Lower corporate rate from 21% to 15%, with bonus depreciation extended. Published estimates of the rate change alone run $595B (PWBM) to $673B (Tax Foundation) over ten years.
Ten-year deficit effect, conventional+$1,562.8Badds to the deficit
With macroeconomic feedback and debt service+$1,768.5BFRB/US-Lite (Federal Reserve calibrated)
Checked against
Published score +$673.1B; this estimate is 129.6% away (unfitted reconstruction). Unfitted reconstruction, 129.6% from $673B (Tax Foundation / Penn Wharton Budget Model) — no constant is fitted to this figure. Published range [$595B, $673B]: the model's $1.55T is outside it by $872B.
Typical error for this kind of policy: +$933.0B to +$2,192.5B, from 4 out-of-sample corporate cases.
Out-of-sample accuracy, corporate: 4 pre-registered rows miss their published scores by 40.3% on average (median 43.0%), worst 53.4% — 1 of the 4 falls inside the average. On this figure that is +$933.0B to +$2,192.5B typical, +$728.2B to +$2,397.3B at the worst row. Read off the out-of-sample tier only — not the calibrated tiers, whose agreement is by construction. It is the observed spread of this model against published scores for this class of policy, not a confidence interval.
Approximate bracket method: this policy cannot run through the return-level microsimulation, so read the shape as directional rather than as a validated table. Groups are ranked by adjusted gross income per return. Average change is in dollars per return; positive means the group pays more tax.
Economic effects
difference from the CBO baseline
GDP, summed over the window+1.72percent-years
Jobs at the largest effect+233K
Deficit effect of the GDP change-$120.7Bmore tax collected
Deficit effect of interest on the debt+$326.5Bmore debt, more interest
Fiscal year
GDP level
Jobs
Unemployment
10-year rate
GDP change, deficit
Interest, deficit
2026
+0.275%
+176K
-0.14 pt
+0.01 pt
-$19.3B
+$4.5B
2027
+0.364%
+233K
-0.18 pt
+0.03 pt
-$25.4B
+$10.4B
2028
+0.322%
+206K
-0.16 pt
+0.04 pt
-$22.6B
+$16.5B
2029
+0.250%
+160K
-0.12 pt
+0.06 pt
-$17.5B
+$22.6B
2030
+0.181%
+116K
-0.09 pt
+0.08 pt
-$12.7B
+$28.9B
2031
+0.127%
+81K
-0.06 pt
+0.09 pt
-$8.9B
+$35.3B
2032
+0.086%
+55K
-0.04 pt
+0.11 pt
-$6.0B
+$41.9B
2033
+0.057%
+37K
-0.03 pt
+0.13 pt
-$4.0B
+$48.5B
2034
+0.038%
+24K
-0.02 pt
+0.15 pt
-$2.7B
+$55.4B
2035
+0.025%
+16K
-0.01 pt
+0.17 pt
-$1.7B
+$62.5B
FRB/US-Lite (Federal Reserve calibrated). Demand effects fade over five to seven years as monetary policy responds; there is no supply-side channel. The dynamic score is the conventional score, plus the deficit effects of the GDP change and of interest on the debt: +$1,562.8B - $120.7B + $326.5B = +$1,768.5B.
Other scoring models
exploratory
Model
Ten-year deficit effect
Method
CBO-Style
+$1,562.8B
Static + ETI behavioral response + optional FRB/US dynamic feedback
TPC-Microsim Pilot
Cannot score this
Corporate rate / international base changes are firm-level; the return-level microsim pilot has no corporate tax module.
No pilot model returned distributional output; the comparison is aggregate-only.
CBO-Style is the model behind the headline above. The TPC-style microsimulation is a pilot: where the two disagree, the gap tells you how much the answer depends on method. It is not an official range.
Limitations
holdout labels follow the locked post-2026-05-02 regression protocol; they are not retroactive historical out-of-sample claims.
Corporate base-shifting, pass-through spillovers, and international interactions are simplified relative to Treasury and JCT models.
The target prices a broader reform than the model builds. Treasury's FY2025 Green Book chapter (report p. 2) states 'The effective global intangible low-taxed income (GILTI) rate would increase to 14 percent under the proposal', and every edition since FY2023 carries that step inside the corporate-rate row, while the factory scored against it sets gilti_rate_change=0.0. The GILTI leg is never printed separately and cannot be differenced across editions, so the reported error measures a scope mismatch as well as a fit.
The target is a rate-PLUS-GILTI row scored against a rate-only shape, and the figure agrees anyway. Treasury's FY2025 Green Book chapter (report p. 2) says 'The effective global intangible low-taxed income (GILTI) rate would increase to 14 percent under the proposal', and every edition from FY2023 carries that step, while `create_biden_corporate_rate_only` sets `gilti_rate_change=0.0`. The GILTI leg's size is never printed and cannot be recovered by differencing editions, so the 3.7% this row reports measures a scope mismatch as well as a fit and the two cannot be split. Recorded as `line_item_differs` by the 2026-09-05 provenance pass; no constant was retuned, and the target did not move because there is nothing to move it to. See `planning/memos/CORPORATE_PER_POINT_YIELD.md` section 2.
The only rate-only Green Book row is FY2022's, and it is now carried beside this one as `biden_corporate_28_fy2022` — same reform, same factory, a per-point yield 36% lower.
Scored across a window offset: the target is Treasury's FY2022-2031 row on a 2021 baseline and this repository carries no 2021 vintage, so the corporate runner's own FY2025-2034 window is used. In reported mode the rate channel is a flat annual, so the model returns the same figure for either decade - which is the residual this benchmark exists to show, not an artefact of the offset.
WINDOW OFFSET, stated rather than adjusted. The target covers FY2022-2031 on a 2021 baseline and this repository carries no 2021 vintage, so the row is scored on the corporate runner's own FY2025-2034 window - the same mismatch `iija_2021_discretionary` records in the out-of-sample manifest. In `reported` mode the offset costs nothing, because the module's rate channel is a flat annual: shifting the window changes which years the same number is stamped on and nothing else. That invariance is itself the thing this benchmark exists to expose - the module's answer to '21% to 28%' is -$1,397.2B whichever decade is asked about, where Treasury's two editions differ by 36% per point. Nothing here is a claim that the model was run on Treasury's window.
Not fitted to this target and never to be fitted to it. `BASELINE_TAXABLE_PROFITS_BILLIONS` is fitted to `biden_corporate_28`; a second constant fitted here would make the pair uninformative. The residual is the finding: Treasury's FY2022 row implies a marginal base of $1,225.5B per year against the module's $1,996.0B, and the FY2025 row $1,928.5B - a 36% move across two editions of one proposal that a fixed base cannot represent. See `planning/memos/CORPORATE_PER_POINT_YIELD.md` sections 4 and 7.
Scope caveat in the other direction from `biden_corporate_28`'s: this row is rate-only and so is the factory, but its baseline is a 2021 profit level, before the 2021-22 profits boom that took SOI's income subject to tax from $1,780.3B (TY2020) to $2,422.1B (TY2021). Part of the residual is that revision, and the memo does not separate it from the modelling error.
The target is a published range rather than a point: PWBM's +$595B and Tax Foundation's +$673.1B for 21% to 15% across all corporations, both FY2025-2034 conventional, with Tax Foundation's carried as the anchor. Until 2026-09-05 the target was this model's own +$1,920B.
Poor, and reclassified rather than retuned: the corporate module's behavioural offset returned an absolute value, so this rate cut booked 12.5% of its static effect as extra deficit rather than as an offset, and the fitted annual was compensating for that. The sign was corrected 2026-09-05 and the row moved 0.1% -> 22.3%; no constant was moved to close it.
The target is no longer this model's own output. The 2026-09-05 provenance pass superseded +$1,920B - provenance model_estimate, 'no official score' by the scenario's own admission - with the PUBLISHED RANGE [+$595.0B, +$673.1B]: PWBM's 'Lower the corporate income tax rate to 15%' (Table 1, 26 August 2024) and Tax Foundation's Table 2 conventional estimate (17 July 2024), the two conventional scores of the same reform on this repository's own FY2025-2034 window, printed side by side in CRFB's 6 September 2024 post. The anchor is Tax Foundation's, because theirs is a standalone analysis of this one reform where PWBM's is a stacked row inside a whole-campaign package. No constant was retuned to the new figure, so the row moved 22.3% -> 121.6% against a published document instead of 22.3% against the model's own answer.
SCOPE, and it is a third of the residual: neither published figure includes bonus depreciation, which this scenario extends. PWBM prints the business provisions separately at -$623B. The module's bonus-depreciation leg is +$294.15B of its +$1,491.8B, so the rate leg alone reads +$1,197.6B - +77.9% against the anchor rather than +121.6%. Not adjusted away: summing two rows of PWBM's table would be constructing a target rather than reading one.
DIRECTION, and it is the rest: Tax Foundation's Options 2.0 is the only document pricing both directions in one edition and one model, and a point of CUT costs 29% more than a point of increase yields ($163.2B against $126.6B). `corporate.py` prices both at the same per-point rate, which is why this row misses further than `biden_corporate_28` does. See `planning/memos/CORPORATE_PER_POINT_YIELD.md` sections 4b and 5b.