CNLI: Convex Net Liquidity Index
The Federal Reserve balance sheet corrected for the two largest non-bank drains, published as a dollar-denominated series with explicit thresholds and sign conventions.
Last reviewed: · Version 1.0 · Identifier CNLI-METHODOLOGY-v1.0
Abstract
The Convex Net Liquidity Index (CNLI) is a daily dollar-denominated measure of the stock of Federal Reserve balance-sheet assets that has been released into the private financial system. It is defined as total Federal Reserve assets (WALCL), minus the overnight Reverse Repurchase Facility take-up (RRPONTSYD), minus the Treasury General Account balance (WTREGEN). This construction isolates the portion of the Fed’s footprint that is actually supporting bank reserves, money-market liquidity, and collateral supply, rather than sitting in an account that the private sector cannot lend against or lever. The index is distributed on a dollar scale with three interpretation bands (Low, Moderate, Elevated). It is not a forecasting model; it is a flow-adjusted stock measure whose directional change leads risk-asset conditions at daily-to-weekly frequency.
1. Motivation
Practitioners frequently cite the Federal Reserve balance sheet (WALCL) as a liquidity proxy. The implicit model is that each dollar of asset purchases creates a dollar of reserves, and each dollar of runoff extinguishes a dollar of reserves. For the 2009 to 2014 period, this identity was approximately right: the Fed expanded assets, reserves rose roughly one-for-one, and risk assets rallied in near-linear correlation with the headline balance sheet.
After 2019, the identity broke. Two large non-reserve uses of the balance sheet appeared, and their combined variation became larger than the week-to-week change in total assets.
The first is the overnight Reverse Repurchase Facility (ON RRP). Introduced as a monetary-policy tool in 2013 and scaled up dramatically in 2021, ON RRP allows money market funds and eligible counterparties to deposit cash with the Fed overnight in exchange for Treasury collateral. Dollars parked at the ON RRP are drained from the banking system. They earn the administered rate but do not circulate as reserves, do not fund private credit, and do not support risk-asset positioning. ON RRP balances exceeded two trillion dollars in 2022-2023 before unwinding.
The second is the Treasury General Account (TGA). The TGA is the Treasury’s operating cash account at the Fed. When Treasury issues debt faster than it spends, the TGA builds up and reserves fall (issuance drains cash from the private sector into the TGA). When Treasury spends faster than it issues, the TGA falls and reserves rise. Around debt-ceiling episodes and quarterly tax dates, the TGA can move by several hundred billion dollars in a matter of weeks.
Both ON RRP and TGA are liabilities on the Fed’s balance sheet in accounting terms, offsetting assets. Headline WALCL is insensitive to whether a dollar of assets is financing reserves or financing the TGA. The CNLI is the minimal correction that separates these two cases.
2. The 2022-2023 paradox
The clearest motivating case is the 2022 to mid-2023 period. Headline WALCL fell from a peak of roughly 8.97 trillion dollars in April 2022 to roughly 8.3 trillion by summer 2023, a reduction of over six hundred billion dollars. On a pure balance-sheet reading this was tightening. Risk assets, after an initial sell-off, rallied strongly through the second half of this period.
The CNLI resolves the paradox. Over the same period the ON RRP drained from roughly 2.55 trillion dollars at year-end 2022 to below 400 billion by late 2023, releasing more than two trillion dollars of private-sector cash that had previously been sterilised. The TGA traced a separate but partially offsetting path around the 2023 debt-ceiling resolution. Net of both drains, CNLI rose for much of this window. Liquidity, properly measured, was expanding even as the headline balance sheet was shrinking.
This is not a retrospective coincidence. It is the mechanism by which the Fed was able to conduct quantitative tightening without inducing the reserves squeeze that many analysts predicted: the RRP drain absorbed the asset runoff. An analyst working off WALCL alone would have been short risk through a strong equity rally. An analyst working off CNLI had the right flow-adjusted signal in real time.
3. Formula and sign conventions
CNLI_t = WALCL_t − RRPONTSYD_t − WTREGEN_t
All three series are in billions of US dollars, point-in-time, non-seasonally adjusted. CNLI is reported at weekly frequency (Wednesdays) to match WALCL, with daily interpolation of RRPONTSYD and WTREGEN for intraweek updates where available.
Sign conventions are critical. Both subtracted terms are entered as positive numbers (the dollar balances themselves). An increase in RRPONTSYD or WTREGEN reduces CNLI. A decrease in either term increases CNLI. WALCL enters additively.
ΔCNLI_t = ΔWALCL_t − ΔRRPONTSYD_t − ΔWTREGEN_t
This identity is useful for attribution: a given week’s change in CNLI can be decomposed into the Fed policy channel (ΔWALCL), the money-market channel (ΔRRPONTSYD), and the Treasury channel (ΔWTREGEN). Convex publishes each channel separately alongside the composite.
4. Component definitions
4.1 WALCL, Federal Reserve Total Assets
WALCL is the total assets line of the Federal Reserve’s H.4.1 statistical release, published every Thursday for the prior Wednesday. It consists primarily of US Treasury securities held outright, agency mortgage-backed securities held outright, loans (including discount-window and emergency facilities), and repurchase agreements. WALCL moves when the Fed buys or sells securities (open market operations, quantitative easing, runoff), extends emergency credit, or executes temporary repo operations.
WALCL does not move when the Fed shifts cash between its own liabilities (for example from reserves to RRP). The CNLI correction captures exactly those shifts.
4.2 RRPONTSYD, Overnight Reverse Repurchase Agreements
RRPONTSYD is the daily outstanding amount of overnight reverse repurchase agreements between the New York Fed and approved counterparties, principally money market funds. Counterparties lend cash to the Fed and receive Treasury collateral overnight, earning the ON RRP administered rate (ON RRP rate). The facility serves as a floor under short-term rates: any counterparty that can earn the ON RRP rate risk-free will not lend at meaningfully lower rates elsewhere.
From the CNLI perspective, the mechanical effect is unambiguous: every dollar parked at the ON RRP is a dollar the private sector is choosing to hold at the Fed rather than deploy into money markets, T-bills, repo, or bank deposits. Those dollars are sterilised. The RRP has ranged from near zero (pre-2021 and late 2024 onwards) to over 2.5 trillion dollars (year-end 2022). RRPONTSYD is available at daily frequency from the New York Fed and is mirrored on FRED with a one-day lag.
4.3 WTREGEN, Treasury General Account
WTREGEN is the balance of the US Treasury’s operating account at the Federal Reserve, reported daily in the Daily Treasury Statement and published weekly on the H.4.1 (with daily availability via the Treasury’s own release). Treasury uses the TGA to manage disbursements: tax receipts flow in, spending and debt redemptions flow out, and debt issuance is the primary balancing mechanism.
A rising TGA drains reserves from the banking system (Treasury is issuing faster than it is spending, and the issuance proceeds sit at the Fed). A falling TGA releases reserves back to the banking system (Treasury is spending faster than it is issuing). The TGA can move by hundreds of billions of dollars around the April 15 tax date, the September 15 quarterly estimated-tax date, and around debt-ceiling resolutions. These swings dominate week-to-week variation in CNLI when they occur.
5. Interpretation thresholds
CNLI is published as a dollar level. The following thresholds are descriptive regime labels, calibrated against the distribution of CNLI observations from January 2015 through March 2026. They are not probability-of-outcome calibrations and should not be interpreted as such.
These thresholds are stock levels. At least as important as the level is thedirection of CNLI over the prior 30 and 90 days. A rising CNLI inside the Moderate band has different implications than a falling CNLI inside the Elevated band. Convex publishes both the level and the trailing change alongside the threshold label.
6. Known limitations
- CNLI is a stock, not a flow. The level of liquidity does not linearly translate into risk-asset returns. A stable CNLI at 5T is not the same as a CNLI that just rose from 4T to 5T. Users should always pair the level with a rate-of-change view.
- The three components are US domestic. CNLI does not incorporate BoJ, ECB, or PBoC balance sheets, or the flow of dollars from the Fed standing repo and FIMA facilities to foreign official accounts. In periods when global dollar liquidity is being driven from outside the US, CNLI will under-count the true global liquidity condition.
- Bank term funding and emergency facilities are included in WALCL but are not functionally equivalent to purchased securities.During the March 2023 regional-bank episode, the Fed’s Bank Term Funding Program added hundreds of billions to WALCL that were collateralised emergency loans, not outright purchases. CNLI counted these as liquidity, which is mechanically correct (reserves were created), but analytically they are short-dated and reversible, not structural.
- RRP composition matters. ON RRP usage is concentrated in government money market funds. If RRP drain happens because MMFs rotate into T-bills rather than deposits, the reserves impact is different from what a naive read would suggest. CNLI treats one dollar of RRP drain as one dollar of released liquidity regardless of where the money goes, which is a useful first-order approximation but not the full picture.
- TGA target changes are exogenous to policy.Treasury’s cash-balance target is set by the Treasury Borrowing Advisory Committee and can change without any Fed action. CNLI will move on TGA rebuilds that have nothing to do with monetary policy. This is a feature, not a bug, the index is measuring net liquidity, not monetary policy stance, but users should attribute large TGA-driven swings to their Treasury source rather than to the Fed.
- CNLI is not a forecasting model. It does not produce probabilities, Sharpe-tested signals, or calibrated forecasts. It is a flow-adjusted state measure. Users building trading or allocation models should treat CNLI as an input feature alongside rates, credit spreads, and positioning, not as a standalone buy or sell signal.
7. Data sources and update cadence
- WALCL: FRED, weekly (Thursday, for prior Wednesday), H.4.1 source
- RRPONTSYD: FRED, daily (business days), NY Fed Desk source
- WTREGEN: FRED, weekly (H.4.1) with daily cross-check against Daily Treasury Statement
CNLI is recomputed on every FRED sync (currently every 6 hours). On days when only RRPONTSYD has updated, the prior-Wednesday values of WALCL and WTREGEN are carried forward, so intra-week CNLI moves are driven by the RRP alone. The full weekly revision is applied on Thursday after H.4.1 releases. The component-level decomposition is published alongside the composite at /indicators/cnli.
8. References
- Board of Governors of the Federal Reserve System. Factors Affecting Reserve Balances of Depository Institutions (H.4.1 Statistical Release).
- Federal Reserve Bank of New York. Overnight Reverse Repurchase Agreement Operations. Policy documentation and operating statements.
- U.S. Department of the Treasury, Bureau of the Fiscal Service. Daily Treasury Statement (TGA balance).
- Copeland, A., Duffie, D. and Yang, Y. (2021). “Reserves Were Not So Ample After All.” NBER Working Paper 29090.
- Du, W., Hebert, B. and Li, W. (2023). “Intermediary balance sheets and the treasury yield curve.” Journal of Financial Economics.
- Logan, L. (2024). “Ample reserves and the Friedman rule.” Remarks at the European Central Bank Conference on Money Markets.