Importance of AI-based Network Analysis in Asset Recovery
"When the $250 million yacht was finally captured in Bali, the United States government couldn't let it bob in the water unattended, so it had to pay for a crew. The $35 million Bombardier jet has been grounded, but it needed an engine test costing up to $25,000. And no one is quite sure what to do with the see-through grand piano now sitting in a supermodel's Malibu home. It won't fit through the door.", (E. A. Harris and A. Stevenson, NYTimes, Dec. 9, 2018)
Asset recovery is one of the hardest tasks at the hands of Law Enforcement Agencies (LEAs). Usually, the problem is much harder than getting a million-dollar piano back from a Victoria's Secret Angel (Not that Miranda doesn't deserve one!). According to World Bank, the amount of money stolen from developing and transition jurisdictions and hidden in foreign jurisdictions each year is approximately $20–$40 billion—a figure equivalent to 20–40 per cent of flows of official development assistance (World Bank, 2007). But still, the amount recovered does not add up to even the amount of law enforcement spending.
But why is it so hard to get the illicit assets back to its owners, and why does it matter in the larger scale of reduction in criminal activity?
How much does crime pay?
According to (Becker, 1968), a crime is committed if the expected utility of a crime is positive, when gains from crime and punishment are considered. The crime utility function U can be written as (Fleming, 2008):
U = y - c - d - P(n+m +pyr)
where y is the total monetary gain from the crime, c is direct costs of crime production, including typical business expenses in terms of wages and purchases, d is the financial equivalent of the distaste for crime, P is the (subjective) probability of conviction, n is fines levied if convicted, m is the monetary equivalent of imprisonment (the disutility of imprisonment per unit time multiplied by its length), p is the probability of successful asset recovery, and r is the discount rate or the ratio of recovered assets.
Since the market mostly determines the cost and revenue of the business, LEAs focus on the conviction and asset recovery rates. Therefore, it can be argued that, if sufficient assets are detected and recovered, the utility function will be negative and crime will not pay. Which brings us to the next question:
How effective is the current asset recovery system?
Not effective at all! According to the Dutch Recovery Agency (CRA), only a small percentage of the cases of asset recovery is finalized, and a mere 22% of the recovery order is recovered by the authorities as shown in the table below (Van Duyne et al., 2014):
Hence in the Netherlands, in almost 18 years between 1995 and 2012, less than 70 million EUR were recovered. Far from being near to the estimations of total illicit assets, it is even arguably less than the cost of the recovery efforts of LEAs. However, the current low percentage of asset recovery is not specific to any single country. The numbers are very similar in other jurisdictions. Detailed statistics in the UK and an excellent overview of issues in existing asset recovery system are presented by researchers at the Northumbria University in (Harvey, 2014).
How can we improve detection with suspicious network analysis?
A recent study in China (Feng, 2019), investigated typical short term transaction patterns in a financial network. These small patterns called the 'motifs' can be thought of as building blocks of a much complex commercial network. The study argues that by using combinations of these motifs, networks such as pyramid schemes can be detected more effectively. To improve suspicious asset detection, perhaps in a similar fashion, the motifs that form the basis of illicit asset hiding activities in a network can be utilized.
How to improve recovery with suspicious network analysis?
Regardless of the detection methods, typically network analysis improves the suspicious entity detection (1) from transactions and CRM features, by investigating the suspicious flow in a network (2). However, some of the entities are not part of the money-laundering activities but just help to hide the assets, hence were not included in the initial detected network. These could be spouses, people with same phone numbers or addresses, and family members, which can be linked with CRM features. Also, the links could be routine one-way transfers of funds, that was somehow left out of the network detection. Extending the suspicious network to related entities as in (3), can help provide information to LEAs to recover more illicit assets. The additional links in (3) are not suspicious activities but rather major asset movements and potential asset hiding links.
Asset recovery is the final step of the anti-money-laundering system. Without its success, regardless of the extend of the detection of criminal activity improvement, criminals will not be affected. Therefore using new technologies and artificial intelligence (AI) in asset recovery is at least as necessary as it is in suspicious activity monitoring.
By Tolga Kurt, Managing Partner at H3M.IO, May 2, 2020
Image by Gerd Altmann from Pixabay