Machine Learning in Finance

This two-day course gives finance professionals a practical introduction to modern machine learning, from framing financial problems and building robust predictive models to deep learning, alternative data and deployment. Participants use OpenAI Codex throughout as an AI development assistant to help build, debug and improve models. The emphasis is on applying ML to real financial use cases and leaving with the skills to develop, validate, deploy and monitor ML workflows in their own workspace.

• Portfolio managers, traders and investment analysts who want to use ML in investment workflows.
• Quants, developers and data professionals working with financial data.
• Analysts and quants
• Risk, model-validation and finance professionals who need to understand modern ML methods.
• Tech-savvy finance professionals who want to build practical ML solutions rather than only study theory.

• Frame real financial problems as machine-learning tasks.
• Build and compare traditional ML and deep-learning models.
• Use structured and unstructured data, including financial text.
• Validate models correctly for financial time series and understand model behaviour.
• Convert predictions into usable financial decisions or signals.
• Use Codex to accelerate ML development and debugging.
• Build, deploy and monitor an end-to-end ML workflow in their workspace.

• Working knowledge of financial markets and common financial data.
• Basic Python familiarity is helpful, but participants do not need to be expert programmers.
• No prior machine-learning or deep-learning experience is required.
• All code, datasets and templates used in the practical sessions are provided.

Solving Financial Problems with Modern ML

Session 1: From Financial Problem to ML Solution — Theory

Focus: How to frame financial questions correctly and choose an appropriate modelling approach.

• Use cases: return forecasting, classification, ranking, volatility/risk prediction and regime detection.
• Financial ML challenges: low signal-to-noise, non-stationarity, leakage and changing regimes.
• Feature engineering for market and macro data.
• Model choices: linear models, Random Forest, XGBoost/LightGBM/CatBoost.
• Time-aware validation and suitable statistical and financial metrics.

Session 2: Build a Financial Prediction Model with Codex — Practical

Focus: Build a complete supervised ML workflow using financial data.

• Prepare leak-free features and prediction targets.
• Build a baseline model and a gradient-boosted model.
• Run time-aware cross-validation and compare model performance.
• Use Codex to assist with coding, debugging, documentation and iteration.

Practical output: A reusable financial ML training pipeline.

Session 3: Deep Learning for Financial Time Series — Theory

Focus: Understand when deep learning can add value beyond traditional ML.

• Neural-network fundamentals and overfitting in financial data.
• Sequence models: LSTM, TCN and Transformers.
• Attention, multi-horizon and probabilistic forecasting.
• When deep learning is useful—and when simpler models are preferable.

Session 4: Build and Compare a Deep-Learning Model — Practical

Focus: Train a sequence model and test whether additional complexity improves out-of-sample results.

• Prepare sequential training data.
• Train an LSTM, TCN or Transformer model.
• Compare performance with the Day 1 gradient-boosted model.
• Use Codex to modify architecture, parameters and diagnostics.

Practical output: A working deep-learning model with a disciplined benchmark comparison.

From Alternative Data to Deployable ML Systems

Session 5: Alternative Data, NLP and Robust Models — Theory

Focus: Learn how to extract signal from unstructured data and assess whether models are genuinely robust.

• Financial text: news, filings, earnings calls and central-bank communication.
• Embeddings and modern text representations.
• Combining text and numerical market features.
• Walk-forward validation, regime analysis, drift and purging/embargo.
• Explainability with feature importance, SHAP and model diagnostics.

Session 6: Build a Multi-Source Financial Model — Practical

Focus: Combine market data and text into one predictive workflow.

• Generate text embeddings and combine them with numerical features.
• Compare market-only, text-only and combined models.
• Use SHAP to understand important drivers.
• Test robustness across time and market regimes using Codex-assisted analysis.

Practical output: A multi-source financial model with explainability and robustness checks.

Session 7: From Prediction to Decision and Production — Theory

Focus: Understand how a research model becomes a usable system in a financial workspace.

• Turn scores and probabilities into decisions, rankings, signals or positions.
• Thresholds, sizing, turnover, transaction costs and risk-adjusted evaluation.
• The ML lifecycle: data → training → validation → deployment → monitoring → retraining.
• Experiment tracking, model versioning, batch/real-time inference and drift monitoring.
• Using Codex to refactor research code, create tests, APIs, dashboards and deployment components.

Session 8: Capstone — Build and Deploy an End-to-End ML Workflow — Practical

Focus: Bring the entire course together into a workflow participants can adapt to their own workplace.

• Define a financial problem and build the data and feature pipeline.
• Train, validate and select a model.
• Convert predictions into a financial decision or signal.
• Package the model for repeatable inference.
• Deploy via a simple API, dashboard or scheduled process.
• Define monitoring and retraining rules using Codex as the development assistant.

Practical output: A reusable template for developing, deploying and monitoring ML models in the participant’s own workspace.

End-of-course capability. Participants should leave able to take a real financial problem and move through the full workflow: frame it → build the data → develop the model → validate it → deploy it → monitor it.