[draft] Align federal SALT to state income-tax liability (#1058)

[PavelMakarchuk]

Draft — do not merge without sign-off on the convention.

What

The emulator's federal SALT deduction uses PE's imputed state_withheld_income_tax (~19% above the actual liability); the TAXSIM-35 binary uses the state income-tax liability. This aligns PE to the liability via a two-sim flow (pass 1 computes state_income_tax; pass 2 set_inputs state_withheld_income_tax = liability and provides the federal columns; state-output columns come from pass 1 so state tax is byte-identical — Hawaii reads state_withheld_income_tax directly). Inert under --disable-salt.

Status / honest caveats

• ✅ Correct + regression-free: NY NY joint 2025 210Kintrec #974 → fiitax = 22921.64 (binary exact); full suite 175 passed; zero state-tax change (verified 0/2000 records differ after routing state columns from pass 1).
• ⚠️ Inert on the eCPS: every eCPS record has mortgage = proptax = 0 → nobody itemizes → the state-income-tax SALT component never enters the deduction. Fed match unchanged (2021 80.4%, 2023 81.2%, 2025 84.5%). It only helps records that itemize (feenberg's NY joint 2025 210Kintrec #974/CA 2025 joint 168Kintrec #973).
• ⚠️ Convention change: withholding-vs-liability is a modeling decision (PE deducts imputed state income-tax withholding (not the liability) as federal SALT — federal residual on itemizing records (#974, #973) #1058) — needs @feenberg's sign-off.
• ⚠️ Cost/size: a ~350-line runner refactor (two-sim) and ~2x compute on the itemizing path.

Given it's inert on the main data and a convention change, I'm not recommending merge — opening as a tested option for review. Fixes the federal residual on #974/#973 if adopted.

Closes #1058 (if merged).

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
policyengine-taxsim	Ready	Preview, Comment	Jul 6, 2026 4:57am

[feenberg]

On Sun, 5 Jul 2026, Pavel Makarchuk wrote: **Draft ? do not merge without sign-off on the convention.** ## What The emulator's federal SALT deduction uses PE's imputed `state_withheld_income_tax` (~19% above the actual liability); the TAXSIM-35 binary uses the state income-tax **liability**. This aligns PE to the liability via a two-sim flow (pass 1 computes `state_income_tax`; pass 2 set_inputs `state_withheld_income_tax` = liability and provides the federal columns; state-output columns come from pass 1 so state tax is byte-identical ? Hawaii reads `state_withheld_income_tax` directly). Inert under `--disable-salt`.

To clear, in production Taxsim iterates 3 times calculating state and federal tax for each other's itemized deduction schedules. This would make comparison testing impossible, so when option(30)==1, there is no iteration and no deduction for income taxes on any state or federal return. I would not favor using an estimate of withholding in place of the iteration for production but I could add 5% of agi to "other deductions" to get a larger number of itemizers, if you thought that was useful. When testing the state treatment of itemizing I have been adding 50,000 for the mortgage deduction on every record. Crude, but helpful.

## Status / honest caveats - ? **Correct + regression-free:** NY #974 ? `fiitax` = 22921.64 (binary exact); full suite **175 passed**; **zero** state-tax change (verified 0/2000 records differ after routing state columns from pass 1). - ?? **Inert on the eCPS:** every eCPS record has `mortgage = proptax = 0` ? nobody itemizes ? the state-income-tax SALT component never enters the deduction. Fed match unchanged (2021 80.4%, 2023 81.2%, 2025 84.5%). It only helps records that **itemize** (feenberg's #974/#973). - ?? **Convention change:** withholding-vs-liability is a modeling decision (#1058) ? needs @feenberg's sign-off. - ?? **Cost/size:** a ~350-line runner refactor (two-sim) and ~2x compute on the itemizing path.

I can see how it would be a 350 line refactor. In Taxsim, I have a loop that is essentially: do i=1,3 call nlaw(d,law) state_tax_in=stata_tax_out enddo It triples the calculating cost for Taxsim but that isn't very significant for Taxsim. I would be glad to reduce 3 iterations to 2, if that helped your costs. Or, you could save the difference between the standard and itemized deductions from the federal calculation, and leave the loop after the first iteration if state tax was less than the shortfall of itemized deductions without state income tax compared to the standard deduction. That would eliminate the loop for most taxpayers.

Given it's inert on the main data and a convention change, I'm **not** recommending merge ? opening as a tested option for review. Fixes the federal residual on #974/#973 if adopted.

Maybe we should concentrate on bugs for the time being. Dan

…

Closes #1058 (if merged). ? Generated with [Claude Code](https://claude.com/claude-code) You can view, comment on, or merge this pull request online at: #1060 -- Commit Summary -- * Align federal SALT to state income tax liability (draft, #1058) -- File Changes -- A changelog.d/fix-salt-liability-alignment.fixed.md (1) M policyengine_taxsim/runners/policyengine_runner.py (352) M tests/test_performance.py (4) A tests/test_salt_liability_alignment.py (86) -- Patch Links -- https://github.com/PolicyEngine/policyengine-taxsim/pull/1060.patch https://github.com/PolicyEngine/policyengine-taxsim/pull/1060.diff -- Reply to this email directly or view it on GitHub: #1060 You are receiving this because you were mentioned. Message ID: ***@***.***>