenue-neutral), so I'd like to be able to modify the reform on-the-fly. The ideal would be something like:

pol.modify_param(param='_UBI1', value=1000, year=2018)

Where pol is an existing reform, e.g. TCJA.

Looking through policy.py and parameters.py I don't a way to do this.

While the current motivation of seeing how things change under TCJA will soon be irrelevant, I think this would be a broadly useful tool for sensitivity analysis among proposed reforms.

[martinholmer]

@MaxGhenis said:

I'm working on analyses that involve modifying reforms, for example removing deductions, and adding UBI to existing reforms. The UBI amount will be calculated in my script (to be revenue-neutral), so I'd like to be able to modify the reform on-the-fly. The ideal would be something like:

pol.modify_param(param='_UBI1', value=1000, year=2018)

Where pol is an existing reform, e.g. TCJA.

Looking through policy.py and parameters.py I don't a way to do this.

While the current motivation of seeing how things change under TCJA will soon be irrelevant, I think this would be a broadly useful tool for sensitivity analysis among proposed reforms.

I agree with you, @MaxGhenis, that the ability to implement a "compound reform" (consisting of two simple reforms) will be "broadly useful." However, I don't think Tax-Calculator needs any enhancements to do what you want to do. My understanding is that there are no problems calling implement_reform twice for the same Policy object. If you have problems with the approach I outline below (which has been used extensively in pull request #1803), please raise another issue. Meanwhile, I'm closing this issue.

The (untested) example below assumes use of Tax-Calculator release 0.14.3, the last in which pre-TCJA policy is current-law policy. It also assumes that the "policy" dictionary derived from the reforms/TCJA_Reconciliation.json file is called `tcja_reform'.

pol = Policy()
pol.implement_reform(tcja_reform)
ubi_reform = {2018: {'_UBI1': [2500], '_UBI2': [10000], '_UBI3': [10000]}}
pol.implement_reform(ubi_reform)

[MaxGhenis]

Terrific, thank you @martinholmer! I used this to considerably simplify a charitable deduction analysis (commit) and verified it produces identical results to creating my own separate reforms.

[martinholmer]

@MaxGhenis, thanks for the feedback on your experience implementing compound reforms by calling the Policy implement_reform method twice. It's good to hear you were able to check the validity of this approach. I hope to add soon a Cookbook recipe that features this technique.

[MaxGhenis]

For posterity, the reform parameters require underscores, so could look like:

ubi_reform = {2018: {'_UBI1': [ubi_amount], 
                     '_UBI2': [ubi_amount], 
                     '_UBI3': [ubi_amount],
                     '_UBI_ecrt': [1.0]}}

I've found copy.deepcopy useful in conjunction with this approach, to create an edited version of an existing reform (e.g. to compare to it). For example here was a doubly-compounded reform after modifying to both repeal the charitable deduction and add a UBI (also in the notebook):

tcja_ubi_no_charitable_policy = copy.deepcopy(tcja_no_charitable_policy)
tcja_ubi_no_charitable_policy.implement_reform(ubi_reform)

[martinholmer]

@MaxGhenis, Thanks for the correction on my original example in issue #1802 and for you deepcopy usage tip here.