When the U.S. Treasury rebuilds its cash balance, where does the money come from? We identify -- rapid rebuild episodes from 2005–2026 and use auction-schedule surprises to trace the causal funding channels. The answer: money market funds, drawing on the Fed's ON RRP facility.
The Treasury General Account (TGA) is the U.S. government's checking account at the Federal Reserve. When Treasury issues new securities—particularly short-term bills—the proceeds flow into the TGA, draining liquidity from the financial system. These "rebuild" episodes can move hundreds of billions of dollars in weeks, yet the transmission channels are poorly understood.
This project asks: who actually provides the funding? Drag to zoom, hover for values, double-click to reset.
Weekly TGA balance (DTS Wednesday close, $B) with -- rebuild events shaded in orange. Use the range slider to zoom into specific periods.
Using a bill-size surprise instrument (see Methods), we find that a one-standard-deviation ($--B) unexpected increase in weekly bill issuance causes MMF Treasury holdings to rise by $--B over 4 weeks (t=--, Newey-West).
Central result: Unexpected bill issuance that rebuilds the TGA is primarily funded by money market funds, with ~75% of the MMF absorption sourced from ON RRP runoff. Bank deposits, reserves, and bank Treasury holdings show no significant causal response.
The interactive charts below compare impulse response functions from two shock specifications. Blue = binary shock (contaminated by pre-trends). Red = bill-size surprise (clean identification). The vertical dotted line marks the shock date. Hover for exact values.
The bill-surprise specification eliminates spurious Bank T&A and Deposits responses while strengthening MMF Treasury and ON RRP channels.
* Significant at 5% with Newey-West HAC standard errors. Binary shock is contaminated by pre-trends; bill-surprise is the clean specification.
A clean shock should show no significant movement before it occurs. The binary shock shows significant placebo leads in -- of 6 channels (--/24 coefficients), versus -- of 6 (--/24) for the bill-surprise specification.
Descriptive counts of the largest positive proxy contribution by era:
Bars summarize the top positive proxy within each event window by shared era bins. Treat these counts as descriptive accounting summaries, not structural identification.
ON RRP declines by $--B per 1-std shock (t=--). The mechanism: MMFs redeem ON RRP balances to buy bills, channeling money from the Fed's balance sheet to the TGA without touching bank deposits.
TGA (blue, left axis) and ON RRP (green, right axis) during the facility era. Orange bands = rebuild events.
When ON RRP holds ≥$100B (green), rebuild shocks transmit differently than when the facility is scarce (red). These are binary-shock IRFs split by regime — hover for exact values.
Green = ON RRP abundant (≥$100B). Red = ON RRP scarce (<$100B). Pre-facility weeks excluded from estimation.
-- rapid TGA rebuild episodes detected using DTS Wednesday-close levels and positive-only 90th percentile thresholds. Hover over events for details.
Bubble size = event duration in weeks. Blue = bill-heavy (≥70% bills). Orange = mixed.
Stacked bars show positive proxy contributions by channel. White dots = actual ΔTGA. Bars exceeding the dot indicate over-explanation from channel overlap.
Weekly panel: -- observations,
-- variables,
----–----.
All values in millions of USD.
TGA from DTS Wednesday closing balances (not FRED WTREGEN weekly average).
H.8 series converted from billions. OFR monthly data forward-filled via merge_asof.
Flagged when weekly ΔTGA or rolling 4w/8w change exceeds the 90th percentile of positive changes only. Contiguous flagged weeks (max 9-day gap) grouped into events.
| Channel | Source | Sign | Interpretation |
|---|---|---|---|
| Reserve drain | FRED WRBWFRBL (H.4.1) | −1 | Reserves ↓ → money → TGA |
| Deposit drawdown | FRED DPSACBW027NBOG (H.8) | −1 | Deposits ↓ → money left banks |
| ON RRP runoff | FRED RRPONTSYD | −1 | ON RRP ↓ → money left facility |
| Bank T&A | FRED TASACBW027NBOG (H.8) | +1 | Banks bought Treasuries |
| MMF Treasury | OFR MMF-MMF_T_TOT-M | +1 | MMFs bought Treasuries |
| Dealer repo | OFR NYPD-PD_RP_T_TOT-A | +1 | Dealers financed inventory |
14-day staleness guard. Foreign holdings (quarterly) excluded from weekly attribution.
surprise = offering_amt − trailing_median(last 8 same-term × reopening-status auctions, min 4). Summed to weekly. Tax-receipt surprise (DTS deposits) as control. A same-term-only grouping is tracked separately as a robustness comparison. Strips ~94% of predictable auction supply.
Δyt+h = α + β(h) · shockt + γ · controlst + εt+h
Baseline (binary shock): 2 shock lags, 1 response lag, 11 month dummies. Bill-surprise specification adds a tax-receipt surprise control. Newey-West HAC SEs (bandwidth = horizon). Placebo tests at h=−4 to h=−1.
Caveats: (1) Identifies issuance-driven rebuilds only. (2) h=−1 ON RRP pre-trend from same-week announcement. (3) Shock persistence (autocorr=0.86) inflates h>8. (4) CMB robustness supports direction but has own pre-trends.
All data from free, official U.S. government sources. No proprietary data.