# MacroVision 62 — Descriptive profiles around corporate-tax changes

## Scope and provenance

Executed 26 October 2026 from the existing v53 panel and v61 screened events.
No new observations were ingested and no causal effect was estimated.
Inputs: `data/processed/oecd_cit_panel_2000_2024_v53.csv` (943 country-years,
38 countries) and `data/processed/oecd_event_candidates_v61.csv`.
The selected sample contains 25 isolated events in 18 countries: 5 increases
and 20 decreases. Isolation concerns changes of at least 1 percentage point;
smaller changes and other reforms can remain in the window.

## Method

Seven annual observations per event, t=-3 through t=+3, normalized by subtracting
corporate-tax revenue/GDP at t=-1. Outputs are **percentage points of GDP**,
not percent changes in revenue or elasticities. All 25 events have balanced windows
(175 event-time rows). Group means weight events equally; countries with multiple
events receive more weight. Sensitivity diagnostics additionally omit one event,
omit one country, and give countries equal weight. No standard errors or causal
confidence intervals are reported. Leave-one-out ranges are not confidence intervals.

## Results

| Relative year | Increases: mean | Increases: median | Decreases: mean | Decreases: median |
|---|---:|---:|---:|---:|
| -3 | -0.0889 | -0.0834 | +0.0401 | -0.0124 |
| -2 | -0.0558 | -0.1360 | +0.0417 | +0.0004 |
| -1 | +0.0000 | +0.0000 | +0.0000 | +0.0000 |
| +0 | +0.0060 | +0.1804 | -0.0092 | -0.0176 |
| +1 | +0.0573 | +0.3371 | -0.0417 | -0.0539 |
| +2 | -0.1715 | -0.2565 | -0.0056 | -0.0777 |
| +3 | -0.0127 | +0.1442 | +0.0076 | -0.0602 |

The direction and magnitude vary across horizons and across mean/median summaries.
At t=+1, the five increase events have mean +0.0573 pp of GDP, but omitting one
event moves that mean anywhere from -0.0800 to +0.3299. At t=+2 the increase-group
mean remains negative in all single-event omissions (-0.3065 to -0.0259), which is
still an uncontrolled descriptive pattern, not evidence that a tax rise lowers revenue.
For decreases at t=+1, equal-event weighting gives -0.0417 while equal-country
weighting gives +0.0186. The sign depends on how repeated-country events are weighted.
Before the event, the increase-group means are -0.0889 at t=-3 and -0.0558 at t=-2
relative to t=-1. Such movement motivates investigation; without a comparison group
it is neither a parallel-trends test nor evidence for/against a causal effect.

## Implication for the automation / benefits thesis

This step does not identify the tax elasticity needed to fund benefits. It supports
keeping fiscal calibration blocked and stress-testing assumed tax responses in
scenarios. It neither proves nor disproves that corporate-tax receipts can finance
transfers in an automated economy. Automation, tax bases and household welfare
are not identified by these profiles.

## Next stage (63) — design before estimating

1. Audit individual event histories, including simultaneous base changes and
   special tax regimes; document event exclusions without choosing them based
   on the observed revenue trajectory.
2. Predefine comparison eligibility and a clean-window rule for controls;
   treatment timing and repeated events require explicit handling.
3. Examine treated-versus-control pre-event paths and country contributions.
   Five increase events sharply limit credible precision.
4. Only after those checks select an estimator suited to the design. Do not
   treat a pooled two-way-fixed-effects event study as automatically causal.

## Reproduce and validate

`python scripts/build_event_profiles_v62.py`

`python -m unittest discover -s calibration -p test_event_profiles_v62.py`

The script rejects duplicate panel rows, duplicate events, incomplete/nonfinite
windows, and inconsistent direction labels. Seven tests independently verify raw
panel differences, baseline normalization, group summaries and omission diagnostics,
as well as malformed-input rejection. JSON output disallows NaN/Infinity.