Summary
Stratifying the Phase-18 test fold by SIC division (first SIC digit) shows that the pooled AUROC of 0.711 is broadly distributed across the three reliable divisions — but with meaningful cell-to-cell variation.
Reliability flags (predeclared)
- RELIABLE — n_pos ≥ 20 strict positives in the cell.
- NO_STRONG_CLAIM — n_pos in [10, 20).
- DESCRIPTIVE_ONLY — n_pos < 10.
Strict-label results by SIC division
| SIC division | Name | n_rows | n_pos | Base rate | AUROC | Lift @ 10% | Quality |
|---|---|---|---|---|---|---|---|
| — | Pooled | 19,410 | 231 | 0.0119 | 0.711 | 2.21× | RELIABLE |
| 0 | Agriculture, Forestry, Fishing | 133 | 5 | 0.0376 | 0.653 | 2.05× | DESCRIPTIVE_ONLY |
| 1 | Mining + Construction | 1,237 | 6 | 0.0048 | 0.706 | 0.00× | DESCRIPTIVE_ONLY |
| 2 | Manufacturing (food / textiles / chem / petr) | 4,933 | 85 | 0.0172 | 0.729 | 2.47× | RELIABLE |
| 3 | Manufacturing (durables / electronics) | 4,960 | 53 | 0.0107 | 0.639 | 0.57× | RELIABLE |
| 4 | Transportation, Communications, Utilities | 1,463 | 15 | 0.0103 | 0.691 | 0.00× | NO_STRONG_CLAIM |
| 5 | Wholesale + Retail Trade | 1,654 | 30 | 0.0181 | 0.742 | 2.34× | RELIABLE |
| 6 | Finance / Real Estate residual | 810 | 6 | 0.0074 | 0.917 | 8.33× | DESCRIPTIVE_ONLY |
| 7 | Services | 3,280 | 18 | 0.0055 | 0.861 | 3.89× | NO_STRONG_CLAIM |
| 8 | Health / Education / Social Services | 936 | 13 | 0.0139 | 0.648 | 2.30× | NO_STRONG_CLAIM |
Reads
- The three RELIABLE divisions span AUROC 0.64 – 0.74; pooled AUROC 0.711 sits in the mid-range.
- Manufacturing-2 (food / textiles / chem / petr) and Wholesale + Retail Trade are the cells where structural distress is most clearly above-chance and consistent with the pooled signal.
- Manufacturing-3 (durables / electronics) is the noticeably weaker reliable cell. Possible reasons: durables / electronics distress often shows distinct patterns the G-aggregate smooths, and many distress events resolve via M&A or going-private rather than Chapter 11 / Item 1.03 distress.
- Flashy AUROCs in DESCRIPTIVE_ONLY cells (Finance / RE residual 0.92 on 6 positives; Services 0.86 on 18) reflect very small N and should not be over-interpreted.
Source basis: public summary of
CORPORATE_PHASE19_SECTOR_STRATIFIED_ANALYSIS.md
and processed/phase19_sector_stratified_results.csv
in the project repository.
Back to Validation · Evidence · Documents index.