How well does (state tax + federal tax) != 0 reflect tax filing?

[MaxGhenis]

For calibrating against SOI targets, we currently filter to tax units with nonzero state+federal income tax, and compare that to the value among All returns in SOI tables. I'm wondering how well this reflects the choices people make to file, and accordingly how well aligned these are.

In particular, the Social Security targets made me curious about this. Publication 1304, Table 1.4 (2021) shows 31.3 million returns with Social Security benefits, of which 21.6 million are taxable returns, and 24.0 million returns have taxable Social Security benefits. I don't think 10 million Social Security beneficiaries receive refundable tax credits; they might just file taxes out of habit.

Calibrating to taxable returns could be a safer bet. Curious what you think @nikhilwoodruff @donboyd5 @martinholmer

FWIW, ChatGPT suggested these reasons people could file non-taxable returns (Claude said basically the same):

1. To Claim Refunds: Many individuals file tax returns even if they don't owe taxes to claim refunds of withheld income or to receive refundable tax credits.
2. To Fulfill Filing Requirements: Some may have to file because their gross income exceeds certain thresholds, or they have special tax situations requiring filing, regardless of the taxability of their Social Security benefits.
3. For State Tax Obligations: Filing a federal return can also be necessary to comply with state tax filing requirements, where their Social Security benefits might be subject to state taxes.
4. To Establish Tax Records: Filing a return creates a record of income and taxes, which can be important for future credit or loan applications, and for determining eligibility for other government benefits or assistance programs.
5. Legal Requirement: Even if the income is not taxable at the federal level, taxpayers may file to adhere to legal obligations, ensuring compliance with all tax laws.

[MaxGhenis]

After looking more into this, I think we should change our approach. The below table (notebook) shows that our projections for SS with respect to taxable returns are unstable. I've starred the columns we target today.

Year	Social Security Benefits ($ billions)*	Social Security Benefits among all returns ($ billions)*	Social Security Benefits among taxable returns ($ billions)	Returns with Social Security Benefits (millions)*	Taxable returns with Social Security Benefits (millions)
2021 (actual)	1,133.2	791.2	582.0	31.3	21.6
2022	1,177.6	836.7	697.0	38.7	31.6
2023	1,346.4	862.7	615.7	33.3	21.0
2024	1,450.1	953.5	696.7	34.3	22.1
2025	1,540.0	1,025.1	753.0	35.2	22.7

I'd suggest we start by switching to taxable return targets (#4284 and #4285). Later on, we could include targets among all returns, but I think only after we impute a probability of filing taxes among tax units with zero net tax (#4286 - we don't need to impute an accurate probability ex ante, since the reweighting will determine it; it's more of a prior).