AI Financial Forecasting Course

Lesson 1 • Data Quality and Preprocessing

Covers missing values, outliers, and normalisation techniques. Prepares learners to deliver clean inputs to any forecasting pipeline.

Lesson 2 • Statistical Foundations for Forecasting

Reviews probability distributions, stationarity, and autocorrelation. Provides the statistical grounding required for model interpretation.

Lesson 3 • Forecasting Workflow and Evaluation

Introduces an end-to-end forecasting workflow from problem definition to deployment. Anchors all subsequent chapters within a repeatable process.

Lesson 4 • Core Concepts in Financial Forecasting

Defines forecasting objectives, horizons, and error metrics. Establishes shared vocabulary used throughout the course.

Lesson 1 • Volatility Modelling with GARCH

Introduces ARCH and GARCH models for conditional variance forecasting. Addresses the heteroskedasticity common in financial return series.

Lesson 2 • ARIMA Model Family

Covers AR, MA, ARMA, and ARIMA specification and estimation. Connects differencing and lag selection to stationarity concepts from Chapter 1.

Lesson 3 • Model Diagnostics and Selection

Applies residual analysis and information criteria to validate classical models. Ensures learners can detect misspecification before advancing to ML methods.

Lesson 4 • Multivariate Time Series Models

Extends univariate methods to VAR and cointegration frameworks. Enables forecasting of interdependent financial variables simultaneously.

Lesson 1 • Hyperparameter Tuning and Cross-Validation

Applies time-series-aware cross-validation and grid search to ML models. Prevents data leakage, a critical concern in financial applications.

Lesson 2 • Feature Engineering for Finance

Transforms raw financial data into predictive features. Directly enables the model training covered in subsequent sections.

Lesson 3 • Regression and Tree-Based Models

Covers linear regression, decision trees, and ensemble methods for return forecasting. Builds intuition for bias-variance tradeoffs in financial contexts.

Lesson 4 • Model Interpretability in Finance

Uses SHAP values and permutation importance to explain ML predictions. Addresses regulatory and stakeholder transparency requirements.

Lesson 1 • Training and Regularisation Best Practices

Covers learning rate scheduling, early stopping, and data augmentation for financial neural networks. Ensures robust generalisation on limited financial datasets.

Lesson 2 • Convolutional Networks for Time Series

Applies 1D CNNs and dilated convolutions to extract local temporal patterns. Complements recurrent models with faster training and parallelism.

Lesson 3 • Neural Network Fundamentals

Reviews feedforward networks, activation functions, and backpropagation. Establishes the building blocks needed for recurrent and attention-based architectures.

Lesson 4 • Transformer and Attention Mechanisms

Introduces self-attention, positional encoding, and Transformer variants for finance. Enables long-range dependency modelling beyond LSTM capacity.

Lesson 1 • Classical NLP Techniques

Applies tokenisation, TF-IDF, and topic modelling to financial documents. Provides baseline text features before introducing transformer-based methods.

Lesson 2 • Large Language Models as Forecasting Tools

Uses LLM-generated summaries and embeddings as forecasting inputs. Addresses prompt engineering and hallucination risks in financial contexts.

Lesson 3 • Financial Text Data Sources

Surveys earnings calls, news feeds, filings, and social media as signal sources. Motivates text-based feature engineering covered in later sections.

Lesson 4 • Transformer-Based Language Models

Fine-tunes BERT-family models on financial text for sentiment and event classification. Builds on Transformer architecture introduced in Chapter 4.

Lesson 1 • Conformal Prediction and Quantile Regression

Covers distribution-free prediction intervals and quantile loss functions. Provides model-agnostic uncertainty bounds compatible with any ML model.

Lesson 2 • Monte Carlo and Simulation Methods

Uses Monte Carlo simulation and bootstrapping to generate forecast distributions. Connects to risk management applications in Chapter 7.

Lesson 3 • Limitations of Point Forecasts

Demonstrates how point forecasts obscure risk and lead to poor decisions. Motivates the shift to distributional and interval forecasting methods.

Lesson 4 • Forecast Calibration and Scoring

Evaluates probabilistic forecasts using proper scoring rules and reliability diagrams. Ensures learners can audit and improve calibration in production systems.

Lesson 1 • ML-Based Credit and Default Forecasting

Trains classification models to predict default probability and credit deterioration. Extends supervised ML from Chapter 3 to credit risk applications.

Lesson 2 • Risk Metrics and Their Forecasting Needs

Defines VaR, CVaR, and drawdown metrics and their data requirements. Frames risk measurement as a forecasting problem requiring probabilistic outputs.

Lesson 3 • Stress Testing and Scenario Design

Designs historical and hypothetical stress scenarios using AI-generated paths. Addresses supervisory expectations for forward-looking stress testing.

Lesson 4 • Market Risk Forecasting with AI

Applies deep learning and GARCH-hybrid models to forecast portfolio volatility and VaR. Integrates volatility modelling from Chapter 2 with neural architectures.

Lesson 1 • Model Versioning and Experiment Tracking

Uses experiment tracking tools to log parameters, metrics, and artifacts. Enables reproducibility and auditability required in regulated financial environments.

Lesson 2 • Compliance and Auditability in AI Systems

Embeds explainability, access controls, and audit logging into deployed models. Addresses governance requirements specific to financial AI applications.

Lesson 3 • Monitoring and Data Drift Detection

Detects feature drift, concept drift, and performance degradation in live models. Prevents silent model failures that can cause significant financial losses.

Lesson 4 • Continuous Integration and Delivery for ML

Applies CI/CD principles to model training, testing, and deployment pipelines. Enables rapid, safe iteration on production forecasting systems.