# MacroVision 64 — Institutional identification audit

26 October 2026. Scope: KOR 2018, SVN 2017, and every candidate control from the
focused v63 comparison sets (13 unique countries, 19 event/control memberships).
This is an issue-focused source audit. It is not a complete legal history or a
certification that every policy change has been found.

## Decision

**The current country-year design does not establish an isolated rate-only causal
effect.** Both treated cases bundle other policies and have prior-year anticipation
risk. Eighteen control memberships have a documented policy or measurement risk;
one (FIN in KOR 2018) has unresolved legal detail. No control or treated event is
approved for causal calibration. A documented issue is not proof that a country
cannot be used in any redesigned analysis; a suitable model may explicitly account
for some issues. No claim is made about the sign or size of those confounders.

No new regression, post-outcome matching or revised revenue estimate is produced.
The v53 panel and v61-v63 selections remain unchanged. Existing comparison gaps
are not relabelled causal effects. The audit script does not read revenue outcomes.

## Treated cases and timing

### Korea 2018

An official ministry announcement dated 2 August 2017 describes the tax package
for the following year. OECD's 2018 survey documents the top-bracket CIT rise,
reduced large-firm R&D credits and a changed investment/mutual-growth tax, alongside
PIT changes and a minimum-wage increase. These jointly affect incentives, taxable
profits or macro conditions; a country-year rate indicator does not disentangle them.

The public 2017 announcement means the v62/v63 t=-1 normalization is potentially
anticipatory. This is an identification risk, not evidence that anticipation actually
changed receipts. A package effect would be a different estimand from the effect
of changing the top CIT rate alone. The effective calendar date and firm tax-year
alignment need tax-return-level verification, including collection lags.

### Slovenia 2017

OECD's reform chronology identifies the 2016 package taking effect in January 2017:
a general CIT increase, PIT bracket/allowance changes and performance-bonus tax
changes. This was a change in the tax mix, not a single isolated policy lever.
The prior-year legislative/package timing similarly puts the 2016 baseline under
an anticipation-risk flag. The exact enactment/announcement timeline and full
base provisions are not certified by this issue screen.

### Pandemic coverage

The original +/-3 windows reach 2021 for Korea and 2020 for Slovenia, so both include
COVID-era outcomes and policy interventions. Truncating at 2019 leaves 2018-19
for Korea (event year and t=+1) and 2017-19 for Slovenia (event year through t=+2).
This may avoid the pandemic period, but cannot resolve bundled treatment or a
potentially anticipatory baseline. No truncation is silently applied to old results.

## Every candidate control was screened

| Country | In comparison set | Finding in relevant window | Evidence status |
|---|---|---|---|
| AUS | KOR | Small-business CIT reduction despite unchanged general 30% rate | Statutory amendment, income-year timing recorded |
| AUT | Both | Research premium increases in 2016 and 2018 | Finance-ministry guidance |
| CHE | SVN | 2020 TRAF regime/base reform, patent box and R&D mechanisms | Federal finance authority |
| CRI | KOR | 2019 fiscal package, VAT/income-tax changes and fiscal rule | OECD official survey |
| CZE | Both | Interest-deduction restriction effective April 2019 | Tax-administration notice |
| FIN | KOR | ATAD-related Act 1237/2018 identified; scope/effective-date detail incomplete | Unresolved; never treated as clean |
| IRL | SVN | Exceptional 2015 GDP/globalisation shift affects denominator comparability | CSO historical release |
| ISL | Both | R&D credit expenditure ceiling expanded in OECD 2019 reporting period | Policy documented; exact legal timing pending |
| LTU | Both | Investment-project profit deduction expanded from 2018 tax periods | Tax authority's statutory text |
| MEX | KOR | Border-region CIT credit for qualifying firms | OECD reform report |
| NLD | Both | Earnings-stripping deduction restriction implemented January 2019 | Parliamentary/finance-ministry record |
| NZL | KOR | BEPS restrictions on inbound related-party interest | Inland Revenue technical bulletin |
| POL | Both | SME CIT cuts in 2017 and 2019 hidden by constant general rate | Finance-ministry historical summary |

The records cover 19 memberships because six countries are shared between the two
comparison sets. Twelve unique controls have documented identification risks; one
has unresolved legal verification. Timing precision and source locations are stored
per finding in `data/metadata/institutional_audit_v64.json`. These are issues to
model or investigate, not quantified revenue effects or automatic universal exclusions.
For example, an R&D credit may have a different budgetary classification across
systems; the audit does not assume a one-for-one change in recorded CIT receipts.

The Irish 26.3% real-growth figure belongs to the historical CSO release vintage.
It flags a multinational-related denominator issue; it is not used to overwrite
current panel GDP, nor confused with a nominal GDP percentage change.

## Audit outputs and safeguards

- `data/metadata/institutional_audit_v64.json`: source register and dated policy findings.
- `data/processed/oecd_institutional_audit_v64.json`: case-level diagnosis and decision.
- `data/processed/oecd_control_audit_v64.csv`: all 19 memberships with explicit status.
- `scripts/audit_institutional_controls_v64.py`: deterministic source-to-membership audit.

Unknown evidence is never labelled clean. Missing sources, inconsistent dates,
measured-impact claims, invalid membership and duplicates are rejected. Source URLs,
locators and retrieval dates are retained, but full original documents are **not**
archived; this remains a provenance limitation. No complete legal audit is claimed
for Finland, for the exact Iceland reform dates, or for the other unreviewed events
outside these two focused cases. The 19 other treated events noted in v63 remain
outside this focused review.

Six new tests check complete focused membership coverage, announcement and pandemic
flags, missing-evidence handling, window boundaries, reference integrity and membership
integrity. The 20 combined v62-v64 tests pass locally.

## Implication for benefits financing

This evidence does not disprove corporate taxation as a funding source. It shows
why this country-year sample cannot yet calibrate its causal fiscal response. Keep
`calibration_allowed=false` and `causal_identification_allowed=false`.

## Next research decision (65)

The priority is a data-availability and estimand decision:

1. For a rate-only effect, assess firm-level exposure data: Korea's taxable-profit
   thresholds, firm tax-year timing, affected versus unaffected firms, and concurrent
   R&D/mutual-growth exposure. Threshold manipulation and endogenous exposure need
   a design of their own; a discontinuity is not automatically valid.
2. For Slovenia, determine whether a bundled tax-mix effect is the actual estimand.
   That requires a newly specified comparison design and macro controls, rather
   than interpreting the current CIT coefficient as a rate elasticity.
3. If suitable exposure data cannot be obtained, preserve the fiscal model as
   scenario analysis with explicitly assumed response ranges. Seek other historical
   episodes before claiming an empirically calibrated tax response.

Do not remove controls until a preferred result appears, and do not promote a small
pre-event gap into a certification of parallel trends.

## Reproduce

`python scripts/audit_institutional_controls_v64.py`

`python -m unittest discover -s calibration -p 'test_*v6*.py'`