State top-tail income mass is uncalibrated: CO M+ AGI at 24% of SOI in buildp-acs-local (fails progressive-bracket scoring)

Summary

Scoring Colorado Initiative 195 (Amendment 87 — graduated income tax with 7.4/7.9/8.4% brackets at $500k/$750k/$1M) on populace-us-2024-buildp-acs-local produced a revenue estimate of $1.116B for TY2027, versus ~$1.9B implied by IRS SOI actuals and ~$1.98B from the Colorado Legislative Council Staff fiscal impact statement. Diagnosis: the CO top tail matches SOI on filer counts but truncates income per filer, and the affected region carries no calibration targets.

Evidence (model vs IRS SOI Historic Table 2, CO TY2023)

• Mean $1M+ filer AGI: $1.27M model vs $3.03M SOI — the $3M–$100M+ filers are absent.
• Income mass above the $1M threshold (the base of an 8.4% bracket): $7.8B model vs $31.7B SOI (24% coverage).
• Meanwhile CO aggregates calibrate tightly in this release: return count −0.02%, AGI +0.12%, taxable income −0.10%, cap gains −0.48%.
• The only per-band CO targets are taxable-interest returns/amounts (all within ~1%), with a top band of 500k_plus and no 500k/750k/1M splits. The federal income_tax_before_credits_amount target for CO runs −3.10% against +0.12% AGI — an independent signal of a thin tail.

Scaling the reform's upper-bracket revenue by the SOI coverage ratio (0.68 on the like-for-like bracket base) moves the estimate to ≈$1.9B, within 5% of the LCS annualized figure — i.e., essentially the entire individual-side gap is tail-income truncation.

Proposal

Add state × AGI-band income-amount targets (at minimum AGI amount and return counts from SOI Historic Table 2, which publishes both by state) for bands up to and including $1M+, to the acs-local target set. Count-only or proxy-only (taxable interest) calibration leaves income mass in the top band unconstrained, which fails exactly the progressive-bracket reforms state ballot measures keep proposing. A 1m_plus amount target alone would have caught this: the model would have needed ~4× the income mass in that band.

If per-band amount targets are too aggressive for the sparse solver, an alternative is a single state-level "AGI above $1M" mass target, which directly pins the quantity top-bracket reforms tax.

Reproduction

• Dataset: populace_us_2024_acs_local.h5 @ populace-us-2024-buildp-acs-local-592ae5d6-20260819T020303Z, CO subset (state_fips 8; 30,079 households), policyengine-us 1.825.0, period 2027.
• Reform: gov.contrib.states.co.progressive_income_tax.in_effect (PR PolicyEngine/policyengine-us#9431).
• Dashboard using this estimate: https://github.com/PolicyEngine/co-initiative-195 (validation tab documents the gap).
• SOI reference: Historic Table 2, CO TY2023 (23in06co.xlsx).

🤖 Generated with Claude Code

https://claude.ai/code/session_016aAzQ2DZmw9BX4GpzU7UKP

Start by locating the acs-local target set and its calibration configuration, then compare the Colorado subset against IRS SOI Historic Table 2, especially the 1m_plus band. Add state-by-AGI-band income-amount and return-count targets, or the stated above-$1M mass alternative, and rerun the CO reproduction to confirm the top-tail income mass is constrained.