Physicists may say otherwise, but it is trade that makes the world go round — at least financially. From supply chain issues to volatility in prices across asset classes, from stocks to crude oil, trade defines much of the movement in the international economy.
With trillions of dollars moving daily across the financial system, the temptation to indulge in surreptitious behaviour is great.
Regulators, compliance officers and banking leaders have long sought effective tools to combat the increasing sophistication of bad actors, whose wrongdoing frequently leads to billions of dollars in financial losses.
Compliance officers and regulators are looking to identify criminal actions such as insider trading, market manipulation, money laundering, violations of sanctions/export controls and trading in others’ accounts more accurately and quickly.
A recent example is the Libor scandal, where traders from banks colluded to set interest rates favourable to the traders rather than the clients. Other examples include front-running trades ahead of client trades, where the fiduciary duty to clients is not followed.
Identifying suspicious activities requires the appropriate monitoring capabilities, especially given the significant communications and transactions trail that must be evaluated. For example, the Financial Times reported an analysis by Behavox, which showed that just 0.0024% of voice-based communications listened to and 0.0002% of texts analysed by Behavox were flagged as “concerning” in 2021.
Despite their low frequency, the consequences for banks in terms of fines for regulatory and compliance actions are steep. Banks were fined $15 billion worldwide for such violations in 2020 alone.
Artificial intelligence joins the fight for trade compliance
Artificial intelligence (AI) is increasingly being used to fight financial transgressions. AI and associated machine learning (ML) or deep learning models provide regulators and compliance officers with new capabilities. These models can handle various data types, run a suite of advanced analytics and contribute an array of outputs to help remove fraud from international trade.
Deep learning language models create a generational leap forward
Some types of fraud cannot be uncovered using transaction ledgers, financial records and other tabular data alone. As an example of scale, Citi processed 9.4 million transactions in 2018, for about $1 trillion in trade, giving the bank a massive dataset of 25 million pages.
In many cases, fraud happens outside these systems in processes involving unstructured data communications such as audio, images and chats. Here, the final source of record data contains minimal markers for identification, so advanced analytics are required to uncover discrepancies.
No team can effectively read, interpret and flag potential wrongdoing within a data set that size in a realistic timeframe. Hence the need for deep learning models and the accelerated computing infrastructure that enables computers to support trade compliance.
Before accelerated computing, training language and unstructured/semi-structured models would take weeks or months. Now, language and vision models can be trained in hours or days, and their outputs can be delivered in seconds.
Given the real-time movement of money, models must be able to execute in milliseconds to prevent financial crimes. Real-time fraud prevention requires an understanding of spoken language. Not just one language, but multiple, in real-time, with the ability to understand context, describe sentiment, identify entities (businesses, people, etc.) and incorporate all of these complex inputs into a fraud-scoring algorithm.
As data sizes increase exponentially, more sophisticated models are trained, requiring more advanced accelerated computing infrastructures to manage trade compliance effectively. In addition to unstructured data, tabular data can be analysed for activities such as front-running trades, insider trading and collusion.
As international trade expands, the use of AI must follow
The level of international trade will continue to grow in the long term as supply chains strengthen and the pandemic eases. As money flows increase, so will the number of bad actors looking to defraud the system for their own financial gain.
Financial regulators, compliance leaders and bank officers must prioritise investment in AI, the premier tool capable of analysing all of the data (structured and unstructured) that powers financial markets.
Where fraudulent activities are detected, compliance must ensure that a fine balance is maintained and that business units can continue to perform trading functions that generate legitimate profits. AI techniques can screen large amounts of data and identify activities/data with advanced algorithms that require further analysis.
Existing systems are often ineffective or unduly flag large amounts of data from legacy systems, referred to as excess false positives. The reduction in false positives alone will create significant efficiencies and cost savings. It will enable financial entities to analyse reams of communications, trade data, and millions of inputs from thousands of sources in real-time. In the meantime, businesses can continue to trade and make profits, leveraging genuine arbitrage opportunities — a win-win for all market participants.
Originally published by Thomson Reuters © Thomson Reuters.