Research Focus

Investigating machine learning applications in portfolio optimization and risk management across multiple asset classes.

Research Focus Areas

Our two primary research themes and their interconnections

ML for Portfolio Optimization

Developing and evaluating machine learning approaches for optimal asset allocation and portfolio construction. This includes:

  • Deep learning architectures for return prediction
  • Reinforcement learning for dynamic rebalancing
  • Multi-asset portfolio construction with ML
  • Transaction cost optimization
  • Constraint handling in ML-based portfolios
Deep Learning Reinforcement Learning Asset Allocation

Forecasting & Risk Management

Applying ML methods to improve risk measurement, monitoring, and prediction. Key areas include:

  • Volatility forecasting with deep learning
  • Tail risk estimation with ensemble methods
  • Real-time risk monitoring systems
  • Uncertainty quantification in risk models
  • Stress testing with ML scenario generation
Volatility VaR/ES Tail Risk

Machine Learning Methods

The computational techniques driving our research

Deep Learning

Neural networks including transformers, LSTMs, autoencoders, and convolutional architectures for financial time series.

Transformers LSTM Autoencoders

Reinforcement Learning

Policy-based and value-based methods for dynamic portfolio allocation and trading strategy optimization.

PPO A2C DQN

Ensemble Methods

Random forests, gradient boosting, and model stacking for robust predictions and feature importance analysis.

XGBoost LightGBM Stacking

Probabilistic ML

Bayesian methods, Gaussian processes, and uncertainty quantification for calibrated predictions and risk assessment.

Bayesian NN GP Conformal

Statistical Learning

Classical statistical models enhanced with modern regularization and feature selection techniques.

Lasso Ridge Elastic Net

Hybrid Approaches

Combining ML with traditional financial models like GARCH, factor models, and optimization frameworks.

ML+GARCH Neural Factors

Open Research Gaps

Key challenges and opportunities in the field

High Priority

Uncertainty Quantification in Deep Learning for Finance

While deep learning models show strong performance, most lack proper uncertainty estimation. Research is needed on calibrated uncertainty for financial decision-making.

uncertainty calibration bayesian deep learning
High Priority

Multi-Asset Reinforcement Learning Portfolio Management

Most RL portfolio papers focus on single asset classes. Research on cross-asset RL strategies with realistic transaction costs is limited.

multi-asset reinforcement learning cross-asset
High Priority

Interpretable ML for Regulatory Compliance

Financial regulations increasingly require model explainability. Bridging advanced ML techniques with regulatory requirements remains challenging.

explainability regulation SHAP
High Priority

Real-Time ML Risk Monitoring Systems

Gap between academic ML models and production risk systems. Research on online learning, concept drift, and real-time inference for risk management.

real-time streaming concept drift
High Priority

Foundation Models for Financial Time Series

Large language models and foundation models are transforming NLP. Similar approaches for financial time series are nascent and require investigation.

foundation model pre-training self-supervised
Medium Priority

Transfer Learning Across Financial Markets

Pre-trained models and transfer learning are underexplored in finance. Research needed on domain adaptation between markets, time periods, and asset classes.

transfer learning domain adaptation cross-market
Medium Priority

Robust Ensemble Methods for Tail Risk

Ensemble methods excel on average but may fail during extreme events. Research on robust ensembles for tail risk and market stress scenarios.

ensemble tail risk robustness
Medium Priority

Causal ML for Portfolio Decisions

Most ML in finance is correlational. Causal inference methods for understanding intervention effects on portfolios are underexplored.

causal inference counterfactual treatment effect

Specific Research Questions

Concrete questions guiding our investigation

Portfolio Optimization

  • How can deep reinforcement learning be extended to handle multi-asset portfolios with realistic constraints?
  • What is the optimal way to incorporate transaction costs and market impact into ML-based portfolio optimization?
  • Can transformer architectures capture cross-asset dependencies better than traditional methods?
  • How should portfolio ML models be adapted during market regime changes?

Risk Management

  • How can we develop uncertainty-aware deep learning models for VaR and ES estimation?
  • What architectures are most effective for real-time risk monitoring with streaming data?
  • Can ensemble methods be made robust to tail events while maintaining overall performance?
  • How should ML risk models be validated to meet regulatory requirements?

Methodology

  • What pre-training strategies work best for financial time series foundation models?
  • How can causal inference be integrated with ML predictions for portfolio decisions?
  • What transfer learning approaches are most effective across financial markets and time periods?
  • How can we develop interpretable ML models that satisfy both performance and regulatory requirements?

Implementation

  • How should ML models be deployed for production risk management with low latency requirements?
  • What monitoring and retraining strategies are needed for ML models in changing market conditions?
  • How can we build hybrid systems that combine ML predictions with traditional financial models?
  • What data infrastructure is required for effective ML-based portfolio management?

Research Methodology

Our approach to rigorous and reproducible research

1

Literature Review

Systematic analysis of existing research using OpenAlex and other academic databases to identify gaps and build on prior work.

2

Method Development

Novel algorithm design combining ML techniques with financial domain knowledge, with theoretical analysis where applicable.

3

Empirical Validation

Rigorous backtesting on historical data with proper train/validation/test splits and out-of-sample evaluation.

4

Industry Testing

Validation in realistic settings at Quoniam with real market data, transaction costs, and practical constraints.

5

Open Science

Publication of code, data pipelines, and reproducible research artifacts under open source licenses.

6

Peer Review

Submission to top academic journals and conferences for rigorous external validation of research contributions.

Data & Asset Classes

The financial data and markets covered in our research

Equities

Global equity indices, individual stocks, factor exposures, and cross-sectional data.

Fixed Income

Government bonds, corporate credit, yield curves, and interest rate derivatives.

Multi-Asset

Cross-asset strategies, correlation dynamics, and multi-asset portfolio construction.

Derivatives

Options, futures, volatility products, and derivative pricing applications.