Latest Research Projects

Schematic overview of the layered approach proposed in this paper for transaction monitoring in HVPS

Finding a Needle in a Haystack: A ML Framework for Anomaly Detection in Payment Systems

This paper introduces an innovative machine learning (ML) framework for real-time transaction monitoring in crucial financial infrastructures. Designed to detect anomalies amid the vast daily volume of transactions and the scarcity of pre-labeled examples of anomalous transactions, our layered approach combines supervised and unsupervised ML techniques. This provides payment system operators and overseers with new tools for safeguarding financial systems.  (full paper link)

Word cloud for the titles of the articles in prominent Econ journals that use ML

Machine Learning for Economics Research: When What and How?

This article provides a curated review of selected papers published in prominent economics journals that use machine learning (ML). The review focuses on three key questions: (1) when ML is used in economics, (2) what ML models are commonly preferred, and (3) how they are used for economic applications. The review highlights the increasing complexity of economic data due to rapid digitalization and the growing literature suggesting that ML is becoming an essential addition to the econometrician’s toolbox. (full paper link)

Interbank payments network for LVTS (left) and Lynx (right)

From LVTS to Lynx: Quantitative Assessment of Payment System Transition

In this project, we assess the impact of Modernizing Canada's wholesale payments system on the behaviour of participants, with a focus on two key changes brought by the shift from LVTS to Lynx. Firstly, we will examine the effects of moving from a hybrid settlement model that combined RTGS and DNS to an RTGS-only system. Secondly, we will investigate the implications of a policy change from discouraging queue usage to encouraging it. Joint with Zhentong Lu, Hiru Rodrigo, Jacob Sharples, Phoebe Tian, and Nellie Zhang (full paper link)

Quantum algorithm, Combinatorial optimization, NP-hard problem, payments system

Proposed schematic of the quantum optimizer as a pre-processor to the payment system

Improving efficiency of payments systems using quantum computing

In this project, we developed an algorithm and ran it on a hybrid quantum annealing solver to find an ordering of payments that reduced the amount of system liquidity necessary without substantially increasing payment delays. Despite the limitations in size and speed of today’s quantum computers, our algorithm provided quantifiable efficiency improvements when applied to the Canadian HVPS using a 30-day sample. Joint with Bank of Canada's PIVOT team and Goodlabs Studio (full paper link)

payments data, machine learning, cross-validation, interpretability, overfitting

Schematic of proposed expanding window approach for cross-validation in time series 

Macroeconomic predictions using payments and machine learning

In this project, we to demonstrate that non-traditional and timely data such as retail and wholesale payments, with the aid of nonlinear machine learning approaches, can provide policymakers with sophisticated models to accurately estimate key macroeconomic indicators in near real-time. Moreover, we provide a set of econometric tools to mitigate overfitting and interpretability challenges in machine learning models to improve their effectiveness for policy use. Joint with J Chapman (full paper link)

reinforcement learning, payment systems, realizing application of RL

Schematic of reinforcement learning in the context of a high-value payments system 

Estimating policy functions in payments systems using reinforcement learning

In this project, we use reinforcement learning (RL) to approximate the policy rules of banks participating in a high-value payments system (HVPS). The objective of the RL agents is to learn a policy function for the choice of amount of liquidity provided to the system at the beginning of the day. Our results show the potential of RL to solve liquidity management problems in HVPS and provide new tools to assist policymakers in their mandates of ensuring safety and improving the efficiency of payment systems. Joint with PS Castro, H Du, R Garratt, F Rivadeneyra (full paper link)

payments data, COVID-19, macroeconomic predictions, financial crisis, machine learning

Comparison of the monthly aggregated Encoded Paper steam during Covid-19

Using payments data to nowcast macroeconomic variables during the onset of COVID-19

In this project, we develop a model to predict the current state of the economy—nowcasting—using retail payments system data and machine learning. The Canadian retail payments data aid in understanding the current state of the economy because they include many types of transactions and are available daily. These data features are ideal for macroeconomic nowcasting during a crisis. The flexibility of machine learning can help capture the large and nonlinear effects of the COVID-19 shock. We find that our model has a significant increase in prediction accuracy. Joint with J Chapman (full paper link)

stochastic PDEs, Uncertainty Quantification, HPC, UQTk, FEniCS, PETSc, MPI, GMSH

Implementational framework of the solvers

Scalable solvers for uncertainty quantification of stochastic PDEs in high-performance computing

In this project, we develop parallel scalable solvers using stochastic finite element methods in conjunction with domain decomposition methods, and implemented using libraries, objects, and routines from MPI, PETSc, FEniCS, UQTk, GMSH, ParaView. Our solver are exercised in uncertainty propagation of large-scale and high-dimensional stochastic PDEs---system matrix with up to 215 million DoFs---on compute-Canada high-performance computing cluster with up to 5000 cores. Joint with M Khalil, C Pettit, D Poirel, A Sarkar (full paper link)