Update IMA deck to the corrected model by MaxGhenis · Pull Request #22 · PolicyEngine/firm-microsim-paper

MaxGhenis · 2026-07-01T22:33:43Z

Summary

Propagates the corrected-model results from #21 into the IMA World Congress deck:

Reform menu: −£698m / −£466m / −£460m / −£230m on the £184.7bn base; taper ≈ 10% band on cost, metric-scoped wording
Anchor slide: full-deregistration (−£248m) and 43%-retention (−£141m) conventions bracket HMRC's −£185m in every year; uprated-counterfactual baseline mechanism (not "fiscal drag")
Bunching slide rewritten as the two-vintage exhibit (E=0 at £85k; spurious spec-robust E≈196k at the £90k band edge) with the documented correction of the earlier draft's E=8,712 artifact
Behavioural slide: level rise has exactly zero intensive-margin offset at every e; band offsets <4%; appendix table updated
Calibration 89.4%, dominated-band mass ≈144k, sweep £230–260m per £5k step on £204.6bn
All figures refreshed from the corrected pipeline; title aligned with the paper's new title
Layout fix so both bunching panels fit; verified by full-deck headless export (29/29 slides)

If the talk hasn't happened yet, Vahid should re-rehearse from this version — the headline claims changed. If it already ran on July 1, this is the version to share via the QR link.

Depends on / follows #21.

🤖 Generated with Claude Code

All headline numbers propagated: menu -698/-466/-460/-230 on the 184.7bn base; anchor bracketing (-248/-141 vs HMRC -185) with the uprated counterfactual baseline path; sweep 230-260m per 5k step on the 204.6bn base; dominated-band mass 144k; calibration 89.4%. Bunching slide rewritten as the two-vintage exhibit: E=0 at 85k vs the spurious spec-robust E~196k at the 90k band edge, with the documented liability-scaling correction (E=8,712 in the earlier draft was an artifact). Behavioural slide rewritten around the region-confined solver: a level rise has exactly zero intensive-margin offset at every e; reduced-rate offsets under 4%. Figures refreshed from the corrected pipeline; deck title aligned with the paper's new title. Layout: the two-panel bunching column self-stretches so both figures fit. Verified by full-deck headless export (29/29 slides render). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

The 2023-24 population's near-threshold density is now calibrated to the OBR Mar-2023 EFO Chart C £1k-band data (firm-microsim-paper#23), restoring the administratively observed bunching profile as an explicit, cited target. Deck numbers propagated: menu -783/-550/-497/-248 on the 184.8bn base; anchor conventions overshoot HMRC early (-366/-209 vs -185) and match by 2026-27 (-125.2 vs -125), sign flip reproduced; dominated-band mass ~160k; bunching slide retold as target-inheritance (E=14,663 inherited from OBR targets at 85k; placebo-regenerate without them -> 0; spurious ~196k at the uncalibrated 90k band edge); behavioural table raise -783 flat, band offsets <4%. Anchor figure y-limits now scale with the data. Verified by full headless export (29/29). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

MaxGhenis · 2026-07-02T06:06:28Z

Updated to the OBR near-threshold-target build (see #23 / #21). Menu: −783/−550/−497/−248 on £184.8bn. Anchor: conventions overshoot HMRC early (−209 retention vs −185), match by 2026-27 (−125.2 vs −125), sign flip reproduced. Bunching slide now tells the target-inheritance story: E=14,663 at £85k inherited from OBR targets; placebo-regenerate without them → 0; spurious ≈196k at the uncalibrated £90k edge. Full-deck export verified.

🤖 Generated with Claude Code

Menu -784/-550/-497/-249; sweep base 200.9bn; anchor 2026-27 match -125.1 vs -125; calibration 89.2%/92.8%; bunching slide retold: E=7,873 target-inherited at 85k; uncalibrated 90k band edge yields b_LLAT=1.23, within 10% of the administrative estimate - coincidental and 4x spec-unstable. Seam caption removed (seam fixed). Verified by full export. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>