VaidhyaMegha/finance-kg

Finance Knowledge Graph

69,437 nodes. 149,387 edges. Listed companies and their filings, US banks, their holding companies and failures, Indian legal entities and their parents, exchanges and exchange rates — from public sources — with a generated layer of institutions, customers, accounts, devices, loans and payments on top.

Built with Samyama Graph. Source and loader: samyama-ai/finance-kg.

Two layers: one real, one generated

Layer	Nodes	Edges
Real — public sources (provenance: "real")	56,043	83,041
Generated (provenance: "synthetic")	13,394	66,346
Total	69,437	149,387

Nobody publishes accounts, payments or loans, so they are invented. Every generated node carries provenance: "synthetic" and source: "generated"; customer and watchlist names are made up, and nothing generated is attached to a real organisation. The payment rings, mule accounts, shared devices and structuring alerts in that layer are planted patterns for demonstrating detection, not observations.

Why a graph

Ownership, control and money movement are chains. In a graph they are one traversal:

// Who ultimately owns what: entities consolidated by Tata Sons
MATCH (e:LegalEntity)-[r:CONSOLIDATED_BY]->(p:LegalEntity {name: 'TATA SONS PRIVATE LIMITED'})
RETURN e.name, r.level LIMIT 10

// Bank holding companies by the number of banks they control
MATCH (b:Bank)-[:CONTROLLED_BY]->(h:HoldingCompany)
WITH h, count(b) AS banks RETURN h.name, banks ORDER BY banks DESC LIMIT 10

// Generated layer: open or escalated structuring alerts, the account and its owner
MATCH (al:Alert {kind: 'structuring'})-[:FLAGS]->(a:Account)-[:OWNED_BY]->(c:Customer)
WHERE al.status = 'open' OR al.status = 'escalated'
RETURN a.account_no, c.name, al.transfers, al.amount_usd ORDER BY al.amount_usd DESC LIMIT 10

Files

Nodes carry an id; edges reference those ids as src and tgt. finance-embedded.sgsnap is the complete graph, including embeddings on 38,235 nodes (10 labels) that the CSVs leave out.

nodes/

File	Rows	Layer
legalentity.csv	16,934	real
filing.csv	14,216	real
company.csv	8,004	real
account.csv	5,411	generated
bank.csv	4,231	real
bankfailure.csv	4,117	real
device.csv	3,653	generated
holdingcompany.csv	3,328	real
customer.csv	3,000	generated
exchangerate.csv	2,457	real
exchange.csv	1,931	real
loan.csv	1,200	generated
industry.csv	340	real
country.csv	249	real
currency.csv	178	real
watchlistentry.csv	60	generated
state.csv	55	real
institution.csv	40	generated
alert.csv	30	generated
regulator.csv	3	real

edges/

File	Rows	Connects
transfer.csv	36,681	Account → Account
located_in.csv	23,022	Bank → State, Exchange → Country, LegalEntity → Country
consolidated_by.csv	18,295	LegalEntity → LegalEntity
filed_by.csv	14,216	Filing → Company
listed_on.csv	8,110	Company → Exchange
signed_in_from.csv	5,563	Account → Device
denominated_in.csv	5,411	Account → Currency
held_at.csv	5,411	Account → Institution
owned_by.csv	5,411	Account → Customer
based_in.csv	5,293	Company → Country, Customer → Country, Institution → Country
supervised_by.csv	4,229	Bank → Regulator
occurred_in.csv	4,117	BankFailure → State
controlled_by.csv	3,548	Bank → HoldingCompany
in_industry.csv	2,931	Company → Industry, Customer → Industry
rate_of.csv	2,457	ExchangeRate → Currency
borrowed.csv	1,200	Customer → Loan
disbursed_to.csv	1,200	Loan → Account
issued_by.csv	1,200	Loan → Institution
guarantees.csv	600	Customer → Customer
uses.csv	253	Country → Currency
branch_of.csv	103	LegalEntity → LegalEntity
part_of.csv	55	State → Country
listed_as.csv	40	Customer → WatchlistEntry
flags.csv	30	Alert → Account
failure_of.csv	7	BankFailure → Bank
same_as.csv	4	Company → LegalEntity

Sources and licences

The code is Apache-2.0. The data is not covered by that licence: each source keeps its publisher's terms.

Source	Supplies	Terms
US SEC EDGAR	Companies, tickers, filings, reported figures	Public domain
FDIC	Banks, holding companies, bank failures	Public domain
GLEIF	Indian legal entities and their parents	CC0 1.0
European Central Bank	Exchange rates	Free reuse, ECB cited as the source
World Bank	Country financial indicators	CC BY 4.0
Wikidata	Stock exchanges	CC0 1.0
datasets/country-codes, datasets/currency-codes	Countries and currencies	Public domain
Generated layer	finance-kg generator	Apache-2.0, with the code

Reuse must cite the ECB for exchange rates and the World Bank (CC BY 4.0) for country indicators.

Usage

from datasets import load_dataset
companies = load_dataset("VaidhyaMegha/finance-kg", "company", revision="v1.0")

docker run --rm -p 8080:8080 -p 6379:6379 public.ecr.aws/f9f6l5u4/samyama-graph:1.1.0
curl -X POST localhost:8080/api/snapshot/import -F "file=@finance-embedded.sgsnap"

Vector search and natural-language query also need the tenant's embed_config and nlq_config, which are not part of a snapshot; see the source repository.

Verification

• Totals reconcile against the snapshot header and the source repository: 69,437 nodes, 149,387 edges; real 56,043 / 83,041, generated 13,394 / 66,346.
• finance-embedded.sgsnap sha256 b0957352fcc5d9691fbe1f65be2737dc3e658231d59616f7ebe1fe044e5481b8 (exported 2026-09-30, engine 1.7.1).
• Imported into an empty tenant on engine 1.7.1: all 69,437 nodes read back, and the queries above answer (Wintrust Financial controls 16 banks; the Tata Sons consolidation chain; the open structuring alerts).
• 0 dangling edges. 263 nodes have no edges: 145 companies, 74 exchanges, 22 currencies, 2 countries, and 20 of the 60 generated watchlist entries, which are not linked to any customer.

Limitations

• The generated layer is a demonstration set: its rings, mules and alerts were planted. Do not infer detection rates or real-world prevalence from it.
• LegalEntity comes from GLEIF's Indian entities and their parents and subsidiaries, not the global LEI set.
• Filings and reported figures are a selection, not all of EDGAR; bank failures are the FDIC's historical list.

Citation

Finance Knowledge Graph, v1.0 (69,437 nodes, 149,387 edges). Real layer: SEC EDGAR, FDIC, GLEIF, ECB,
World Bank, Wikidata. Generated layer: synthetic. Built with Samyama Graph.
Source: https://github.com/samyama-ai/finance-kg