A public-data research project studying how Supplementary Leverage Ratio pressure shapes large-bank Treasury holdings and balance-sheet behavior. The pipeline is built entirely from free regulatory filings and market data, covering FFIEC Call Reports, FR Y-9C, FR Y-15, NY Fed dealer statistics, and FINRA TRACE.
The Supplementary Leverage Ratio requires large banks to hold Tier 1 capital against all on-balance-sheet assets without risk-weighting. Because SLR treats Treasuries the same as corporate loans in the denominator, leverage-constrained banks face a real cost to holding Treasuries. That link matters for Treasury-market intermediation, monetary-policy transmission, and capital-rule design.
When the Fed temporarily excluded Treasuries and reserves from SLR in April 2020, it created a natural experiment for measuring how leverage-ratio relief affects bank Treasury demand. SLR Watch uses that event as its causal core and extends the question with a broader constraint-regime analysis.
The repo now carries an explicit 2020 insured-bank treatment map. That makes the direct-treatment design methodologically cleaner, but it also broadens the balanced coverage core enough that the earlier positive Treasury signal largely disappears. The current read is cautionary rather than confirmatory.
Universe A: all insured-bank descriptive universe (2019Q1-2021Q4)
- 5,488 insured-bank filers
- 61,626 bank-quarter observations
Universe B: SLR-reporting insured banks
- 37 entities
- 426 bank-quarter observations
Universe C: treatment-definable SLR sample
- 20 entities
- 233 bank-quarter observations
- Requires a usable 2019Q4 baseline for treatment assignment
Universe D: primary causal core
- 35 balanced-coverage entities
- 420 balanced bank-quarter observations
- The treatment-defined Treasury estimable cores inside Universe D still use 19 entities / 228 observations
- Low-headroom Treasury result: -0.15pp (p = 0.813)
- Covered-bank / direct-eligibility Treasury result: +0.40pp (p = 0.570)
Universe F: flagship per-parent clustered inference (16 parent clusters)
- 16 entities
- 192 bank-quarter observations
- Low headroom: -0.59pp (p = 0.690)
- Covered bank / direct eligibility: -0.23pp (p = 0.911)
Under parent-level clustering, neither Treasury split shows a reliable positive effect. The broader direct-eligibility treatment is still the cleaner regulatory split, but the current evidence is inconclusive rather than supportive.
The diagnostics are cleaner than the coefficient signal. In Universe D, the Treasury pre-trend joint p-values are 0.826 for the low-headroom split and 0.933 for the covered-bank split; in the clustered flagship sample they are 0.163 and 0.266. The three-placebo grid is also quiet for both Treasury rows, with covered-bank placebo p-values of 0.492, 0.364, and 0.293. But the leave-one-parent-out ranges still cross zero for both treatments, so the right public framing is that the map-backed direct-treatment design is cleaner while the Treasury effect itself is currently weak.
Five extension reports still help characterize how balance-sheet pressure reallocates activity, even though the broadened direct-treatment Treasury signal is weak:
- Reallocation, not expansion. Constrained banks shifted balance-sheet capacity toward Treasuries while reducing deposits and loans. Low-headroom banks: Treasury +2.47pp, deposit growth -3.52pp, loan growth -1.65pp.
- Safe-asset composition shift. The strongest safe-asset reallocation is in the low-headroom split: Treasury share of safe assets rises by +9.11pp relative to controls, while the covered-bank / direct-eligibility split is roughly flat to slightly negative.
- Family-level transmission. Bank and parent Treasury-share changes moved in the same direction in 73.3% of linked quarter-over-quarter comparisons across 16 parent families.
- Trading-balance-sheet tradeoff. Low-headroom banks reduced trading-asset share by 1.71pp while Treasury holdings rose by 2.47pp, consistent with leverage-capacity reallocation.
- Policy-regime context. A longer quarterly panel (2019-2026) places the 2020 event within a broader safe-asset absorption story across pre-exclusion, exclusion, post-exclusion, and QT-era regimes.
A live decomposition module extends the project from "did SLR relief matter in 2020?" to "which balance-sheet constraint matters in which regime?" It compares leverage headroom, duration-loss pressure, and funding stress across insured-bank and parent/IHC panels through 2025Q4, using a scorecard built from SLR headroom, Treasury unrealized-loss proxies, deposit runoff, repo reliance, deposit funding gaps, liquid buffers, and HTM mix.
Key results:
- In the 2022-2023 duration-loss window, duration loss is the dominant bucket for insured banks in 65.9% of observations and for parents/IHCs in 63.0%.
- By late QT normalization, insured banks still lean duration loss at 42.1%, while parents/IHCs now lean back toward leverage at 35.4%.
- Linked parent-bank families match on the dominant constraint in 64.8% of family-quarters during the duration-loss window; both bank and parent are duration-loss dominant in 48.6% of those observations.
A first interaction-regression layer provides supporting evidence on the parent panel: higher duration pressure is associated with higher Treasury share in the 2022-2023 window (coefficient 0.027, p = 0.001) and lower Treasury share during late QT normalization (-0.034, p = 0.026). Bank-side interaction results are weaker. This layer supports the descriptive decomposition rather than constituting a separate causal claim.
| Source | Use | Cadence |
|---|---|---|
| FFIEC Call Reports | Main insured-bank panel | Quarterly |
| FR Y-9C (Chicago Fed / NIC) | Parent holding companies | Quarterly |
| FR Y-15 Snapshots | GSIB surcharge context | Annual |
| OFR Bank Systemic Risk Monitor | Surcharge enrichment | Mixed |
| NY Fed Primary Dealer Statistics | Market context overlay | Weekly |
| FINRA TRACE Treasury | Market context overlay | Weekly / Monthly |
config/ Rule regime definitions, variable registry, source manifest
src/slr_watch/ Python package: rules, headroom, ingestion, panels, analytics
data/ Raw, staged, and derived data (generated by pipeline)
output/reports/ Event-study results and extension reports
tests/ Unit and integration tests
site/ Static GitHub Pages site
# Install (core analysis — reproducible, pinned dependencies)
pip install -r requirements.lock && pip install -e ".[dev]" --no-deps
# Or install with latest compatible versions
pip install -e ".[dev]"
# Optional: browser automation for FR Y-9C downloads
pip install -e ".[browser]"
python -m playwright install chromium
# Run tests
python -m pytest -q
# Download and stage data (example quarters)
python -m slr_watch.cli download-call-reports --quarter 2020Q1
python -m slr_watch.cli stage-call-reports --quarter 2020Q1
python -m slr_watch.cli download-fry9c --quarter 2025Q4
python -m slr_watch.cli stage-fry9c --quarter 2025Q4 --input data/raw/fry9c/2025Q4/BHCF20251231.ZIP
# Build panels
python -m slr_watch.cli build-crosswalk
python -m slr_watch.cli build-fdic-institutions
python -m slr_watch.cli build-all-insured-panel
python -m slr_watch.cli build-insured-panel
python -m slr_watch.cli build-parent-panel
# Run analysis (or `make reproduce` to run all steps below at once)
python -m slr_watch.cli run-event-study
python -m slr_watch.cli run-reallocation-report
python -m slr_watch.cli run-safe-asset-absorption-report
python -m slr_watch.cli run-parent-transmission-report
python -m slr_watch.cli run-treasury-intermediation-report
python -m slr_watch.cli run-policy-regime-panel-report
python -m slr_watch.cli run-constraint-decomposition-report
python -m slr_watch.cli build-site-dataThe event-study pipeline now writes output/reports/event_2020/sample_manifest.csv, sample_ladder.csv, treatment_roster.csv, methodology_memo.md, and gpt_pro_next_steps_prompt.md so the exact descriptive universe, causal sample ladder, treatment-map classification, exclusion reasons, and next-question handoff are explicit.
If you need to inspect which banks sit in the current 19-bank Treasury core or the 16-bank clustered flagship sample, use output/reports/event_2020/treatment_roster.csv; the site keeps that roster in a collapsed audit note instead of putting another visible table in the main narrative.
See python -m slr_watch.cli --help for the full command set.
For current FR Y-9C quarters, the NIC downloader first tries browser automation. If FFIEC blocks that flow, it automatically checks for the expected ZIP in the raw quarter folder, the current working directory, and ~/Downloads before failing, so the practical fallback is to download the ZIP manually and rerun the same command.
A static research microsite with interactive charts and detailed findings is available at smkwray.github.io/slrwatch (deployed from site/).
cd site && python -m http.server 8000See LICENSE.