Research

BidBridge

Do primary dealers act as short-run balance-sheet bridges when Treasury supply exceeds end-investor demand?

01

The Research Question

When the U.S. Treasury issues large volumes of debt and end-investor demand is weak, primary dealers must absorb the gap onto their own balance sheets. We call this the bridge function—temporary warehousing of supply until it can be distributed to final holders.

This project builds an 846-week auction-week panel (2010–2026) from five public data sources to measure this intermediation channel. We identify supply shocks from pre-auction announcements and trace their impact on dealer inventories using local projections with HAC inference.

The central finding: dealers absorb +$7.3 billion in inventory when announced supply exceeds historical norms—and 2.1x more during quantitative tightening, when the Fed is no longer absorbing supply itself.

Observable Signatures

  1. Heavier issuance or clustered refunding supply
  2. Weaker auction metrics or higher dealer awards
  3. Larger dealer positions and financing usage
  4. Persistence followed by normalization of inventory
02

Key Findings

Supply Shocks Drive Immediate Dealer Accumulation

When announced Treasury supply exceeds its historical 75th percentile, dealers accumulate +$7.3 billion in inventory in the same week (p<0.001). The shock is identified from pre-auction information only—offering amounts known before the auction clears.

The impulse response is front-loaded: most absorption happens at horizon 0, with partial reversal over subsequent weeks as dealers distribute to end-investors.

$M at h=0 (full sample)
$M at h=0 (QT period)
$M at h=0 (non-QT)

Quantitative Tightening Amplifies the Bridge Effect 2.1x

During QT periods (Oct 2017–Sep 2019, Jun 2022–Dec 2025), the supply-shock response is 2.1 times larger than outside QT. When the Fed is shrinking its balance sheet, dealers must absorb supply that the Fed would otherwise have purchased.

This is identified via an interaction term (shock × QT regime) on the full contiguous panel, with the total QT effect computed via the delta method. QT periods are defined from Fed announcement dates, not realized balance-sheet changes.

Larger Auctions Attract More End-Investors

Weekly supply volume and dealer allotment share are strongly negatively correlated (r = −0.81). This inverts the naive hypothesis: bigger auction weeks bring in more institutional investors, so dealers take a smaller percentage—even as they absorb more in absolute terms.

Dealer share has declined secularly from 70% (2013) to 37% (2026), reflecting the growth of investment fund and foreign participation.

Dealers Absorb Most at the Short End

Dealer absorption follows a clear maturity gradient: 62% of Bills, 38% of short coupons (2–3Y), 30% of belly (5–7Y), 27% of long bonds (10–30Y), and just 24% of TIPS.

This is consistent with dealers acting as short-duration warehouses with high turnover. The maturity-bucket panel with bucket and week fixed effects confirms within-week cross-sectional variation in absorption patterns.

Refunding Weeks Show Clear Inventory Spikes

Quarterly refunding weeks (containing 10Y + 30Y package) show +$11.2B inventory change vs −$0.3B in ordinary weeks (p<0.001, Welch's t-test). Refunding weeks also have higher supply but lower bid-to-cover (3.02 vs 3.17).

03

Supply & Dealer Activity Over Time

Explore the full 2013–2026 weekly time series. Bridge episodes (red markers) cluster during periods of heavy supply and stress. Hover for details, drag to zoom.

04

Methodology

LP

Local Projections

Jordà-style local projections estimate the dynamic response of cumulative dealer inventory change to ex ante supply shocks at horizons h = 0…12 weeks.

cum_Δinvt,t+h = αh + βh·shockt + θh·shockt×soft_demandt-1 + Γh·Xt-1 + ut+h
  • Shock: Announced supply > expanding p75 (pre-auction only)
  • Controls: All lagged one week (supply, dealer share, trend, SOMA change)
  • Inference: HAC (Newey-West) via statsmodels, bandwidth = h+1
  • Regimes: Interaction shock × QT (announcement-dated), delta method for total effect
  • Contiguity: Only includes horizons where t+h is exactly h weeks after t
FE

Panel Fixed Effects

Maturity-bucket panel with bucket and week fixed effects, exploiting within-week cross-sectional variation across the yield curve.

Δpositionb,t = αb + τt + β·supplyb,t + θ·supply × soft_demandb,t-1 + εb,t
  • Panel: Balanced 6-bucket × 675-week grid using directly observed NY Fed maturity bands (n = 2,584 in estimation)
  • Buckets: Bills, short coupon (2–3Y), belly (5–7Y), long (10–30Y), TIPS, FRN
  • Inference: Driscoll-Kraay headline surface + clustered-by-bucket robustness
OLS

Extended OLS Regressions

Descriptive OLS with HC1 robust standard errors. Extended specification includes SOMA changes, bank holdings, and time trend as controls.

  • R² = 0.168 (extended specification, n = 673)
  • Refunding effect: +$3.9B inventory (p = 0.099)
  • SOMA effect: Fed buying $1B reduces dealer inventory by $92M
  • Subsample splits: Pre/post-2020, QT/non-QT, refunding/ordinary
Z

Bridge Episode Detection

Bridge episodes flag weeks of unusually large dealer absorption using only backward-looking information.

  • Heavy supply: Awarded amount > 52-week rolling median
  • Positive accumulation: Inventory change > 0
  • Unusual size: Z-score > 1 on trailing 13-week window
  • Result: 85 episodes across 675 dealer-observed weeks (12.6%)
05

Regression Results

Extended bridge regression with HC1 heteroskedasticity-robust standard errors (n = 673, R² = 0.168). Dependent variable: weekly dealer inventory change ($M).

VariableCoefficientStd. Errort-statp-value

*** p<0.001, ** p<0.01, * p<0.05. HC1 robust standard errors. All monetary values in millions USD.

06

Data Sources

All data is publicly available from U.S. federal agencies. The pipeline fetches, harmonizes, and merges five sources into an 846-week auction-week panel with 33 columns.

Fetch
5 public APIs
Harmonize
CUSIP merge, lag alignment
Panel
846 weeks × 33 cols
Metrics
Bridge episodes, z-scores
Site Data
JSON for charts + tables
Analysis
LP, FE, persistence, pressure monitor
07

Annual Summary

Treasury issuance has grown from $5.9T (2013) to over $30T annually (2025), while dealer allotment share has fallen from 70% to 37%. Bridge episodes are becoming more frequent and larger.

YearAuctionsAwarded ($B)Dealer ShareAvg Inventory ($M)Bridge Episodes
08

Stress Regimes & Additional Analysis

Bridge Episode Rate by Regime

Risk-off periods (high auction tails) show the highest bridge episode frequency at 17.0% vs 11.2% baseline. QT periods identified from Fed announcement dates (Oct 2017, Jun 2022, ending Dec 2025).

Persistence of Inventory Accumulation

After bridge episodes, excess detrended inventory decays with a half-life that is now estimated on abnormal inventory (removing the secular $98B→$500B trend). The detrended R² is substantially higher than the raw estimate.

09

Upcoming Auction Pressure

A compact near-term monitor for the next 1–4 auction weeks. Scores combine recent bridge intensity, weak-end-investor absorption, and the size mix of scheduled supply into a transparent 0–1 pressure scale.

How the score and category work

composite_pressure_score = 0.40 × supply_size_score + 0.20 × bill_share_score + 0.20 × recent_bridge_rate + 0.20 × recent_weak_demand_rate

low is below 0.33, medium is 0.33 to below 0.66, and high is 0.66 or above.

Week Ahead Offering ($B) Supply Bill Share Recent Bridge Recent Weak Demand Score Category

Scores are normalized to 0–1. The monitor is written by bidbridge run-all and exposed through the site-data JSON payload.