Forecasting with Large Language Models

A Complete Guide to Time Series Prediction from Classical Methods to LLMs

Chapter 1: What Is Time Series Forecasting?

1. The Value of Prediction
2. What Makes Time Series Different
3. Core Terminology
4. The Forecasting Landscape
5. How to Use This Book

Chapter 2: Understanding Time Series Data

1. Components of a Time Series
2. Visualizing Time Series
3. Stationarity and Why It Matters
4. Autocorrelation and Partial Autocorrelation
5. Seasonality Detection and Characterization

Chapter 3: Preparing Time Series Data

1. Data Quality and Cleaning
2. Handling Missing Values
3. Temporal Train/Validation/Test Splitting
4. Feature Engineering for Time Series
5. Scaling and Transformations

Chapter 4: Evaluating Forecasts

1. Forecast Error Metrics
2. Probabilistic Forecast Metrics
3. Validation Strategies
4. The Importance of Baselines
5. Avoiding Evaluation Pitfalls

Chapter 5: Classical Statistical Methods

1. Naive and Simple Models
2. Exponential Smoothing Family
3. ARIMA and SARIMA
4. Automatic Model Selection
5. When Classical Methods Win

Chapter 6: Machine Learning for Time Series

1. Framing Time Series as Supervised Learning
2. Tree-Based Models for Forecasting
3. Feature Engineering at Scale
4. Handling Multiple Series
5. Practical Strengths and Limitations

Chapter 7: Deep Learning Foundations for Time Series

1. Why Neural Networks for Sequences
2. Recurrent Neural Networks
3. LSTM and GRU Architectures
4. Practical Deep Learning for Time Series

Chapter 8: Transformers and Modern Deep Learning Architectures

1. Attention Mechanisms for Time Series
2. Positional Encodings and Sequence Order
3. Temporal Fusion Transformer
4. Patch-Based and Frequency-Domain Models
5. Benchmark Performance on M4/M5/GluonTS

Chapter 9: Large Language Models Enter Forecasting

1. How Can Text Models Forecast Numbers?
2. Prompting Strategies for Time Series
3. Zero-Shot and Few-Shot Forecasting
4. Numerical Reasoning Limitations
5. Prompt-Based vs Model-Based Approaches
6. When LLMs Are the Wrong Tool

Chapter 10: Pretrained Time-Series Foundation Models

1. The Foundation Model Paradigm for Time Series
2. Chronos: Probabilistic Forecasting with Autoregressive Models
3. TimesFM and Google’s Approach
4. Time-LLM and Temporal Adaptation Layers
5. Tokenization and Representation of Numerical Sequences

Chapter 11: Adapting LLMs for Forecasting

1. Fine-Tuning Strategies
2. Parameter-Efficient Adaptation
3. Retrieval-Augmented Forecasting
4. Hybrid Architectures
5. Multimodal Context Integration

Chapter 12: Advanced Forecasting Scenarios

1. Multivariate Time Series
2. Long-Horizon Forecasting
3. Covariates and Exogenous Variables
4. Hierarchical and Grouped Time Series
5. Irregularly Sampled and Missing Data

Chapter 13: Probabilistic Forecasting and Uncertainty

1. Why Uncertainty Matters
2. Classical Prediction Intervals
3. Quantile Regression and Pinball Loss
4. Deep Probabilistic Methods
5. Conformal Prediction for Time Series

Chapter 14: Production Systems and MLOps for Forecasting

1. From Prototype to Production
2. Data Pipelines for Forecasting
3. Model Serving Architectures
4. Monitoring and Alerting
5. Model Retraining and Lifecycle Management
6. Cost and Efficiency Considerations

Chapter 15: Future Directions and Advanced Topics

1. Where the Field Is Heading
2. Emerging Architectures and Techniques
3. Benchmarking and Evaluation Challenges
4. LLMs and the Human Forecaster
5. Ethical and Societal Considerations
6. Preparing for What Comes Next

Conclusion: The Forecasting Journey Ahead

References