Running synthesized files through Tax-Calculator

[donboyd5]

Issue:

• We are creating a synthesized file from the raw PUF for 2011.
• We'd like to compare certain complex variables computed from the synthesized data with their counterparts in the raw data, to help us evaluate file quality.
• Most notable are adjusted gross income and various tax measures.
• They are too complex to calculate from scratch, so we plan to run the synthesized and raw files through Tax-Calculator and compare the resultant computed variables.
• The first tax year we can compute in Tax-Calculator is 2013, not 2011.

Proposed solution:

• I propose to (and have started to) run the raw and synthesized files through Tax-Calculator with the tax year set to 2013, and will examine the results.
• There were some tax-law changes between 2011 and 2013, the biggest of which were top-rate increases. However, I think this approach is reasonable nonetheless because the same 2013 law would be applied to both the raw 2011 PUF and the synthesized 2011 PUF. Our concern will be how far apart the two files are, not how different the two tax laws are. The approach has the advantage of being tractable and non-cumbersome.

Two questions, especially for @andersonfrailey:

1. Do we think this makes sense?
2. I'm thinking that we should do the same for the 3 tax reforms @andersonfrailey put together (run reforms vs. "current law" defined as 2013 law). Does that makes sense? @andersonfrailey, to do that, do I just need to change 2019 to 2013 in the json files?

@feenberg, are you using Tax-Calculator, and if so what tax year are you using?

[andersonfrailey]

@donboyd5 I think this makes sense. And you're correct that all you would need to do is change 2019 to 2013 in the reform files.

[feenberg]

On Wed, 12 Dec 2018, Don Boyd wrote:

@feenberg, are you using Tax-Calculator, and if so what tax year are you using?

I am using taxcalc.sas, and comparing to 2011. Is the synthetic data for a later year? If so what can I use for comparison? I think we should simulate the same year we have data for, and then age that, not age first and then simulate. Otherwise testing is rather confusing.

I will have results tomorrow.

dan

[donboyd5]

That's good - 2011 is the synthesis year.

[feenberg]

I have added are a large number of non-calculated variables to the tax reforms previously tested.

synscore4.txt

That is for each of the income and deduction items mentioned, the table shows the difference in the revenue score between PUF and synth4 when $100 is added to that item. I am still working on getting standard errors on the PUF scores.

[feenberg]

Apparently Github doesn't like the Linux newline indicator. Here it is with Windows newlines: synscore4.txt