Practica
Five hands-on projects plus the capstone. Each practicum ships with the assignment scaffold and a fully worked reference build, so you can see what a finished project looks like before you start your own.
| # | Practicum | Deliverable | Session | Status |
|---|---|---|---|---|
| P0 | Environment + the sealed envelope | Working Python env; a naive current-quarter GDP nowcast, committed and sealed until demo day | 1 | in development |
| P1 | Weekly economic tracker | A weekly activity index from FRED data, à la Lewis et al. (2022) — reference build · journal | 2 | live |
| P2 | Seasonal adjustment at high frequency | Hourly electricity load, seasonally adjusted three ways, published as an interactive decomposition | 3 | in development |
| P3 | A daily sentiment index | News-text sentiment via lexicons and LLM annotation, validated against a market outcome | 4 | in development |
| P4 | The foundation-model horse race | Gradient boosting vs. zero-shot Chronos/TimesFM on macro data, with vintage-honest backtesting | 5 | in development |
| ★ | Capstone: the nowcasting dashboard | A live GDP nowcast with uncertainty bands, benchmarked against GDPNow | 7–8 | in development |
Design principles (fixed across all practica):
- Real-time discipline everywhere. Backtests run on archived vintages (ALFRED), never on today’s revised data.
- Everything reproducible. Each practicum is a self-contained repo scaffold with pinned dependencies;
make all(or one notebook, run top to bottom) regenerates every figure. - Public artifacts. Every practicum ends in something you can put on a website: an interactive chart, a small dashboard, a readable report.