In network science, it’s fairly common to take a graph of something and be able to apply techniques like neighbourhood analysis or clustering which then reveal something about the data. This comes up in all kinds of real world examples from social networks to music preferences etc. What the authors are saying (I think) is that they tried the same tricks on the Panama data and got nothing useful out of it. The standard graph techniques don’t explain anything useful about what’s going on. I would guess (from previous experience working on similar things at opencorporates.com a few years ago) that this is because these networks aren’t constructed according to the real world - instead they’re a response to specific tax rules. For example if you use a specific tax avoidance scheme like the Irish dog leg then the company structure will be defined by the laws of the countries involved. Another reason would be that intermediaries like company formation agents are likely to show connections where they don’t exist in reality. If I use the same company formation agent as say, Donald Trump, it doesn’t follow that we’re related financially. Finally the Panama papers were just about hiding assets in many cases, so you’ll have these chains of ownership going through as many countries as they could be bothered to work with. This is just to give law enforcement etc. a hard time. Again I wouldn’t expect graph analysis to make sense of that on its own. If anyone is interested, banks in the US are required to document their corporate structures so there’s good data available on their structures. It suffers from the same issues though - it’s hard to tell from looking at a graph of companies where the bad stuff is happening.