Offshore Leaks — Mapping the Shadow Economy

Graph algorithms over 800k+ offshore entities in Neo4j — finding the "offshore factories"

The question. The ICIJ Offshore Leaks database (Panama Papers and successors) links more than 800,000 offshore entities and 750,000 people and companies across 200+ countries. Relational queries struggle with “who is connected to whom through what” — so we loaded it into a graph database and asked the network itself. UC Berkeley MIDS data engineering final project; I led the team.

Louvain community detection reveals the "offshore factory" pattern: dandelion-shaped clusters where a single intermediary mass-produces shell entities.

How it works. The ICIJ dump is loaded into Neo4j and interrogated with graph algorithms: degree centrality and PageRank to find the load-bearing nodes, then Louvain community detection to expose cluster structure. We paired the graph analysis with NoSQL architecture cases — where MongoDB (heterogeneous future leak ingestion, geo/text search) and Redis (real-time risk leaderboards) would extend an investigative platform.

The finding. Centrality and PageRank both point at Portcullis TrustNet (BVI) Limited — founded by a former Cook Islands Solicitor General who helped write the offshore laws, then sold navigation of them. Community detection around it exposed an assembly line: 80+ entities jointly formed by Portcullis (intermediary) and signed by DirectCorp (officer) — a deliberate, repeatable shell-company production pattern, visible only at the graph level.

Team. Led by me, with Ryan Castillo, Ani Sreekumar, and Jenny Park.

Data note. The ICIJ Offshore Leaks database is publicly available from ICIJ under the Open Database License; we analyze it and link to it rather than redistribute it.

Links. Repository — notebooks, the findings deck, and graph-algorithm results.