What is a transactions graph and how to use it
Every graph has two main components – nodes (also called vertices) and edges. Nodes are the objects that are linked together by the edges. In the graph for transactions analysis the nodes represent bank’s clients and their external partners (these are the clients of the other banks who pay or receive money from the clients) and the edges represent payments or transactions between the clients inside the bank or between the clients and external partners. Graphs could be directed, i.e. edges have the direction and undirected when edge shows only the connection between edges without information about the direction. For the transactions graph it is important to know who is the payer and who is the receiver of the money thus it is more useful to keep it directed. All the transactions between any pair of the clients or the client and his partner are grouped together and are shown as a single edge to make the graph more readable but keep all the information.
To get the first impression about the client and his activity with the help of a graph we need to take a few steps:
Choose the period for the transactions to explore, e.g. one year
Choose the depth of the connections – how distant clients we want to see in the graph. Distance is defined by the number of hops from the client of interest. Usually it is enough to limit the graph by two hops, then there’ll be the central client, his direct neighbors and the neighbors of neighbors
Choose the money cut-off or the threshold for the turnover between the nodes. Small turnover usually does not bear significant money laundering risks and the reduction of the insignificant edges makes the graph more readable. Threshold could be set as a percentage of the client’s turnover to keep all the edges at the scale of this client
Applying the above mentioned criteria to the transactions we can create the graph that shows all the cash flows of the client, his direct neighbors and neighbors of the neighbors. Just looking at this graph you can get the expression of how this client works:
who are his main partners
what is the geography of the payments
what is the main business of the partners of the client (if payments are categorized)
is this client connected with other clients (directly or via common external partners)
does the client have previously marked as high risk / suspicious clients or partners in the direct or indirect neighbors
is there something suspicious in the cash flows: long chains when money go from client to client multiple times; densely connected groups of clients; circle paths when money go in close ring
The graph like described above can be created in a seconds with the proper software thus saving many hours to AML officers. Beside that there are multiple risk ratios derived from the graph that can be used in the AML risk scoring system or on a standalone basis to run automated screening of the clients.