Dynamic Modeling and Forecasting in Big Data

Most machine learning models emphasize cross-sectional data, while traditional time-series models typically focus on a small number of variables and low-frequency data.

This course develops the skills and tools needed to analyze big data that are rich in both variables and time.

We cover both structural and reduced-form approaches.

Topics include dynamic regression models, dynamic factor models, vector autoregressions (VAR), error-correction models, dimensional-reduction techniques for high-dimensional datasets, state-space models, and methods for decomposing trends, cycles, and seasonality in high-frequency data.

The course incorporates recent advances in artificial intelligence (AI) for dynamic modeling and forecasting, including transformer architectures, probabilistic deep learning, diffusion models, and large language models (LLMs).

Students learn how AI methods complement structural econometric approaches in high-dimensional and real-time forecasting environments.

The course is taught primarily in RStudio, with supplemental use of Python.

The course emphasizes data-driven and case-based applications, with relatively less focus on mathematical derivations and formal theory.