 Fenreg lab is proud to present proposed research into the use of federated machine learning to improve the effectiveness, efficiency and fairness of the global bank secrecy act and anti money laundering regimes. We aim to evaluate the use of federated machine learning in the context of transaction monitoring as one prototype for implementation across BSA and AML disciplines. This research will help to address threshold questions about whether using federated machine learning respects privacy and data localization laws and results in metrics by which we can measure the performance and inclusion effects of the technology. The research will also help to evaluate the development of federated machine learning approaches across BSA and AML disciplines, including an important topic of customer onboarding, where it may be useful to use non traditional information to verify potential customers identities. We see a unique opportunity to reimagine the role of the central bank of the future as one that promotes responsible adoption of advanced technologies and actively supports financial inclusion. Imagine an approach to improving BSA and AML risk detection that leverages sophisticated technology to increase the predictiveness and efficiency of risk detection systems that enables firms and regulators to learn from each other without sharing sensitive or protected data. That enhances firms ability to identify accurately real risks and to reduce unfounded risk reporting and to improve firms risk reward calculus when making decisions about whether to serve specific markets. So what's the global context. The BSA AML framework serves vital national interest in preventing those intent on doing harm, whether that's through terrorism money laundering fraud or human trafficking from using the global financial system for those illegal purposes. But not withstanding enormous investment in and attention to BSA and AML risk detection systems. The system is broken firms invest enormous resources to satisfy BSA and AML requirements, but get little feedback on the quality of their risk reporting. Worldwide firms spend $181 billion on financial crime compliance. However, of the estimated $2 trillion laundered every year globally, we catch less than 1% of it. So what's the formula for de-risking. Two key factors have driven firms out of operating in certain markets. One, the inherent difficulty and additional cost of risk rating activity that occurs outside mainstream forms of finance, especially in connection with money services businesses. Two, the regulatory risks and the reputational impact for firms related to being connected to illicit financial activities. As a result, globally correspondent banking relationships have fallen 25% since 2009. 25% of banks in emerging markets have reported correspondent banking losses. What's the true cost of that. It's the consumers. Consumers are all too often left behind, especially those in jurisdictions that firms categorically refuse to serve because operating there brings too much risk and too little reward. Or when individuals identities cannot be verified using traditional methods. Retrenchment of correspondent banking is enormously important to financial inclusion because correspondent banking relationships relate both directly to remittance transfer services and foreign direct investment. This creates no small problem. Almost 2 billion adults around the world are underbanked, excuse me unbanked. Approximately 30% of the adult population globally does not have an account from which they can safely transact, save or access credit. And importantly, humanitarian organizations have lost access to financial services as a result of de-risking. But the underbanked population globally, if served, could be worth as much as $380 billion to the firms that are able to serve these markets. So what's happening today? Here's an illustration of transaction reporting. Each institution's picture of risk patterns reflects its own experience with illicit activities. Institutions won't necessarily know about patterns that their competitors are picking up, or what government knows about which transactions flagged are actually suspicious or actually real. Firms get little timely feedback on the accuracy of the reports they submit, and the result is that firms lack the most powerful information for improving their risk detection capabilities. That is timely information about confirmed problems. So imagine transaction monitoring that is more effective, efficient and fair. In this scenario, the central bank acts as a utility like hub. Powerful computers combined with smart algorithms could be deployed to evaluate data at different institutions to understand risk patterns. Algorithms would move between the participating firms to pick up the patterns and learn from the risk at each institution. All of this could be done without sharing sensitive or protected data. So what is federated machine learning? The central bank creates a classification algorithm. That algorithm trains on each participating firms data. The central bank develops a key model and model parameters that reflect insights from all participating firms and from the data in possession of the government. The central bank distributes the key model and model parameters to the participating firms while the data stays in each of the institutions. So core benefits. There are many. I'll call out just a few. Federated machine learning expands the data set on which the detection algorithm is trained without exchanging any of that underlying data. The accuracy of the process used to evaluate individuals and businesses at the application stage and to detect illicit transaction activity is markedly improved. The obstacles to collaboration among firms are reduced and firms are able to enter and serve new markets responsibly. So what do these tech solutions offer for financial inclusion goals and the global regulators? First, increased confidence in the effectiveness of transaction monitoring and in those systems abilities to identify real illicit activities. Improved deficiencies for the firms, the institutions and the regulators and greater ability to adhere to jurisdictional data privacy goals and expectations. Finally, an open door for firms to responsibly serve more markets and to let more good customers into the financial system. So how do we get there? We need research now to establish whether federated machine learning promise can be implemented in ways that deliver BSA and AML regimes that are more efficient, more effective and enable greater financial inclusion.