Discussion 2 – Benchmarking & Forecasting

Discussion 2 – Benchmarking & Forecasting#

This session pivots from dashboards to the theory and tooling behind the Financial Time-Series Forecasting Repository (FTSFR).

Slide Talk: Why a Financial Forecasting Benchmark?#

  • We’ll walk through the slides in slides_ftsfr.pdf.

  • Topics include: the motivation for a dedicated finance benchmark, dataset design, train/test protocol, and how the team evaluates competing models.

  • Bring questions about asset classes, evaluation metrics, or how the benchmark might align with your own research interests.

Repository Walkthrough#

  • Tour the FTSFR repo structure (src/, _data/, _output/, datasets.toml, subscriptions.toml).

  • Review environment setup, .env credentials, and the key doit scripts you’ll use (dodo_01_pull.py, dodo_02_forecasting.py).

  • Highlight where forecasting jobs drop their predictions and error metrics so you can wire them into the Streamlit apps.

Hands-On Forecasting#

  • Kick off at least one forecasting job (or use the provided sample outputs if compute is limited).

  • Inspect the generated CSV/parquet files and verify the schema matches what your dashboard expects.

  • Capture any TODOs needed to plug the forecasts into app_01.py or app_04_crsp.py before Discussion 3.

Suggested Next Steps#

  • Run a full forecasting batch overnight and stash the outputs for integration.

  • Start weaving forecasts into your dashboard—new tabs, comparison charts, or narrative callouts.

  • Note any blockers (data access, compute, modeling questions) so we can troubleshoot them in the next session.