P1 · A Weekly Economic Tracker from FRED
Practicum brief
The instructor build — a weekly activity tracker built from four public FRED series, scaled to GDP units, benchmarked against the official Weekly Economic Index. Live data, executed code, every figure reproducible. That page is the standard this practicum is graded against; the build journal records how it actually went, dead ends included.
Motivation
Lewis et al. (2022) showed that ten weekly series, one factor, and disciplined scaling produce an index that tracks U.S. GDP growth closely enough to be useful in real time — the Weekly Economic Index (WEI), now published by the Dallas Fed. It is the cleanest possible demonstration that the factor-model machinery of Session 2 pays rent. Your task is to build a small sibling of it, end to end, from public data only.
The deliverable
A reproducible report (Quarto notebook or equivalent) that:
- pulls at least four weekly U.S. series from FRED programmatically (no hand-downloaded files) — the instructor build uses initial claims, continued claims, bank credit, and consumer loans; you are encouraged to find better ones (EIA energy volumes, rail carloads, anything public and weekly);
- transforms each to a stationary, seasonally comparable form (52-week log differences, following Lewis et al.) and standardizes;
- extracts a common factor (principal component at minimum; the Kalman filter’s one-factor model from Session 1 for distinction);
- scales the factor to GDP units so the index reads as “GDP growth over the past year of x%”;
- benchmarks against the official WEI: correlation, divergence episodes, and an honest paragraph on why yours differs (fewer series, different sectors, no revisions handling);
- plots the COVID collapse and recovery — the episode that made weekly trackers famous.
Definition of done
Stretch goals
- Replace the principal component with the one-factor Kalman filter model and compare the two factors — where do they disagree, and why?
- Pull claims vintages from ALFRED and compute the tracker as it would have read on a specific 2020 Thursday versus how it reads today.