Central-Bank Operations, Reserves & the Repo/Money-Market Complex

Branch: Money & Monetary Systems (Finance-Plus Ring) · Node MS1.03 · ~18 focused hours · Mastery gate ≥85%

M7.02 already taught you what a central bank is and for, its three hats, its mandate, its policy rate and corridor, the mechanics of QE and QT, and how to read a policy statement. That module gave you the current numbers too: the RBI's SDF/repo/MSF corridor, the Fed's target range, the size its balance sheet reached at the 2022 QE peak. This module does not repeat any of that. It takes the institution M7.02 built and hands you its actual books, because "the central bank cut rates" and "the central bank's balance sheet is $6.7 trillion" are journalism; the T-account underneath them, the arithmetic of a haircut, and the chain of collateral that turns one bond into three or four bonds' worth of financing are mechanics, and mechanics is what this module owns. Everywhere M7.02 described a lever narratively, this module puts a number through it: what does a central bank's balance sheet actually look like, line by line, and why do the RBI's and the Fed's look nothing alike; what is a reserve balance, mechanically, and why does it sit at a floor; what exactly happens, in double-entry, when a bond is repo'd, when the central bank buys $60 billion of bonds, when a money-market fund's net asset value stops being a dollar. This Ring is additive institutional/CFA breadth, it presents the toolkit and the plumbing alongside, not in place of, the program's process-over-P&L, concentration-over-diversification philosophy.

One standing warning before you start, because the program's own coordination notes flag this branch by name: the money-and-plumbing region is the single highest AI-over-trust risk in this corpus. A chat assistant will summarize a central-bank balance-sheet release or a repo-market note fluently, confidently, and (because the vocabulary sounds technical and the summary sounds authoritative) persuasively. It is also exactly the kind of content a model can get plausibly wrong in ways that are invisible in the prose: a T-account that doesn't actually balance, a haircut applied to the wrong side of a trade, a "central bank printed money" headline laundered into a paragraph that sounds like an explanation. The discipline this module drills, and will keep drilling, is that the "money view" (whose balance sheet just expanded, whose just contracted, and by how much) is a by-hand skill you verify against a primary document, never a paraphrase you accept because it reads smoothly. Every number below is Python-verified against either a real, named, dated primary source or an explicitly labeled synthetic construction, hold yourself to the same standard in the mini-project.

Learning objectives

By the end you can:

1. Read an actual central-bank balance sheet as a T-account, the Federal Reserve's H.4.1 release and the Reserve Bank of India's Weekly Statistical Supplement / Annual Report, reconcile each to its reported total, and explain why the RBI's book is dominated by foreign-currency assets and gold while the Fed's is dominated by domestic Treasury and mortgage securities.
2. Define a reserve balance precisely as a central-bank liability and derive, from arbitrage rather than assertion, why interest on reserve balances (IORB) is a floor: no bank rationally lends reserves overnight for less than it could earn risk-free by leaving them at the central bank.
3. Distinguish a floor system from a corridor system operationally, not as two labels but as two different relationships between the quantity of reserves supplied and the quantity demanded, and read the Fed's H.4.1 "factors supplying / absorbing reserves" framework, in which reserves are the balance sheet's residual, not a chosen target.
4. Turn every standing facility and open-market operation into a literal balance-sheet entry (an outright OMO purchase, a repo add, a reverse-repo drain, the SDF, the MSF, the discount window) so that "the central bank injected liquidity" becomes two ledger lines that must balance.
5. Run the QE and QT balance-sheet arithmetic end to end, entry by entry, for both a bank counterparty and a non-bank counterparty, reconnecting (in literal T-accounts, not narrative) the reserves-versus-broad-money distinction M7.01/M7.02 already established.
6. Compute a repo haircut, an initial margin, an effective leverage ratio, and a mark-to-market margin call from first principles, and read the difference between general-collateral (GC) repo and a security trading "on special."
7. Explain a money-market fund's architecture (WAM/WAL, government versus prime funds, amortized-cost stable NAV versus floating NAV) and reconstruct, numerically, how a fund's NAV "breaks the buck."
8. Trace the shadow-banking collateral chain: rehypothecation, collateral velocity, and the haircut spiral that turns a modest haircut increase into a large, fast deleveraging, the exact mechanism the module's Case Lab sequences.
9. (R10 duality objective.) Use an AI assistant to draft a summary of a central-bank balance-sheet release or a money-market primary document, then verify every figure against the actual named release (Guardrail 1–5, Trace through Cite) because in this region a fluent AI paraphrase and a correct one are indistinguishable until you check.

The duality, stated once (R10). Objectives 1–8 are the understanding gate (the balance sheet, the floor, the haircut arithmetic, the collateral chain) and they are what the mastery check grades. Objective 9 is the productivity payoff: AI is a genuinely useful first pass at reading a dense central-bank release or a fund prospectus, and a genuinely dangerous one to trust unverified, because this is a region where a confident, well-formatted, wrong balance sheet reads exactly like a correct one. The branch's angle is draft-then-verify-against-the-primary-source: let the tool draft the summary, then you open the actual H.4.1, the actual RBI Annual Report, the actual fund fact sheet, and check every figure yourself. No tool ever buys a pass on that check.

Prerequisites & connections

Builds on. M7.01 gave you the endogenous-money view (loans create deposits, base money versus broad money) and the double-entry mechanics of a single bank's balance sheet; this module assumes you can read a T-account without prompting and does not re-derive why banks create money. M7.02 gave you the central bank's mandate, the policy rate and its corridor (the RBI's SDF/repo/MSF levels and the Fed's target range, IORB and ON-RRP, referenced below exactly as M7.02 stated them, never re-derived), the OMO/CRR/SLR/forward-guidance toolkit narratively, the QE/QT channels and the reserves-versus-broad-money argument, and the M0–M3 aggregates. This module is the "one level down" complement M7.02 itself points toward: everywhere that module said what a tool does, this one shows the balance-sheet entry that is the tool. M5.01/M5.02 gave you a commercial bank's own balance sheet (CRR, SLR, CASA, the HTM/AFS/FVTPL investment classification) which is the counterparty side of every repo and OMO entry here. M1.01's double-entry bookkeeping is the literal skill in constant use throughout.

Feeds forward. MS1.04 (the crown of this branch) takes the domestic plumbing built here (repo, collateral, the shadow-banking chain) offshore: the eurodollar system, FX-swap funding, CLS settlement, and CCP clearing are this module's mechanics operating across a currency border rather than within one central bank's balance sheet. FI1.04 already flagged the connection in the other direction, its structured-credit and MBS machinery is funded, in practice, through the repo market this module builds; a dealer warehousing an MBS position before it is securitized is running exactly the collateral-and-haircut arithmetic of Worked Examples 3 and 6 below. FI1.03 bootstraps a curve whose very first point is the overnight repo/OIS rate this module explains the formation of. M7.03's yield-curve signal-reading and M7.06's FX-and-eurodollar macro both sit one level up from the mechanics here; this module does not re-derive either.

One-sentence version. A central bank's balance sheet is a literal, always-balancing T-account whose composition tells you what kind of monetary system you are looking at; its policy rate is enforced by paying interest on reserves at a floor and bracketing the market with standing facilities, not by rationing a fixed quantity of money; and the repo market that sits beneath it (where collateral is haircut, re-pledged, and financed through a chain of money-market funds and dealers) is the plumbing that turns "the central bank hiked" into a price every borrower in the economy actually pays, and that same plumbing, run in reverse under stress, is how a modest haircut increase becomes a systemic deleveraging in 72 hours.

Case Lab: A Repo Run in 72 Hours (CL-MS.2) sequences exactly that reversal (a haircut/collateral spiral, a money-market fund breaking the buck, and the central-bank facility response against the IORB floor) using the same arithmetic Worked Examples 3, 5 and 6 build below. Open it after Worked Example 6.

Core teaching content

Why a balance sheet, not a headline: what "one level down" means here

A newspaper reports that "the Fed's balance sheet is $6.7 trillion" the way it reports a company's revenue: as a fact to be impressed by, not a statement to be checked. An analyst reads it the way you have been trained to read every other balance sheet in this program, as a claim that must reconcile, line by line, to a total, with every line meaning something specific about who owes whom. A central bank is, after all, only an unusually powerful bank: it has assets, it has liabilities, the two sides are equal by construction (double-entry does not bend for institutions with the power to create their own liabilities), and its liabilities happen to be the safest, most liquid instrument in the financial system: currency and reserves. Reading that balance sheet as a T-account, rather than as a headline number, is the single skill this module opens with, because everything that follows (the floor, the OMO, the QE entry, the repo, the haircut) is just this same T-account skill applied to a smaller and smaller balance sheet, one desk down, until you reach a single repo trade between two counterparties.

Standing facilities and open-market operations, as literal entries

M7.02 named every tool in this list; this section is where each one becomes two ledger lines. An outright OMO purchase (the central bank buys a government bond and keeps it): central-bank assets, securities +X; central-bank liabilities, reserves (credited to the selling bank) +X. An outright OMO sale: the mirror image, both −X. A repo (the central bank lends cash overnight against collateral it does not keep, a temporary reserve injection): central-bank assets, "securities purchased under agreements to resell" +X (a receivable, not an outright holding); central-bank liabilities, reserves +X; and the entry unwinds the next day, both −X, when the bank buys its collateral back. A reverse repo (the central bank borrows cash overnight, draining reserves, the ON-RRP facility from the previous section): central-bank liabilities swap composition, reserves −X and "reverse repurchase agreements" +X; total liabilities unchanged, but the reserve component specifically shrinks. The SDF (India: banks deposit surplus cash at the RBI, unsecured, earning the floor rate) and IORB (US: the same idea, paid on the reserve balance itself) both work on the liability side only, they set the price at which a given quantity of reserves sits, without changing that quantity. The MSF/discount window/SRF (the ceiling: banks borrow against collateral at a penalty-ish rate) is a mirror of the repo entry above, initiated by the bank rather than the central bank, and priced at the corridor's top rather than the policy midpoint.

Notice what every single one of these operations shares: the size of the central bank's balance sheet only changes for the outright purchase/sale and (temporarily) for the repo/reverse-repo entries; the SDF, IORB, MSF and discount window change the composition and price of the liability side, never its total. That is the precise, mechanical content behind M7.02's claim that a central bank "sets the price of reserves, not their quantity by fiat", the standing facilities are pure pricing tools, and only the OMO family actually moves the balance sheet.

QE and QT, entry by entry

M7.02 gave you QE's three channels (duration/portfolio-balance, signalling, portfolio-rebalancing) and the reserves-versus-broad-money argument narratively. Here it is as literal double entries, which is where the argument becomes undeniable rather than merely persuasive. Take a $60 billion QE purchase of Treasury securities and run it two ways, because the counterparty is what decides whether broad money moves at all.

Case A, the seller is a bank, trading out of its own portfolio. Fed: securities +$60,000mn, reserves (credited to that bank) +$60,000mn, the Fed's balance sheet grows on both sides. The bank: securities −$60,000mn, reserves +$60,000mn, a pure asset swap on the bank's own book; its total assets are unchanged, and no new deposit is created anywhere. Broad money: $0 created.

Case B, the seller is a non-bank (a pension fund) that banks at "Bank X." Fed: securities +$60,000mn, reserves (credited to Bank X) +$60,000mn, identical to Case A on the Fed's own book. But now Bank X's balance sheet actually grows: reserves +$60,000mn (asset) and a brand-new deposit for the pension fund +$60,000mn (liability). The pension fund's own book shows an asset swap (securities out, deposit in), but that deposit is new, spendable broad money that did not exist before the trade. Broad money: $60,000mn (=$60bn) created.

Base money (reserves) rose by exactly $60bn in both cases. Only Case B moved broad money, and only because the seller was outside the banking system. This is M7.01/M7.02's reserves-versus-broad-money point, finally nailed down in accounting entries rather than asserted in prose: whether QE creates spendable money at all depends entirely on who is on the other side of the trade, a fact no amount of "the Fed printed $60bn" headline-reading can surface.

QT, in its simplest net form, is the mirror image: when a Treasury security the Fed holds matures and is not reinvested, the Fed's balance sheet shrinks (securities −$60,000mn, reserves −$60,000mn) both sides down, balancing exactly, the reverse of a QE entry. (One honest simplification: the real-world sequence runs through the Treasury General Account as an intermediate step, the Treasury refinances the maturing bond by selling new debt to the private market, whose buyers pay via reserves into the TGA, and the TGA then pays down the Fed's maturing holding, so the reserve drain actually happens the moment the private market absorbs the new issuance, not at the instant of maturity. The net two-line entry above is the correct final answer; the TGA is where an analyst watches the timing of that drain in real time, which is exactly why a rising Treasury General Account balance is itself a reserves-draining event even before the Fed's own balance-sheet total moves, the mechanism behind the September 2019 US repo-rate spike discussed below.)

Since M7.02's own text put the Fed's balance sheet at "roughly $9 trillion by 2022" (FRED series WALCL), and this module's Worked Example 1 pulls the actual July 2026 total of $6,747,378mn ($6.747tn), you can already state the balance-sheet story precisely: ≈$2.253 trillion below the 2022 peak, a 25.0% decline, a real and Python-checkable number, not a description. Read it as history, not as run-off still under way: the FOMC ended balance-sheet reduction on 1 December 2025 and began reserve-management purchases of Treasury bills that month, so the July 2026 total sits a little above the trough (check on federalreserve.gov).

Haircuts and collateral: initial margin, effective leverage, and the margin call

A haircut is the percentage by which a lender discounts collateral below its market value before lending against it, the buffer that protects the cash lender if the collateral's price falls before the trade unwinds. It is the single most important number in the entire repo market, because it is also, mechanically, a leverage dial: a 2% haircut on a bond means a dealer can finance 98% of that bond's value with borrowed cash and needs to fund only the remaining 2% itself, an effective leverage of 1 ÷ 0.02 = 50 times on that position. Raise the haircut to 10% and the same dealer can only finance 90%, cutting leverage to 10x; raise it to 25% (a crisis-level haircut on lower-quality collateral) and leverage falls to 4x. A haircut is therefore not an insurance premium, it is a deleveraging lever, and the fact that haircuts fall in booms (cheap, abundant financing, rising leverage) and spike in busts (expensive, scarce financing, forced deleveraging) is exactly the procyclical amplifier Adrian and Shin's work on financial-system leverage describes, and the mechanism behind every real repo-market crisis, 2008 included.

Worked Example 3 below runs the full numeric sequence, collateral value, cash raised, required collateral to raise a target amount of cash, the implied leverage, and a next-day margin call when the collateral's price falls and its market value drops below what the outstanding cash requires. Hold the shape of that sequence here: cash lent = collateral value × (1 − haircut); equivalently, required collateral = cash needed ÷ (1 − haircut); and a margin call fires whenever the collateral's current market value falls below that same required-collateral figure, demanding either fresh collateral or a partial unwind of the cash. Every one of these is ordinary arithmetic, but it is arithmetic that determines, in a real crisis, whether a dealer survives the week.

The shadow-banking chain: rehypothecation, collateral velocity, and the haircut spiral

"Shadow banking" is a term of art (coined and popularized by researchers including Paul McCulley and, in its systematic mapping, Zoltan Pozsar) for credit intermediation that performs genuinely bank-like functions (maturity transformation, liquidity transformation, leverage) entirely outside the regulated deposit-taking banking perimeter: money-market funds, repo desks, securitization vehicles, and the dealers that connect them. None of it is shady or illegal; all of it is heavily documented by the Financial Stability Board's annual shadow-banking (now "non-bank financial intermediation") monitoring reports. What it lacks, by design, is the backstop a regulated bank has almost by default (deposit insurance, routine discount-window access) which is exactly why, when it seizes up, it seizes up fast and needs its own emergency facilities built from scratch under time pressure (2008's alphabet-soup of Fed lending programs; 2020's commercial-paper and money-market facilities were the same improvisation, faster).

The chain's defining mechanism is rehypothecation: a dealer that receives a bond as collateral is often permitted, under the terms of the agreement, to re-pledge that same bond onward to raise its own financing, and the counterparty who receives it may do the same again. Each re-pledge applies a fresh haircut, so the same underlying $100 million bond can finance more than $100 million of cumulative borrowing across the chain, a phenomenon the IMF's Manmohan Singh formalized as collateral velocity: the ratio of cumulative financing raised to the original collateral's value. Worked Example 6 computes this precisely at three different haircut levels and shows the same chain collapsing as haircuts rise, the haircut spiral that is the actual mechanism of a repo-market deleveraging: no one has to sell a single bond outright for the system's total financing capacity to shrink dramatically; raising every haircut in the chain by a few points is entirely sufficient, and it happens in days, not quarters. This is the precise machinery the Case Lab sequences under a 72-hour clock.

Reading the plumbing as an analyst: what to actually watch

Tie the whole module together the way M7.02 §4.10 tied the policy rate to every valuation you build: this plumbing is not trivia, it is a set of live dashboard indicators. Watch the EFFR-minus-IORB gap (or the WACR-versus-repo gap in India): a gap that is small and stable says reserves are ample and the floor is holding cleanly; a gap that widens or turns volatile is the market's own signal that reserves are getting scarcer than the central bank believes, exactly what happened in the well-documented September 2019 US repo-rate spike, when overnight repo rates briefly jumped toward 10% (roughly 6 percentage points above the prevailing fed funds range) because a coincidence of heavy corporate tax payments and Treasury settlement drained reserves further than the "ample reserves" assumption anticipated, forcing the Fed to resume active reserve-supplying operations it had stopped years earlier. Watch ON-RRP facility usage: a large balance sitting there says money funds have more cash than attractive private-market places to put it (reserves/collateral abundant); a shrinking balance says that cash is finding its way into repo, bills, or bank deposits instead, often a precursor to reserves tightening elsewhere. Watch the Treasury General Account: a TGA rebuild (common right after a debt-ceiling resolution, as new debt issuance is absorbed before the cash is spent back out) drains reserves from the banking system exactly the way QT's fuller mechanics describe, independent of anything the Fed itself is doing. None of this replaces M7.02's transmission-mechanism or policy-statement-reading skills, it is the plumbing-level cross-check that tells you whether the "ample reserves" story the central bank is telling itself is still actually true.

{
  "skill": "Summarizing a dense central-bank balance-sheet release or a money-market/repo primary document without letting a fluent paraphrase substitute for the actual figures",
  "use": "Have an AI assistant produce a first-pass summary of a central-bank statistical release (the Fed's H.4.1, the RBI's Weekly Statistical Supplement) or a fund's prospectus/fact sheet — which lines changed, by how much, in which direction — as a starting map, then verify every single figure against the primary document yourself before it enters any analysis.",
  "tools": ["Chat assistants — Claude, ChatGPT, Gemini", "Document/PDF AI run directly over the downloaded H.4.1 PDF or the RBI Annual Report's Statement of Accounts", "A notebook environment to independently re-total the T-account"],
  "prompt": "Here is this week's [H.4.1 release / RBI Weekly Statistical Supplement]. Summarize which balance-sheet lines changed since last week/last release, by how much, and in which direction. Then list every dollar/rupee figure you cited so I can check each one against the source document myself.",
  "verify": "A summary that reads fluently and cites plausible-looking numbers is not the same as one where every number is correct — this is exactly the region where a wrong figure is invisible in the prose. Re-total the T-account yourself from the actual release (assets must equal liabilities plus capital, to the dollar/rupee) and confirm every cited change against the primary document's own prior release, never against the AI's restated version of it (Guardrail #1 Trace, #2 Match, #3 Recompute).",
  "diy": "You must already be able to read a central-bank balance sheet as a T-account and know which lines are base money versus not — this Core section is exactly that skill. AI accelerates finding what changed; it cannot certify that its summary of the change is correct.",
  "market": "US"
}

{
  "skill": "Using AI to draft a repo/haircut or collateral-chain calculation without letting a plausible-looking number substitute for a verified one",
  "use": "Ask an AI assistant to set up or check a repo haircut, margin-call, or collateral-velocity calculation — then treat its output as a draft to be independently recomputed, not as the answer, because this arithmetic is exactly the kind a model can get subtly wrong (haircut applied to the wrong side, a margin call computed against the wrong reference value) while still looking completely correct.",
  "tools": ["Chat/code assistants — Claude, ChatGPT with code execution", "A notebook environment (Jupyter/Colab) to independently rerun every formula"],
  "prompt": "A dealer posts Rs 100 crore face value of a G-Sec (price 102) as collateral in a tri-party repo at a 2% haircut. Compute: (1) the cash the dealer can raise, (2) the collateral required to raise exactly Rs 100 crore of cash, (3) the implied effective leverage, and (4) the margin call if the bond's price falls 1.5% the next day. Show each formula, not just the final numbers.",
  "verify": "Recompute all four figures yourself from the definitions in this Core section (cash = collateral x (1-haircut); required collateral = cash / (1-haircut); leverage = collateral/margin; margin call = required collateral value minus new market value) before trusting any AI-produced figure — an AI can silently apply the haircut to the wrong base or compare the margin call against the wrong reference value while producing a table that looks perfectly formatted (Guardrail #3, Recompute, is the entire discipline here).",
  "diy": "You must already be able to state and apply the haircut/leverage/margin-call formulas from memory — Worked Example 3 below is exactly that drill. AI can format the table; it cannot own whether the arithmetic inside it is right.",
  "market": "IN"
}

Common mistakes & how experts think differently

"Every central bank's balance sheet looks basically the same." It does not, and the reason is structural, not incidental: the Fed's book is ~96% domestic Treasury and MBS holdings because the US issues the currency the rest of the world holds as a reserve asset; the RBI's book is roughly 71% foreign-currency assets and gold because India manages a currency that is not the world's reserve currency and must hold a war chest to defend it. Reading "the RBI's balance sheet grew 20.6%" as equivalent in kind to "the Fed's balance sheet is $6.7 trillion" (rather than as two structurally different animals) is the novice's error; the expert asks what is this balance sheet's composition telling me about this country's monetary role before comparing sizes at all.

"Reserves are a quantity the central bank sets and holds fixed." The H.4.1 release's own title ("Factors Affecting Reserve Balances") reveals the opposite: reserves are computed as the residual of everything else on the balance sheet (total assets minus currency, minus reverse repos, minus the TGA and other liabilities and capital), not chosen directly. A modern central bank targets the price of reserves (via IORB/ON-RRP or the SDF/MSF corridor) and lets the quantity float as a plug, the opposite of the textbook "the central bank controls the money supply by controlling reserves" story M7.01/M7.02 already retired at the broad-money level; this module retires its narrower, balance-sheet-level cousin.

"A repo is just a collateralized loan, call it what it economically is and move on." Economically, yes; legally, a repo is a sale-plus-repurchase, and that legal form is why repo counterparties get a bankruptcy safe harbor allowing immediate seizure and sale of collateral. The expert holds both facts at once: the safe harbor is exactly what makes repo cheap and abundant in calm markets, and exactly what makes a repo run self-reinforcing and fast in stressed ones, the same legal feature produces both effects, at different points in the cycle.

"A haircut is basically an insurance margin, a fixed cost of doing business." A haircut is a leverage dial: 1 ÷ haircut is the effective leverage a piece of collateral supports, so a haircut move from 2% to 10% is a leverage move from 50x to 10x on the identical bond, with no change in the bond's own price. Haircuts fall in booms and spike in busts, a well-documented procyclical amplifier (Adrian and Shin), which is why "collateral quality didn't change, but the haircut did" is one of the fastest ways real-world leverage collapses in a crisis, without a single forced asset sale.

"Money-market funds are basically bank deposits: safe, stable, insured." They are not deposits and (outside the rare exception of an MMF explicitly holding insured bank deposits as an asset) are not insured by DICGC or the FDIC; a stable $1.00 NAV was historically an accounting convention (amortized cost), not a guarantee, and pre-2016-reform prime funds could and did "break the buck" when credit losses pushed their true, mark-to-market value below the $0.995 shadow-NAV threshold. The expert treats a stable-NAV fund's $1.00 sticker price as a useful convention with a real, if rare, failure mode, not as cash.

"Shadow banking" means shady or illegal banking. It is a term of art for bank-like credit intermediation (maturity transformation, leverage, liquidity provision) conducted outside the regulated, deposit-insured banking perimeter (money funds, repo, securitization), entirely legal and monitored annually by the Financial Stability Board. The expert's concern is not legality; it is that this system provides deposit-like liquidity promises without deposit-like backstops, so it needs its own emergency facilities, built under time pressure, exactly when it is under the most stress.

"QT is just QE run backwards, so it's automatically gentle and safe." The arithmetic is symmetric, but the real-world mechanics are not risk-free: QT interacts with Treasury issuance and the Treasury General Account in ways that can produce sudden, localized reserve scarcity even while the aggregate reserve level still looks "ample" system-wide, precisely what the September 2019 US repo-rate spike (overnight rates briefly approaching 10%) demonstrated in public, forcing the Fed to resume active reserve-management operations. The expert watches the repo-rate-versus-IORB/EFFR gap as a real-time stress gauge throughout a QT cycle, not just the headline balance-sheet total.

"The corridor's width tells you how tight monetary policy is." Corridor width (the RBI's SDF-to-MSF spread, or the Fed's target-range width) is a plumbing design choice (how much day-to-day overnight-rate volatility the central bank is willing to tolerate) entirely separate from the level of the policy rate, which is the actual stance of policy. A wide corridor at a low rate is looser policy than a narrow corridor at a high rate; conflating "how bracketed is the market" with "how restrictive is policy" is a category error the expert never makes.

Worked examples

Worked example 1: Reading the Federal Reserve's balance sheet as a T-account (US, $)

You are handed the Fed's H.4.1 release dated July 23, 2026 (data as of July 22, 2026) and asked to reconcile it and diagnose the balance-sheet story since the 2022 peak (QT ended on 1 December 2025, and reserve-management bill purchases have added to the total since).

Solution. Step 1: reconcile the assets. Securities held outright = UST $4,515,661mn + MBS $1,944,788mn + federal agency debt securities (the small residual) $2,348mn = $6,462,797mn, matching the release exactly. Add other Fed assets (repos made, gold/SDR certificates, discount-window loans, premises) of $284,581mn: total assets = $6,462,797mn + $284,581mn = $6,747,378mn.

Step 2: reconcile the liabilities and capital. Currency in circulation $2,470,901mn + reserve balances $3,062,149mn + reverse repurchase agreements $352,703mn = $5,885,753mn of named liabilities; the residual (Treasury General Account, other deposits, capital) = $6,747,378mn − $5,885,753mn = $861,625mn. Total liabilities and capital: $5,885,753mn + $861,625mn = $6,747,378mn: matches total assets exactly. The T-account balances.

Step 3: reserves as the plug. Using H.4.1's own "factors supplying versus absorbing reserves" framing: supplying factors (total Fed assets) = $6,747,378mn; absorbing, non-reserve factors (currency + reverse repos + TGA/other/capital) = $2,470,901mn + $352,703mn + $861,625mn = $3,685,229mn. Reserves-as-plug = $6,747,378mn − $3,685,229mn = $3,062,149mn, exactly the reported reserve balance. Reserves are not a chosen input here; they are what is left over once every other line is accounted for.

Step 4, the QT story, as history. M7.02's own text already established the 2022 peak at "roughly $9 trillion." Decline from the peak, most of it run-off before QT ended on 1 December 2025: $9,000,000mn − $6,747,378mn = $2,252,622mn (≈$2.25 trillion), a 25.0% decline from the peak (as of the 2026-07-23 H.4.1 release; the peak figure is M7.02's rounded "roughly $9 trillion", so the percentage inherits that rounding. Pull the current release from the Federal Reserve's own site, federalreserve.gov/releases/h41/, before quoting either number, because both move every week.) Checkable, and perishable: the arithmetic is exact on the stated inputs and the inputs expire.

Worked example 2: Reading the Reserve Bank of India's balance sheet as a T-account (India, ₹)

The RBI's FY26 Annual Report (year ended 31 March 2026) reports a total balance sheet of ₹91.97 lakh crore, up from ₹76.25 lakh crore in FY25, with domestic assets at 29.1% of the total (up from 25.7%) and foreign-currency-assets-plus-gold-plus-overseas-loans at 70.9% (down from 74.3%). Reconcile the split and explain the direction of the share shift. Every reported figure here is taken from the RBI's Annual Report for the year stated: the total balance-sheet size, the domestic and foreign asset shares, the gold holding and the surplus transferred to the government. Verify each against the RBI Annual Report for the year in question (rbi.org.in) before quoting it anywhere. Each is restated annually and several are revised. The derived figures below recompute exactly from those inputs and do not change; what changes is the inputs.

Solution. Step 1: reconcile the growth. (₹91.97 − ₹76.25) ÷ ₹76.25 = 20.62%, matching the reported 20.6%; the absolute increase, ₹91.97 − ₹76.25 = ₹15.72 lakh crore, matches the reported figure exactly.

Step 2, split both years into the two buckets. FY26: domestic = ₹91.97 × 0.291 = ₹26.76 lakh crore; foreign+gold+overseas = ₹91.97 × 0.709 = ₹65.21 lakh crore (sum: ₹91.97 lakh crore, checks). FY25: domestic = ₹76.25 × 0.257 = ₹19.60 lakh crore; foreign+gold+overseas = ₹76.25 × 0.743 = ₹56.65 lakh crore.

Step 3, explain the share shift with the sub-line growth rates. The domestic bucket grew (₹26.76 − ₹19.60) ÷ ₹19.60 = 36.6%; the foreign+gold+overseas bucket grew (₹65.21 − ₹56.65) ÷ ₹56.65 = 15.1%. Because the smaller bucket (domestic) grew faster than the larger one (foreign+gold+overseas), and both grew slower or faster than the 20.6% blended total in a way that mechanically shifts weight toward the faster grower, domestic's share of the total balance sheet rose even though it remained the smaller bucket in absolute terms. The reported sub-line growth confirms the mechanism: domestic investments +44.9% pulled the domestic bucket up fast, while foreign investments (the majority of the "foreign" bucket) grew a modest +7.9%, even as gold, a minority sub-line inside that same bucket: surged +63.8%. A fast-growing minority (gold) could not offset a slow-growing majority (foreign investments) within its own bucket.

Step 4, the gold detail and the Issue/Banking Department split. RBI's 880.52 tonnes of gold split 312.32 tonnes (Issue Department) + 568.20 tonnes (Banking Department), confirming the Section-23 bifurcation is a live, reported structural feature, not a historical curiosity. Banking-Department gold alone rose in value from ₹4.32 to ₹7.06 lakh crore, a 63.4% increase (reported: 63.6%, the small gap being rounding in the source figures), pro-rating to an estimated total RBI gold value (Issue-department gold value inferred by tonnage, at the same market price, since only the Banking-Department figure was directly reported) of ₹7.06 + [₹7.06 × (312.32÷568.20)] = ₹7.06 + ₹3.88 = ≈₹10.94 lakh crore total, roughly 16.8% of the entire foreign+gold+overseas bucket. Step 5, the surplus check. Surplus transferred to the government: (₹2.87 − ₹2.69) ÷ ₹2.69 = 6.69%, matching the reported 6.7%.

Worked example 3: A repo trade: haircut, initial margin, leverage, and a margin call (India, ₹)

"Kaveri Gilts Trading (synthetic)" wants to raise ₹100 crore of overnight cash by posting a 10-year G-Sec, currently priced at 102 (per 100 face), in a TREPS-style tri-party repo at a 2% haircut, referencing an overnight rate of 5.30% (Act/365): close to the RBI's repo rate of 5.25% in mid-2026, per M7.02 (5.50% after the 7 October 2026 hike).

Solution. Step 1, cash raised from a fixed face value. ₹100 crore face at a price of 102 is worth ₹100cr × 102/100 = ₹102.00 crore of collateral. At a 2% haircut: cash lent = ₹102.00cr × (1 − 0.02) = ₹99.96 crore.

Step 2, collateral required to raise exactly ₹100 crore. Reframe: required collateral = cash ÷ (1 − haircut) = ₹100cr ÷ 0.98 = ₹102.0408 crore. The dealer's own money in the trade (the initial margin) = ₹102.0408cr − ₹100cr = ₹2.0408 crore. Effective leverage = required collateral ÷ initial margin = ₹102.0408cr ÷ ₹2.0408cr = 50.0×, a 2% haircut supports fifty-to-one leverage on the position.

Step 3: overnight interest. At 5.30% (Act/365): ₹100cr × 0.0530 ÷ 365 = ₹1.4521 lakh for the one night.

Step 4, the margin call. The next day the bond's price falls 1.5% (a rate back-up): new price = 102.00 × (1 − 0.015) = 100.470. New collateral value = ₹100cr face × 100.470/100 = ₹100.4700 crore, but the required collateral value, unchanged, is still ₹102.0408 crore. Shortfall = ₹102.0408cr − ₹100.4700cr = ₹1.5708 crore, demanded as fresh collateral or a partial unwind of the outstanding cash.

Step 5: specialness. If the general-collateral (GC) rate is 5.30% but this specific bond is in high demand (say, cheapest-to-deliver ahead of a futures expiry), it might trade "on special" at 4.75%: specialness of 5.30 − 4.75 = 55 basis points, or, in a genuinely acute squeeze, as low as 0.50%, specialness of 5.30 − 0.50 = 480 basis points. The lower special rate is the price a short seller pays, in foregone interest, to guarantee delivery of that exact bond.

Worked example 4: QE and QT, entry by entry (US, $)

The Fed conducts a $60 billion Treasury purchase. Show the full T-accounts for (a) a bank seller and (b) a non-bank seller, then show the QT mirror image.

Solution. Case A: bank seller. Fed: securities +$60,000mn, reserves (credited to the bank) +$60,000mn. Bank: securities −$60,000mn, reserves +$60,000mn, a pure asset swap; the bank's total assets are unchanged and no new deposit exists anywhere. Broad money created: $0.

Case B, non-bank seller (a pension fund) banking at "Bank X." Fed: securities +$60,000mn, reserves (credited to Bank X) +$60,000mn, identical to Case A on the Fed's own book. Bank X: reserves +$60,000mn (asset), new deposit for the pension fund +$60,000mn (liability). Bank X's balance sheet genuinely grows. Pension fund: securities −$60,000mn, deposit at Bank X +$60,000mn, an asset swap on its own book, but the deposit itself is new, spendable broad money. Broad money created: $60,000mn.

Reconciliation. Base money (reserves) rose $60,000mn in both cases: QE always creates reserves. Broad money rose only in Case B, and only because the seller sat outside the banking system. This is exactly why a decade of QE (2008–2019, heavily bank-intermediated and demand-constrained) coincided with below-target inflation, while any episode combining QE with direct, non-bank-targeted transfers (fiscal cheques landing straight in household deposit accounts) mechanically creates far more broad money per dollar of central-bank purchase, the same point M7.02 made narratively, now visible in the entries themselves.

QT, net. A $60bn maturing Treasury holding, not reinvested: Fed securities −$60,000mn, reserves −$60,000mn, both sides fall, the exact mirror of the QE entry, balance sheet shrinking by $60bn.

Worked example 5: A money-market fund breaks the buck (US, $)

"Northbridge Prime Reserve Fund (synthetic)" holds $1,000,000,000 of amortized-cost assets against 1,000,000,000 shares outstanding, priced at a stable $1.00 NAV. It holds $50,000,000 of commercial paper issued by "Cardinal Brothers Holdings (synthetic)," which defaults; recovery is estimated at 20 cents on the dollar. Does the fund break the buck?

Solution. Step 1, the loss. Loss = $50,000,000 × (1 − 0.20) = $40,000,000.

Step 2, the shadow NAV. Mark-to-market assets = $1,000,000,000 − $40,000,000 = $960,000,000. Shadow NAV = $960,000,000 ÷ 1,000,000,000 shares = $0.9600.

Step 3, the verdict. The breaking-the-buck threshold is a shadow NAV of $0.995; $0.9600 is well below it, so the fund must break the buck, repricing from $1.00 to $0.96 per share and, historically, suspending redemptions while the position is worked out.

Step 4, the historical anchor, kept separate. The real-world case this mirrors is the Reserve Primary Fund, whose NAV fell to $0.97 on September 16, 2008, after writing off roughly $785 million of Lehman Brothers commercial paper, a 3-cent-on-the-dollar break, the only US prime fund to break the buck in that crisis, and a widely documented episode (SEC, Treasury, and GAO post-crisis reporting). This module's $0.96 figure is a synthetic, round-number construction built for clean arithmetic, deliberately distinguished from the real historical $0.97 print.

Step 5, the floating-NAV contrast. Under the 2016-reform floating-NAV regime that now applies to institutional prime funds, this same $40mn loss produces no discrete "break": the NAV simply marks down smoothly to $0.9600 and the fund keeps transacting at that marked price, no redemption cliff, no first-mover advantage, no suspension. The arithmetic is identical; the institutional structure around it is what changed, and that structural change is the entire point of the reform.

Worked example 6: The shadow-banking chain: rehypothecation and the haircut spiral (US, $)

Dealer A holds a $100,000,000 bond and repos it to raise cash; the counterparty re-pledges (rehypothecates) that cash's underlying value onward through a four-hop chain, each hop taking a fresh haircut. Compute the chain's cumulative financing and collateral velocity at three haircut levels: 2%, 10%, and 25%.

Solution. Step 1, the 2% (calm-market) chain. Hop 1: $100mn × 0.98 = $98.000mn. Hop 2: $98.000mn × 0.98 = $96.040mn. Hop 3: $96.040mn × 0.98 = $94.119mn. Hop 4: $94.119mn × 0.98 = $92.237mn. Cumulative financing = $98.000 + $96.040 + $94.119 + $92.237 = $380.396mn. Collateral velocity = $380.396mn ÷ $100mn = 3.80×, the original $100mn bond has financed $380mn of cumulative borrowing across the chain.

Step 2, the 10% (tightening) chain. Hop 1: $90.000mn; Hop 2: $81.000mn; Hop 3: $72.900mn; Hop 4: $65.610mn. Cumulative = $309.510mn; velocity = 3.10×.

Step 3, the 25% (crisis) chain. Hop 1: $75.000mn; Hop 2: $56.250mn; Hop 3: $42.188mn; Hop 4: $31.641mn. Cumulative = $205.078mn; velocity = 2.05×.

Step 4, the spiral, quantified. Raising every haircut in this chain from 2% to 25% shrinks cumulative financing from $380.396mn to $205.078mn, a ($380.396 − $205.078) ÷ $380.396 = 46.1% collapse in achievable system-wide financing from the identical underlying $100mn bond, with no change whatsoever in the bond's own market price. This is the mechanism, not a side effect of one: raising haircuts across a collateral chain is, by itself, a full deleveraging event, and it can happen in days because it requires no one to actually sell anything, only to re-price the terms on which the same collateral keeps circulating. This is precisely what the Case Lab, A Repo Run in 72 Hours (CL-MS.2), asks you to sequence against a real clock, alongside a money-fund break (Worked Example 5's mechanics) and the central-bank facility response that eventually arrests it.

Practice set

Work each problem before reading its solution. Figures are synthetic unless a primary source is named; numeric answers to about ±1% unless stated otherwise.

P1 (guided). A hypothetical central bank's weekly release shows: securities held outright $4,200,000mn, other assets $180,000mn, currency in circulation $1,650,000mn, reverse repos $95,000mn, and reserve balances $2,100,000mn. Find the residual (TGA + other liabilities + capital) and confirm the balance sheet balances.

Solution. Total assets = $4,200,000mn + $180,000mn = $4,380,000mn. Named liabilities (excl. residual) = $1,650,000mn + $95,000mn + $2,100,000mn = $3,845,000mn. Residual = $4,380,000mn − $3,845,000mn = $535,000mn. Check: $1,650,000 + $95,000 + $2,100,000 + $535,000 = $4,380,000mn: balances.

P2 (guided). Using the same figures as P1, compute reserves as the "plug" from the factors-supplying/absorbing framework, and confirm it matches the reported $2,100,000mn.

Solution. Supplying factors = total assets = $4,380,000mn. Absorbing, non-reserve factors = currency + reverse repos + residual = $1,650,000 + $95,000 + $535,000 = $2,280,000mn. Reserves-as-plug = $4,380,000 − $2,280,000 = $2,100,000mn: matches exactly.

P3. A central bank's target range is 4.00%–4.25%. Following the Fed's standard operating practice (ON-RRP at the range floor; IORB set inside the range, typically 10–15bp below the ceiling), state a plausible IORB and ON-RRP rate, and explain why EFFR-equivalent should trade a few basis points below IORB rather than above it.

Solution. ON-RRP ≈ 4.00% (the range floor). IORB ≈ 4.10%–4.15% (inside the range, near but below the 4.25% ceiling). The overnight rate trades a few bp below IORB, not above it, because non-bank lenders (e.g., the FHLBs in the US) cannot earn IORB themselves and are willing to lend fed-funds-equivalent cash at a shade below it; banks that borrow that cash and redeposit it at IORB arbitrage the gap closed from below, never from above, since no rational bank would lend for less than it can earn risk-free at the central bank.

P4. A corridor system has a floor of 6.00% and a ceiling of 6.75%, symmetric around the policy rate. Find the policy rate and the corridor width in basis points.

Solution. Policy rate = midpoint = (6.00 + 6.75) ÷ 2 = 6.375%. Width = 6.75 − 6.00 = 0.75 percentage points = 75 basis points.

P5. A dealer posts $50,000,000 face value of a bond priced at 98 (per 100 face) as collateral at a 3% haircut. Compute the cash raised.

Solution. Collateral value = $50,000,000 × 98/100 = $49,000,000. Cash raised = $49,000,000 × (1 − 0.03) = $47,530,000.

P6. A dealer needs to raise exactly $75,000,000 in a repo at a 4% haircut. Find the required collateral value, the initial margin, and the effective leverage.

Solution. Required collateral = $75,000,000 ÷ (1 − 0.04) = $75,000,000 ÷ 0.96 = $78,125,000. Initial margin = $78,125,000 − $75,000,000 = $3,125,000. Leverage = $78,125,000 ÷ $3,125,000 = 25.0× (equivalently, 1 ÷ 0.04 = 25×).

P7. Using P6's trade, the collateral's market value falls 4% the next day. Compute the margin call.

Solution. New collateral value = $78,125,000 × (1 − 0.04) = $75,000,000. Required collateral value is unchanged at $78,125,000. Margin call = $78,125,000 − $75,000,000 = $3,125,000 of additional collateral (or a partial unwind) demanded.

P8. GC repo trades at 5.10%. A specific bond trades "on special" at 3.90%. State the specialness in basis points and explain who is paying it and why.

Solution. Specialness = 5.10% − 3.90% = 120 basis points. The party short that specific bond (needing to borrow and deliver it) is effectively paying the 120bp (accepting a below-GC lending rate on their cash) in exchange for guaranteed access to the exact security they must deliver; the cash lender accepts less interest because they value receiving that specific bond more than the extra yield.

P9. A central bank buys $40 billion of corporate bonds directly from an insurance company (a non-bank) that banks at "Meridian Bank." Construct the full T-account sequence and state how much broad money is created.

Solution. Central bank: securities +$40,000mn; reserves (credited to Meridian Bank) +$40,000mn. Meridian Bank: reserves +$40,000mn (asset); new deposit for the insurer +$40,000mn (liability). Meridian's balance sheet grows. Insurer: corporate bonds −$40,000mn, deposit at Meridian +$40,000mn (asset swap on its own book, but the deposit is new spendable money). Broad money created: $40,000mn, because the seller was a non-bank.

P10. A central bank lets $25 billion of its held bonds mature without reinvesting. State the net effect on its own balance sheet.

Solution. Securities −$25,000mn (asset); reserves −$25,000mn (liability). Both sides fall by $25,000mn; the balance sheet shrinks by that amount, the mirror image of a $25bn QE purchase.

P11. A prime fund holds three positions: $200mn at 1 day, $300mn at 45 days, $500mn at 80 days. Compute the fund's WAM and state whether it satisfies a 60-day WAM cap.

Solution. Total book = $200mn + $300mn + $500mn = $1,000mn. WAM = (200×1 + 300×45 + 500×80) ÷ 1,000 = (200 + 13,500 + 40,000) ÷ 1,000 = 53,700 ÷ 1,000 = 53.7 days: inside a 60-day cap.

P12. A stable-NAV fund holds $500,000,000 of assets against 500,000,000 shares. It holds $30,000,000 of paper from a defaulted issuer with an estimated 40% recovery. Compute the shadow NAV and state whether the fund breaks the buck.

Solution. Loss = $30,000,000 × (1 − 0.40) = $18,000,000. Mark-to-market assets = $500,000,000 − $18,000,000 = $482,000,000. Shadow NAV = $482,000,000 ÷ 500,000,000 = $0.9640: below the $0.995 threshold, so the fund breaks the buck and must reprice to $0.964.

P13. A $200,000,000 bond is rehypothecated down a 3-hop chain at a 5% haircut per hop. Compute the cumulative financing raised and the collateral velocity.

Solution. Hop 1: $200,000,000 × 0.95 = $190,000,000. Hop 2: $190,000,000 × 0.95 = $180,500,000. Hop 3: $180,500,000 × 0.95 = $171,475,000. Cumulative = $190,000,000 + $180,500,000 + $171,475,000 = $541,975,000. Velocity = $541,975,000 ÷ $200,000,000 = 2.71×.

P14. Using P13's chain, the haircut jumps from 5% to 20% (a stress event). Recompute the cumulative financing and state the percentage collapse.

Solution. At 20%: Hop 1 = $200,000,000 × 0.80 = $160,000,000; Hop 2 = $160,000,000 × 0.80 = $128,000,000; Hop 3 = $128,000,000 × 0.80 = $102,400,000. Cumulative = $160,000,000 + $128,000,000 + $102,400,000 = $390,400,000. Collapse = ($541,975,000 − $390,400,000) ÷ $541,975,000 = 28.0%, a substantial financing collapse from the same bond, purely from re-pricing the chain's haircuts.

P15. Classify each of the following as (i) base money, (ii) broad money but not base money, or (iii) neither: (a) currency in circulation; (b) a bank's reserve balance at the central bank; (c) a household's savings-account deposit; (d) a central bank's holding of a government bond.

Solution. (a) Currency in circulation, base money (a central-bank liability, part of M0). (b) A bank's reserve balance, base money (also a central-bank liability, the other half of M0 alongside currency). (c) A household's savings deposit, broad money but not base money (a commercial-bank liability, created by lending, counted in M2/M3 but not M0). (d) A central bank's bond holding, neither (it is an asset of the central bank, not a monetary liability at all).

P16. In one or two sentences, explain why a repo market can seize up faster than an ordinary secured-lending market when confidence cracks.

Solution. A repo's legal form (a sale-plus-repurchase, not a loan) gives the cash lender a bankruptcy safe harbor (the right to seize and immediately sell the collateral without waiting on an automatic stay) so the instant confidence cracks, the rational move for every lender is simultaneously to refuse to roll the trade and to be ready to sell the collateral into the market, with no legal friction slowing anyone down; that same safe harbor that makes repo cheap and liquid in calm markets is what makes the exit door frictionless, and therefore fast, in stressed ones.

Applied mini-project

Read two real central-bank balance sheets, build a haircut schedule, and diagnose the plumbing.

1. Pull the primary sources. Download the Federal Reserve's current H.4.1 release (federalreserve.gov/releases/h41/current/) and the RBI's most recent Weekly Statistical Supplement or the Annual Report's Statement of Accounts (rbi.org.in). Use whatever is current when you do this, the module's own figures (July 2026 for the Fed, FY26 for the RBI) will already be out of date by the time you read this.
2. Rebuild each as a T-account. For each central bank, lay out a clean two-column table (assets; liabilities & capital) that reconciles exactly to the reported total, the way Worked Examples 1 and 2 do. Classify every line as base money (M0) or not, per M7.01/M7.02's definitions, do not re-derive those definitions, apply them.
3. Compute reserves as the plug for the Fed's current release using the factors-supplying/absorbing framework, and confirm your figure matches the reported reserve-balance line exactly.
4. Build a four-asset haircut schedule. Choose four collateral types (a sovereign bond, a PSU/agency bond, an investment-grade corporate bond, and a listed equity), assign your own clearly-labeled illustrative haircut to each (state your reasoning, liquidity and credit quality, not a guess), and for a fixed $10 million (or ₹10 crore) cash need, compute the required collateral, the initial margin, and the effective leverage for all four side by side.
5. Read one real MMF-like vehicle. Pull the public fact sheet of one US government or prime money-market fund, or one Indian liquid/overnight debt mutual fund, and either read its disclosed WAM or reconstruct it from the disclosed portfolio if a maturity breakdown is given.
6. Write the diagnosis. In half a page, state whether reserves currently look ample or scarce relative to the corridor (cite the EFFR/WACR-versus-IORB/repo gap and the ON-RRP or SDF facility's usage level as your two data points) and name one thing you would watch for early warning of repo-market stress, tying it explicitly to the Case Lab's sequence.

Deliverable and self-check. One worksheet containing both reconciled T-accounts, the reserves-as-plug computation with its match confirmed, the four-asset haircut schedule with all three figures per asset, the fund WAM read, and the closing diagnosis paragraph. Score yourself out of 10, 2 points each for: (a) both T-accounts reconcile exactly to their reported totals, with every line sourced and dated; (b) the reserves-as-plug figure matches the reported reserve balance; (c) the haircut schedule is internally consistent (required collateral, margin, and leverage all correctly related by the same formula) across all four assets; (d) the fund's WAM is correctly read or computed and checked against the relevant regulatory cap; (e) the diagnosis names a specific, current EFFR/WACR-versus-floor gap and a specific facility-usage data point, not a general impression. Pass at 8/10 with no zero on any item.

Reading & resources

Everything below is an anchor to verify against and learn from, never a primary source to quote a live figure from, for that, always pull the central bank's or fund's own current release.

The money-view anchor (start here):

• Perry Mehrling, The Economics of Money and Banking, free Coursera course; the single best introduction to reading a balance sheet as a "who owes whom" money view. [Free] [Beginner–Intermediate]
• Perry Mehrling, The New Lombard Street: How the Fed Became the Dealer of Last Resort, the modern central bank as market-maker, not just rate-setter. [Paid] [Advanced]
• Zoltan Pozsar's shadow-banking and money-market plumbing notes (various, Fed/Credit Suisse/private-sector research years), the map of the collateral chain this module builds arithmetic for. [Free] [Advanced]

Primary documents (check every current figure here, never from memory):

• Federal Reserve, H.4.1 statistical release, "Factors Affecting Reserve Balances", the Fed's own weekly balance sheet. [Free] [Intermediate]
• FRED (St. Louis Fed) series WALCL, TREAST, WSHOMCB, RESPPLLD, IORB, SOFR, for pulling the same data programmatically. [Free] [Beginner–Intermediate]
• Reserve Bank of India, Weekly Statistical Supplement and Annual Report (Statement of Accounts), the RBI's own balance sheet, both departments. [Free] [Intermediate]
• Bank of England, "Money creation in the modern economy" (2014): already your anchor from M7.01; re-read its balance-sheet exhibits with this module's T-account discipline. [Free] [Intermediate]

Repo, collateral, and shadow banking:

• Stigum and Crescenzi, Stigum's Money Market: the standing reference for US money-market and repo mechanics. [Paid] [Advanced: reference]
• Gorton and Metrick, "Securitized Banking and the Run on Repo", the academic account of the 2007–08 repo-market run. [Free (NBER working paper)] [Advanced]
• Manmohan Singh (IMF), papers on collateral velocity and rehypothecation in the shadow-banking system. [Free] [Advanced]
• Adrian and Shin, "Liquidity and Leverage", the procyclical-haircut/leverage mechanism this module's spiral example builds on. [Free (Fed working paper)] [Advanced]
• Financial Stability Board, annual Global Monitoring Report on Non-Bank Financial Intermediation ("shadow banking monitoring report"), the sober, regulator's-eye map of the sector's size and risks. [Free] [Intermediate–Advanced]
• CCIL (India), documentation on the Tri-Party Repo Dealing System (TREPS) and the negotiated dealing system. [Free] [Intermediate]
• ICMA, European Repo and Collateral Council publications on haircuts and margining conventions. [Free] [Intermediate–Advanced]

Money-market funds:

• SEC, 2016 Money Market Fund Reform and 2023 amendments (rule text and adopting releases), the floating-NAV/liquidity-fee framework directly. [Free] [Advanced]
• SEC and GAO post-crisis reporting on the Reserve Primary Fund's 2008 break. [Free] [Intermediate]

Tooling (for the Modern Data Analysis callout and the mini-project):

• pandas-datareader / fredapi documentation: pulling FRED series programmatically. [Free] [Beginner–Intermediate]
• pandas documentation, organizing and reconciling balance-sheet time series. [Free] [Beginner]

Do this, not just read: pull this week's actual H.4.1 release and this month's actual RBI Weekly Statistical Supplement, reconcile each to its own reported total the way Worked Examples 1 and 2 do, and only then start the mini-project, reading someone else's already-reconciled table teaches far less than reconciling a live one yourself.

Teach it back & journal

Feynman prompt. Explain to someone who has just read a headline saying "the Fed's balance sheet grew by $X billion this week" why that sentence, by itself, tells them almost nothing, using the reserves-versus-broad-money T-account from Worked Example 4 (a bank-seller QE purchase creates $0 of new spendable money; a non-bank-seller purchase of the identical size creates the full amount). Then explain, using the haircut-as-a-leverage-dial idea, why a headline reading "haircuts on repo collateral have risen" is actually a deleveraging story even though not a single bond has been sold. If your listener can, unprompted, ask "was the seller a bank or a non-bank?" the next time they read a QE headline, you have installed this module's central reflex.

Journal prompt. Pull this week's actual Fed H.4.1 release or RBI Weekly Statistical Supplement and reconcile it fully, the way Worked Examples 1 and 2 do, without looking at this module's own numbers as a crutch. Log: did every line reconcile cleanly on the first attempt, or did you have to hunt for a residual/plug line the way this module did for the Treasury General Account? Did the reserves-as-plug computation match the reported figure exactly, and if not, what did you miss? Then write a paragraph connecting this module's central discipline (a central bank's press release is a claim to be reconciled, not a headline to be believed) to the Guardrail's Trace-Match-Recompute-Date-check-Cite sequence from AI0.01: which of those five checks did you find yourself running without being told to, and which one would you have skipped if this module had not drilled it?

This module's flashcards and mastery quiz are wired into the app: see the node's Quiz and Reviews.

Applied mini-project rubric recap

Score the mini-project above out of 10 (2 points per criterion a–e listed there); pass at 8/10 with no zero on any item. This is the module's written-deliverable check alongside the Form A/B mastery gate, both are expected, neither substitutes for the other.