# Phase 70 — public-filing descriptive tax case study

## What this pilot measures

Nine separate-statement issuer-years in three purposively selected Korean companies. The inputs are the independently transcribed public filing references from phases68/69. This script does not read the licensed KoTaP CSV, supplied indicators or reconstructed LEV/SIZE. Only the bounded public reference sample is analyzed.

- Accounting expense ratio = income tax expense / pretax accounting profit.
- Cash payment ratio = income taxes paid / pretax accounting profit, using the positive cash outflow convention documented in the source references.
- Gap = cash ratio minus expense ratio, in percentage points.

Neither ratio is a statutory rate or an observed domestic taxable-income base. Expense recognition and cash payments are different accounting quantities. Differences can reflect timing and other reconciliation items; this pilot does not identify their firm-specific causes. Cash payments need not relate solely to the same year's profits. Ratios are not clipped, sign-reversed, imputed or labeled tax avoidance.

## Annual observations

All values below are percentages, except the gap in percentage points. Display rounding does not affect calculation precision.

| Company | Year | Expense / pretax profit | Cash / pretax profit | Cash minus expense (pp) |
|---|---:|---:|---:|---:|
| Dong Wha | 2015 | 6.53 | 22.70 | +16.17 |
| Dong Wha | 2016 | 26.37 | 5.73 | −20.64 |
| Dong Wha | 2017 | 27.92 | 9.37 | −18.55 |
| Hyundai | 2017 | 17.09 | 28.71 | +11.62 |
| Hyundai | 2018 | 47.20 | 87.00 | +39.81 |
| Hyundai | 2019 | 13.24 | 14.11 | +0.88 |
| Samsung | 2016 | 21.36 | 15.17 | −6.19 |
| Samsung | 2017 | 21.17 | 10.37 | −10.79 |
| Samsung | 2018 | 26.09 | 22.86 | −3.23 |

A large cash ratio in a low-profit year should not automatically be interpreted as a change in the statutory rate or evidence of reform effects. We retain these observations and the positive denominator requirement instead of silently trimming them.

## Multi-year calculation changes the result

For each company's observed three-year window, the ratio of sums divides total expense (or cash payments) by total pretax profit. It weights annual ratios by pretax profit. The arithmetic mean gives each year equal weight. These are different descriptive summaries.

| Company and observed window | Expense: ratio of sums | Expense: annual mean | Cash: ratio of sums | Cash: annual mean |
|---|---:|---:|---:|---:|
| Dong Wha, 2015–2017 | 26.20% | 20.27% | 8.91% | 12.60% |
| Hyundai, 2017–2019 | 18.39% | 25.84% | 27.97% | 43.28% |
| Samsung, 2016–2018 | 23.48% | 22.87% | 16.91% | 16.14% |

The windows differ. This table is not a simultaneous three-company ranking and is not an event-study comparison. There is no all-nine-row “national average.” The sharp Hyundai difference illustrates sensitivity to weighting in this particular sample; its cause is not estimated.

## One common calendar year: 2017

All three companies are observed in2017. Summed expense divided by summed pretax profit is **20.86%**; summed cash payments divided by summed pretax profit is **11.79%**, a cash-minus-expense gap of **−9.07pp**.

Equal-company means are **22.06%** and **16.15%**, respectively. Pretax-profit weights are Samsung92.08%, Hyundai7.76%, Dong Wha0.16%. The summed ratio is thus largely driven by Samsung. This is a comparison within three selected separate statements, not a representative Korean corporate tax burden, government revenue measure, confidence interval or forecast.

## Sources and reproducibility

- [Samsung DART receipt20190401004781](https://englishdart.fss.or.kr/dsbh002/viewer.do?rcpNo=20190401004781): separate statements2016–2018, public source amounts in million KRW.
- [Hyundai DART receipt20220217001012](https://englishdart.fss.or.kr/dsbh002/viewer.do?rcpNo=20220217001012): separate statements2017–2019, million KRW, amended filing. It remains distinct from the preliminary AGM notice discussed in phase67.
- [Dong Wha DART receipt20180618000261](https://englishdart.fss.or.kr/dsbh002/viewer.do?rcpNo=20180618000261): separate statements2015–2017, KRW.

The existing reference JSONs contain receipt IDs, units, column locators and snapshot hashes. Phase70 records hashes of both reference files and converts all accounting amounts to KRW before summing. Manual transcription and filing-version caveats remain. This uses publicly reported accounting facts; it does not resolve the broader dataset's licensing.

Run `python scripts/build_public_case_study_v70.py` without network or licensed raw inputs. Outputs are `data/processed/public_case_study_v70.json` and `data/processed/public_case_annual_v70.csv`. Tests cover currency normalization, ratios of sums versus means, duplicate issuer-years, invalid denominators, statement scope and the common-year intersection. Run `python -m unittest discover -s calibration -p 'test_*v[67]*.py'`.

## What follows

This completes a bounded descriptive pilot. Next build a documented bridge between tax expense, current/deferred components and actual cash payments for one issuer, using tax-note and cash-flow disclosures. Only then expand a rule-based sample. Nothing here unlocks a causal reform estimate, a national extrapolation, the full-panel descriptive analysis or fiscal calibration for the automation scenario.