---
license: cc-by-4.0
pretty_name: "tbg-kit: synthetic ownership network of Thai-style business groups"
language: [en, th]
size_categories: [10K<n<100K]
task_categories: [graph-ml, tabular-classification]
tags: [business-groups, ultimate-ownership, pyramids, corporate-governance, ownership-network, thailand, synthetic]
configs:
  - config_name: ultimate_owners
    data_files: results/ultimate_owners.csv
    default: true
  - config_name: entities
    data_files: data/processed/entities.csv
  - config_name: holdings
    data_files: data/processed/holdings.csv
  - config_name: board_seats
    data_files: data/processed/board_seats.csv
  - config_name: groups
    data_files: results/groups.csv
  - config_name: group_members
    data_files: results/group_members.csv
  - config_name: fi_nexus
    data_files: results/fi_nexus.csv
---

# tbg-kit: synthetic ownership network of Thai-style business groups

> **Synthetic data.** Every company, person, ticker and number here is fictional, generated by
> [`tbg-kit`](https://github.com/thanatchon36/tbg-kit) (`python -m tbg synth`, seed 42). It mimics the structure of
> Thai public disclosures so the pipeline can be tested and taught; it says nothing about real Thai firms or people.
> Any resemblance of a generated name to a real person or company is coincidental.

Companion dataset to the software [tbg-kit](https://github.com/thanatchon36/tbg-kit) and the working paper
*Mapping Thailand's Business Groups: Ownership, Control, and the Finance-Business Nexus* (Thanatchon Chongmankhong).
It contains one complete run of the pipeline on the synthetic world: the raw inputs as extraction leaves them, the
resolved ownership network, and the analysis (ultimate owners, business groups, pyramids, bank–group links).
Explore the same network online: <https://thanatchon36.github.io/tbg-kit/explorer.html>.

## Summary

- 12,063 entities: 527 listed firms, 7,244 companies, 3,485 persons in 2,783 families, 782 foreign companies
- 13,267 ownership edges and 3,019 board seats
- 421 multi-firm business groups, 46 with several listed firms
- At the 20% threshold, 63% of listed firms are family-controlled, 17% widely held, 4% state, 12% unresolved

## Files

| Path | Rows | Content |
|---|---:|---|
| `data/raw/universe/listed_companies.csv` | 527 | Listed companies (SET + mai style universe) |
| `data/raw/shareholders/top10.csv` | 5,158 | Top-10 shareholder registers |
| `data/raw/subsidiaries/extracted.csv` | 9,419 | Subsidiary tables as extracted from One Reports |
| `data/raw/directors/directors.csv` | 3,026 | Board members |
| `data/raw/dbd/boj5.csv` | 171 | BOJ.5 shareholder lists of private holding companies |
| `data/raw/fi/fi_licenses.csv` | 58 | Financial-institution licences |
| `data/raw/gleif/th_entities.csv` | 58 | GLEIF legal entities |
| `data/raw/gleif/relationships.csv` | 42 | GLEIF parent relationships |
| `data/manual/seeds.yaml` |  | Group labels (labels only) |
| `data/processed/entities.csv` | 12,063 | Resolved entities: companies, persons, families, funds |
| `data/processed/holdings.csv` | 13,267 | Ownership edges after entity resolution |
| `data/processed/board_seats.csv` | 3,019 | Person–company board seats |
| `data/processed/parent_links.csv` | 42 | Financial-group and GLEIF parent links |
| `data/processed/person_family.csv` | 3,485 | Person → family assignment |
| `results/ultimate_owners.csv` | 25,548 | Ultimate owner, control and cash-flow rights per firm and threshold |
| `results/control_pairs.csv` | 40,204 | All owner–firm pairs with control and cash-flow rights |
| `results/groups.csv` | 544 | Business groups with size, depth, wedge, market value |
| `results/group_members.csv` | 8,852 | Group membership of every entity |
| `results/group_ties.csv` | 3,603 | Shared directors and cross-ownership between groups |
| `results/fi_nexus.csv` | 766 | Bank group ↔ business group links |
| `results/cycles.csv` | 3 | Cross-holding cycles |
| `results/robustness.csv` | 27 | Results under alternative thresholds and settings |
| `results/summary.json` |  | Headline numbers used in the paper |

## Load

```python
from datasets import load_dataset
uo = load_dataset("thanatchon36/tbg-kit", "ultimate_owners", split="train")
holdings = load_dataset("thanatchon36/tbg-kit", "holdings", split="train")
```

## Key definitions

- `holdings.pct`: direct ownership of `owner` in `owned`, in percent. Edges come from SET top-10 registers, One Report
  subsidiary tables, DBD BOJ.5 lists, Pillar 3 reports and manual input (`source`).
- `ultimate_owners`: per firm and control threshold (10%, 20%, 50%), the ultimate owner `uo_id`, control rights `cr`
  (block-wise weakest link), cash-flow rights `cf` (M(I−M)⁻¹), `wedge = cr − cf`, chain `depth` and pyramid flags.
- `groups` / `group_members`: business groups around each ultimate owner at the 20% threshold.
- `fi_nexus`: links between bank groups and business groups through ownership, shared directors and fund holdings.

## Reproduce

```bash
git clone https://github.com/thanatchon36/tbg-kit && cd tbg-kit
make setup && make hf        # writes this folder to dist/hf/
```

`SHA256SUMS` lists the checksum of every file.

## License and citation

Data: CC BY 4.0. Software: MIT. Cite the software with its
[`CITATION.cff`](https://github.com/thanatchon36/tbg-kit/blob/main/CITATION.cff) (DOI on Zenodo).