BRDA Living Data · Handbook Diffs
Handbook Temporal Diffs
Supervisory handbooks don’t change all at once. When the Fed updates its Commercial Real Estate guidance, which sections changed and which stayed the same? This panel computes section-level cosine similarity between consecutive versions of regulatory handbooks, flagging high-drift sections where the supervisory stance shifted.
Corpus Coverage
| Corpus | Version Pairs | Diff Pairs | High Drift (<0.5) |
|---|---|---|---|
| CFPB Supervisory | 32 | 100 | 100% |
| DFAST Scenarios | 12 | 296 | 53.7% |
| Fed FSR | 14 | 771 | 84.8% |
| OCC Risk Perspectives | 27 | 570 | 86.3% |
How It Works
- Align sections — Group chunks by (section heading, version date) to identify corresponding sections across editions
- Vectorize — TF-IDF on section text with regulatory-domain stop words
- Score — Cosine similarity + Jaccard similarity between aligned section pairs
- Flag drift — Sections with cos_sim < 0.5 are marked as high-drift for further analysis
White Paper: The Handbook Signal
Do handbook revisions predict supervisory actions? When agencies significantly revise examination guidance on a topic, enforcement actions in that domain tend to follow. This panel quantifies the magnitude of handbook changes and provides the timing identification needed for predictive event studies. High-drift sections are natural treatment events for difference-in-differences designs.
Applications: Regulatory change detection, examination priority forecasting, compliance burden estimation, supervisory stance shift identification