Introduction

The Operational Efficiency in Finance Laboratory (OEFin Lab) is a research facility established under the Institute for Digital Technology and Economy (BK Fintech), Hanoi University of Science and Technology. The laboratory focuses on optimization, applied mathematics, operational optimization, finance, and business applications.

The development of information technology has triggered the digital transformation revolution in the financial sector and reshaped the competitive landscape of the banking industry. This transformation has profoundly affected the operational efficiency of many state-owned and commercial banks. The increasing banking competition has disrupted traditional market competitive structures, creating stronger impacts on banks' operational performance.

The global financial industry is undergoing comprehensive integration of digital transformation technologies as well as adoption of new technologies such as artificial intelligence, cloud computing, big data, and the Internet of Things. This integration is driving a new technological revolution that is reshaping business operations. The digital transformation revolution in finance has promoted the digitization of traditional finance, with banks playing important roles in this process.

Unlike traditional financial services that rely mainly on financial institutions and physical branches to conduct business operations, digital transformation in finance relies on online services, core digital technologies such as mobile payments and big data analytics, and has strong spatial and temporal independence. It breaks the public's dependence on physical branches and physical environments in traditional financial institutions, reduces the threshold for people, organizations, and businesses to access financial services.

Research Directions

1. Predictive Control Optimization Models for Smart Logistics Operations under Demand and Travel-Time Uncertainty

• • Research question: How can spatio-temporal forecasting be integrated with real-time re-optimization—including routing, inventory management, and fleet coordination—to simultaneously minimize costs, service-level violations, and emissions while maintaining feasibility when forecasts are inaccurate?
  • Objective: Develop a predictive control framework for multi-echelon supply chains in which forecasting models are trained using actual operational losses rather than statistical errors, together with probabilistic service-level guarantees formulated as chance constraints.
  • Key research areas: Modeling dynamic routing and inventory problems under stochastic demand and travel times; spatio-temporal forecasting across transportation networks; rolling-horizon re-optimization and recovery mechanisms for unexpected events; learning combinatorial algorithms guided by operations research knowledge; balancing the multiple objectives of cost, service quality, and emissions; developing a digital twin to validate policies before deployment; and conducting experiments on last-mile delivery problems in Vietnamese cities.
  • Technologies and techniques: Stochastic and robust optimization, chance-constrained programming, model predictive control (MPC), decision-focused learning, spatio-temporal graph neural networks, neural combinatorial optimization, reinforcement learning, column generation, OR-Tools, simulation optimization, and multi-objective optimization.

(PhD candidate profile: Applicants should have a background in operations research, applied mathematics, computer science, or industrial and systems engineering, as well as experience with optimization solvers and heuristic methods. Experience working with transportation or GPS data is preferred) - Tran Ngoc Thang

2. Heterogeneous Temporal Neural Network Models for Approximating the Search Space of Multi-Objective Evolutionary Algorithms, with Applications to Multi-Objective Smart Logistics Operations

• • Research question: Can a heterogeneous temporal representation of a population across generations be learned to approximate evolutionary operators and evaluation functions, thereby substantially reducing the number of expensive objective-function evaluations required to approach the Pareto front in computationally intensive optimization problems such as combinatorial optimization?
  • Objective: Develop a heterogeneous temporal graph neural network that serves both as an uncertainty-aware surrogate model and a learned offspring-generation operator. The model should be compatible with widely used multi-objective evolutionary algorithm (MOEA) frameworks while providing guarantees regarding diversity and convergence.
  • Key research areas: Representing populations and generations as dynamic heterogeneous graphs with different node types—including individuals, objectives, constraints, and reference points—and edge types representing dominance, neighborhood, and parent–offspring relationships; training models to predict objective values, dominance rankings, and the potential of offspring, together with uncertainty estimates; designing learned crossover, mutation, and environmental-selection operators; developing mechanisms to prevent diversity collapse and strategies for surrogate-model management; evaluating the proposed methods using DTLZ, WFG, and RE benchmark suites; and applying them to multi-objective neural architecture search, simulation-based engineering design, and Pareto-front generation for mixture-of-experts (MoE) models.
  • Technologies and techniques: Heterogeneous temporal graph neural networks, graph transformers, permutation-invariant networks, surrogate-assisted evolutionary multi-objective optimization, Bayesian optimization, Gaussian processes, uncertainty quantification, NSGA-III, MOEA/D, learning to optimize, and generative models for decision spaces.

(PhD candidate profile: Applicants should have a background in evolutionary optimization and machine learning, together with strong programming skills and experience organizing large-scale experiments. Publications or experience involving multi-objective optimization benchmark suites are preferred) - Tran Ngoc Thang

3. Fisher-Geometric Statistical Monitoring for Decision-Focused Demand Forecasting under Distribution Shifts, with Applications to Risk-Aware Inventory Control

• • Research question: How can a Fisher-geometry-based monitoring mechanism be developed to detect distribution shifts that are meaningful to neural demand-forecasting models, distinguish harmless drift from drift that increases forecasting and inventory-decision risks, and determine the appropriate level of model adaptation?
  • Objective: Establish theoretical foundations and develop algorithms for adaptive, decision-focused forecasting in which Fisher–Rao geometry is used to quantify changes in forecast distributions under temporal dependence; link the magnitude of distributional drift to forecasting risk and inventory-decision risk; and select the optimal, cost-aware adaptation action.
  • Key research areas: Developing Fisher-geometric drift statistics for neural forecasting; adjusting for temporal dependence and decomposing drift by forecasting horizon and model layer; developing mechanisms to confirm harmful drift based on forecasting loss and inventory costs; designing adaptation actions such as retaining, recalibrating, fine-tuning, retraining, or switching models; integrating forecasting with inventory control that accounts for shortage costs, overstocking costs, and Conditional Value at Risk (CVaR); and establishing guarantees regarding false alarms, stability, and dynamic regret based on Fisher-path variation.

(PhD candidate profile: Applicants should have a strong background in probability and statistics, sound mathematical reasoning, programming skills, and an interest in research at the intersection of mathematical theory and artificial intelligence) - Nguyen Thi Ngoc Anh

4. Statistical Decision Intelligence for Financial Operations under Distribution Shift and Dynamic Risk

• • Research question: How can statistical methods be developed to detect, quantify, and forecast changes in financial distributions and risks over time, while determining when such changes are sufficiently significant to require adjustments to operational decisions?
  • Objective: Develop a Statistical Decision Intelligence for Financial Operations framework that connects mathematical statistics, machine learning, and optimization to support the progression from distribution-shift detection to risk quantification and optimal decision selection. The research aims to establish mathematical guarantees concerning error, stability, risk, and regret in non-stationary environments.
  • Key research areas: The research focuses on three pillars:
  • Dynamic Statistical Monitoring: Sequential inference, change-point detection, confidence sequences, and Fisher geometry for early warning.
  • Risk-Aware Statistical Learning: Modeling distribution shifts, tail risk, Value at Risk (VaR), Conditional Value at Risk (CVaR), and uncertainty.
  • Forecast-to-Decision Optimization: Connecting forecasts to decisions through statistical decision theory, stochastic and distributionally robust optimization, and dynamic regret, with applications in credit, liquidity, capital allocation, and financial risk management.

(PhD candidate profile: Applicants should have a strong background in probability and statistics, sound mathematical reasoning, programming skills, and an interest in research combining mathematical theory with artificial intelligence) - Nguyen Thi Ngoc Anh

5. Real-Time Statistical Monitoring for the Early Detection of Economic Cycle Turning Points under Data Publication Lags and Revisions

• • Research question: How can a sequential statistical monitoring mechanism be developed to detect changes in economic conditions—such as a transition from growth to contraction—at an early stage when economic data are released asynchronously, frequently revised, and the information available at the time of decision-making differs from the final data observed later?
  • Objective: Develop a statistical process monitoring framework for real-time economic data in which monitoring signals follow the strict-vintage principle and use only the information actually available at each point in time. The research will compare and develop fast-warning mechanisms, exponentially weighted moving average (EWMA), cumulative sum (CUSUM), and multivariate monitoring methods; establish threshold-calibration mechanisms that enable these methods to maintain comparable false-alarm rates; and assess their ability to detect economic downturns at an early stage.
  • Key research areas: The research focuses on four components:
    Evaluation criteria include detection rates, early-warning lead time, detection delay, false-alarm probability, and robustness to data revisions.
  • • (1) Constructing vintage datasets and features that reflect deterioration, momentum, and common signals across multiple economies or indicators.
    • (2) Developing sequential monitoring statistics and control charts suitable for temporally dependent data.
    • (3) Calibrating warning thresholds through bootstrap methods or simulation, with control of false-alarm probability over a finite horizon instead of directly applying classical statistical process control assumptions.
    • (4) Conducting evaluations using both Monte Carlo simulations and real economic data, such as the OECD Composite Leading Indicators, industrial production data, and similar macroeconomic indicators.

(PhD candidate profile: Applicants should have a background in probability and statistics, time-series analysis, or econometrics, along with proficiency in Python or R. This topic is suitable for candidates interested in research at the intersection of statistical process monitoring, time-series analysis, real-time macroeconomic data, and decision support) - Nguyen Huu Du

6. Sequential Statistical Monitoring for Change and Risk Detection in Financial Markets

• • Research question: How can sequential statistical monitoring methods be developed to detect meaningful changes in financial market conditions and risk levels at an early stage when the data exhibit temporal dependence, non-normal distributions, heavy tails, and extreme observations?
  • Objective: Develop a statistical process monitoring framework for financial time series that combines robust and nonparametric statistics with sequential monitoring mechanisms to improve the detection of small, persistent, or anomalous changes in risk distributions. The research will also develop threshold-calibration mechanisms to control false alarms under dependent and non-Gaussian data conditions.
  • Key research areas: Investigating control charts and sequential monitoring methods suitable for financial data; developing statistics based on ranks, U-statistics, robust scores, or risk measures; studying information-accumulation mechanisms such as EWMA, CUSUM, and their extensions; developing calibration methods under serial dependence and heavy-tailed distributions; and conducting evaluations through theoretical analysis, Monte Carlo simulations, and real-world data such as stock returns, market indices, volatility measures, and other financial risk indicators.

(PhD and master’s candidate profiles: Applicants should have a background in probability and statistics, statistical inference, and time-series analysis, together with proficiency in R or Python. This topic is suitable for candidates interested in research at the intersection of statistical process monitoring, sequential analysis, and financial time series) - Nguyen Huu Du

7. Simulation of Transportation Systems and Electricity Supply for Electric Vehicles Using GIS Data and Multi-Agent Simulation

• • Integrate Python and MATSim to simulate transportation and electric-vehicle electricity supply systems at the district level in Hanoi.

Nguyen Trung Dung

Research Problems

1. Financial Forecasting Challenges

Problem: Traditional forecasting methods based on historical data and statistical models are prone to errors and time-consuming, especially when time series are highly volatile and unstable.

Approach: Integration of deep learning and machine learning models based on artificial intelligence to improve forecasting accuracy. Development of hybrid models combining traditional methods with AI approaches.

Applications:

• Banking: Predicting customer default risk for better lending decisions
• Investment: Portfolio optimization and risk assessment
• Market Analysis: Short-term and long-term market trend predictions

2. Digital Transformation Impact Assessment

Problem: The varying impact of financial digital transformation on different banks' operational efficiency requires comprehensive analysis and customized solutions.

Research Focus:

• Measuring the effectiveness of digital transformation initiatives
• Analyzing competitive pressure effects on different bank sizes
• Developing specialized solutions for banks with different technological orientations
• Human resource optimization in digital finance environments

3. Demand Forecasting Optimization

Problem: Selecting appropriate and effective models (low error, high speed, simplicity) for specific problems and datasets requires continuous research optimization.

Methodology:

• Time series analysis models for direct real-world applications
• Hybrid model development combining multiple forecasting approaches
• Integration of traditional econometric methods with deep learning approaches
• Multi-model ensemble systems for improved accuracy

4. Production Planning and Control

Problem: Complex production scheduling problems requiring multi-stage optimization with sequence-dependent setups and resource constraints.

Solutions:

• Genetic algorithms combined with Lagrange multipliers
• Hybrid simulated annealing and genetic approaches
• Variable neighborhood search and particle swarm optimization
• Stochastic programming for uncertain demand scenarios

Members

Core Faculty Members

• Leadership: TS. Nguyễn Thị Ngọc Anh - Associate Professor

Senior Researchers:

• TS. Nguyễn Thị Xuân Hòa - Associate Professor
• TS. Trần Ngọc Thăng - Ph.D.
• TS. Nguyễn Văn Hạnh - Ph.D.
• TS. Nguyễn Hữu Du - Ph.D.
• TS. Nguyễn Trung Dung - Ph.D.
• TS. Thái Minh Hạnh - Ph.D.
• TS. Nguyễn Thúc Hương Giang - Ph.D.
• TS. Trần Văn Đức - Ph.D.
• TS. Đỗ Bá Lâm - Ph.D.
• TS. Nguyễn Mạnh Cường - Ph.D.

Extended Research Team

Additional Expertise:

• Leading experts in finance and operational optimization
• Extensive research team including PhD students, master's students, and undergraduate students
• Collaborative researchers from three affiliated units of Hanoi University of Science and Technology: 
• • School of Applied Mathematics and Informatics
  • School of Economics and Management
  • School of Information and Communication Technology

Research Infrastructure

Facilities:

• Shared workspace at Institute of Technology and Economics, 6th floor, Ta Quang Buu Library
• Modern infrastructure with simulated financial laboratories
• Advanced software tools including MATLAB, Python, R, Java, C, C++
• Big Data management systems (Apache Hadoop)
• Cloud-based virtual server systems for project-specific computing
• Specialized software supporting operational optimization research

Collaboration Network

Domestic Partnerships:

• Banks and Fintech companies
• Universities in related Fintech fields
• Banking associations and enterprises

International Cooperation:

• International universities
• International research funding organizations
• Global financial institutions and research centers