Quantitative Easing Explained: How QE Works and Its Risks

A central bank normally steers monetary conditions through a short-term policy interest rate. Quantitative easing adds purchases of financial assets, financed by creating reserves, to influence conditions further along the financing chain. Those assets commonly include government bonds; eligible assets differ across countries and programmes.

Why buy bonds? If borrowing remains expensive or spending remains weak after substantial rate cuts, purchases can put downward pressure on longer-term financing costs. The aim is to support demand and the central bank’s employment or inflation objectives. QE is often associated with rates near their effective lower bound, where further cuts have limited practical scope. That bound need not be exactly zero. Bank of England’s introduction and ECB’s programme explanation describe this purpose.

Buying a bond already held by an investor exchanges one financial asset for another. It does not build a factory or purchase someone’s groceries directly. The connection to everyday activity runs through financing, portfolios, confidence and subsequent spending decisions. Keep three questions separate when reading a QE announcement: what the central bank purchases, how it pays, and why those transactions might change behaviour.

The Bank of Canada explains settlement balances as a form of central bank deposit. A household’s savings balance and its bank’s reserve balance belong to different balance sheets.

“Printing money” is an imprecise description of electronic reserve creation. Saying “QE only creates reserves” can also mislead: purchases from non-banks can simultaneously create deposits. Ask whose money balance changed. That question resolves much of the confusion.

An announcement normally specifies eligible securities and a purchase plan. The central bank or its dealing arm then buys through approved counterparties under the programme’s rules. Government bond purchases in the examples here concern the secondary market, where existing securities trade, rather than a payment to the government for newly issued debt.

At settlement, the central bank receives the security and credits the relevant reserve account. If the ultimate seller has no reserve account, its commercial bank handles the payment and credits its deposit. A dealer can intermediate; the dealer and the final economic seller need not be the same entity. The Bank of Canada’s operational account explains its purchases and reserve funding.

The examples below use a fictional $100 transaction. Dollars are units for arithmetic, not a claim about a particular Federal Reserve settlement. Assume the bond’s carrying value equals its sale price, settlement is immediate, and there are no fees, capital gains, taxes or other simultaneous transactions. Real operations have additional counterparties and accounting detail.

Suppose Bank A owns a government bond worth $100. The central bank buys it for $100. These are changes, not the institutions’ complete balance sheets. Each row follows assets = liabilities + equity.

The central bank’s balance sheet grows by $100 on each side. Bank A swaps a security for reserves; its total assets remain unchanged in this example. Its customers receive no new deposit from this transaction. Its equity does not increase merely because it sold the bond at carrying value.

Balance the arithmetic before proposing an economic story: Bank A’s −100 + 100 equals zero, and the central bank’s new asset equals its new liability. This purchase creates reserves, but it does not by itself create a household loan or a customer deposit. The bank might subsequently alter its portfolio or lending terms; that is a further decision. The Bank of England’s 2009 explanation distinguishes bank and non-bank sellers.

Now suppose a fictional pension investor owns the $100 bond and banks with Bank A. It sells the bond to the central bank, using Bank A for settlement.

There are two new balances, with different holders. The investor receives a deposit; the bank receives reserves. Bank A’s balance sheet expands because the deposit is its liability. The investor exchanges a bond for a deposit without becoming $100 wealthier at the assumed unchanged price. Market-wide valuation effects are a separate issue.

In this simplified closed example, customer deposits increase by $100 without a new loan. Specific monetary aggregates may exclude particular institutional deposits, so “deposits increased” does not imply that every published broad-money measure rises by exactly $100.

If the investor buys another asset from a customer at another bank, the deposit can move to that seller and reserves settle between banks. Moving a deposit is different from extinguishing it. This does not determine whether the recipient spends on goods, buys another asset or repays debt. Our numbers are an original teaching example; the Bank of England’s money-creation paper explains the underlying distinctions.

Banks can transfer reserves to other institutions with reserve accounts. Ordinary customers do not borrow those balances directly. When a bank originates a loan, it normally records a loan asset and credits a customer deposit. Payments may then require reserve transfers between banks. Bank of England’s account of money creation explains why lending is more than redistributing pre-existing savings.

For a separate fictional $20 loan, Bank A records loan +$20 on the asset side and borrower deposit +$20 on the liability side. That loan creation entry does not subtract $20 from its reserve balance. A later payment is another entry. Trace those stages instead of assuming one container of money is emptied into another.

Credit decisions still depend on creditworthy demand, capital, liquidity management, risk, funding costs and expected profitability. A reserve increase is not a fixed multiplier of new lending. Carpenter and Demiralp’s Federal Reserve research examines why the standard reserve-to-loan multiplier is an inadequate account of US transmission. QE can affect credit indirectly, but more reserves alone do not specify how many loans will be made.

These are interacting mechanisms. D’Amico and coauthors’ Federal Reserve study analyses scarcity and duration effects on Treasury term premiums. The ECB explains portfolio rebalancing and signalling. Effects depend on the instruments bought and how readily investors substitute between assets.

A useful conceptual decomposition is long nominal yield ≈ average expected short nominal rates + term premium. It is a simplification, not a direct measurement of either component. St. Louis Fed’s discussion explains these channels and the role of expected inflation in real rates.

Lower benchmark yields may ease mortgage or corporate financing, but lenders’ margins and credit spreads can offset that movement. Stocks and currencies respond to earnings, risk and global news too. QE therefore does not guarantee a particular stock return, currency direction or stock–bond correlation.

Take a fictional default-free zero-coupon bond paying $100 exactly ten years from now. It pays no intervening coupons. With annual compounding, its price is P = 100 / (1 + y)^10.

The yield falls 50 basis points, or 0.50 percentage points. The price rises about $3.3355. This is checked arithmetic with chosen inputs, not an estimate of QE’s effect. Download the reproducible calculation and data checker.

For this bond, modified duration at 4% is 10 / 1.04 = 9.6154. The first-order approximation, −duration × yield change, gives +4.8077%. The exact result differs because the price–yield relationship is curved. Our script checks both results.

This is an instantaneous price comparison holding the maturity and payment fixed. It excludes elapsed time, transaction costs and funding. A coupon bond needs all its cash flows discounted. A Treasury par yield series is a curve statistic, not the price or return of this zero-coupon bond. A realized investment return depends on entry and exit prices, coupon receipts and, where relevant, reinvestment. Neither a yield level nor a basis-point change is itself a strategy return.

The Federal Reserve’s balance-sheet timeline records purchases, reinvestment and maturity extension separately. Its May 2022 runoff plan illustrates caps on non-reinvested principal. A cap is a maximum, not a promise that runoff equals that amount each month.

Distinguish the flow of purchases from the stock of securities still held and the holdings investors expect in future. Ending purchases need not remove all influence of the existing portfolio. Ihrig and coauthors explicitly model expected future holdings.

QT is not necessarily QE with an equal and opposite market effect: stress, surprise and implementation differ. Reserves also depend on other central bank liabilities, including currency and government accounts. Total assets and reserve balances need not move one-for-one. Bank of England’s 2022 review cautions against mechanical symmetry.

These are selected milestones, not current policy settings. Programme labels and operating targets differ across central banks.

“QE1”, “QE2” and “QE3” commonly describe US purchase rounds; “QE4” is an informal label often applied to pandemic purchases. Compare the actual announcements rather than treating those names as globally standardized policies.

The evidence is more informative than either “QE always works” or “QE does nothing.” The Bank of England’s 2022 review finds support for effects on yields and financing, with strength varying across episodes. Estimating GDP and inflation effects is harder: purchases respond to the same shocks that damage economic activity.

Federal Reserve research on duration and scarcity provides evidence for term-premium transmission, while research on expected holdings shows why the anticipated future portfolio matters. These are author studies with models and assumptions, not universal policy constants.

Before accepting a claimed effect, ask what was measured: an announcement-day yield change, a longer-horizon model estimate or an economic forecast relative to a no-QE scenario. A fall in unemployment after purchases is not automatically a measured QE effect. An economy that remains weak after QE may nevertheless be stronger than it would have been without it. The missing comparison is the difficult part.

Inflation and calibration. QE aims to support demand when inflation or activity is too weak. If stimulus persists as supply constraints bind, demand support can add inflation pressure. That does not identify how much of a particular inflation episode came from QE rather than fiscal transfers, energy shocks or other forces. CBO’s 2022 analysis explains how outcomes depend on economic capacity and policy response. “Reserves stayed inside banks” does not establish that QE cannot affect inflation.

Unequal household effects. Asset owners may gain from valuation support; borrowers, savers and workers face different effects. Bunn, Pugh and Yeates’ UK study estimates small effects on standard relative inequality measures for 2008–2014 monetary easing, with substantially different cash gains and age patterns. It considers QE together with rate cuts. That result is not proof that QE alone has the same distributional effect in every country or period.

Risk-taking and market functioning. Lower yields can encourage investors to seek more risk; Jeremy Stein’s 2013 Federal Reserve speech discusses this incentive. Large holdings can also affect securities availability and trading conditions. Evaluate leverage, liquidity and market structure, not just headline equity prices.

Public-sector interest exposure. Interest-bearing reserves can replace longer-term fixed-rate debt held privately. If short-term rates rise above portfolio earnings, central bank net income and transfers to government can fall. Selling a bond below its purchase price creates a different accounting issue from paying more interest on liabilities than assets earn. Federal Reserve analysis of balance-sheet interest-rate risk examines these mechanics; its projections are historical scenarios. CBO discusses uncertain net budget effects. A loss does not by itself establish policy failure: macroeconomic benefits and fiscal costs require separate assessment.

Consider the 3 November 2010 QE2 announcement. We retrieved these daily par-yield observations from the US Treasury’s 2010 XML feed, checked on 3 October 2026. Values are annual percentages; spread = 10-year minus 2-year.

From 2 to 3 November, the 10-year yield rose 4 basis points. From 3 to 4 November, it fell 14 basis points. Across 2 to 4 November, it fell 10 basis points, and the spread narrowed 9 basis points. Our CSV snapshot and calculation output make the arithmetic inspectable.

These observations establish neither that QE raised yields on announcement day nor that it caused the following decline. Different windows change the description. Treasury par yields are indicative curve estimates based on market quotations, not transaction prices or bond-return indexes. See Treasury’s statistics description.

Use the three-row exercise to practise careful measurement, then expand it into a research notebook.

1. Define the question first. For example: how did 2-, 5- and 10-year yields and their spreads move around a specified purchase announcement? This is descriptive. “How much did QE cause yields to fall?” is a separate identification question.
2. Build an announcement log before examining outcomes. Record publication date, timestamp, timezone, asset eligibility, announced amount, purchase horizon and accompanying rate guidance. Distinguish the announcement from later purchase settlement. Record what investors could know at the time.
3. Freeze source data and units. Save the official series, retrieval date and missing-value convention. One percentage point equals 100 basis points. Do not forward-fill a holiday as if a new market observation occurred.
4. Order observations in trading time. Set the last available close before the announcement as a baseline. Inspect whether an announcement falls before or after the daily observation. Intraday claims require appropriately timed intraday data, not this daily feed.
5. Choose windows in advance. Report the announcement day, next trading day and a prespecified wider interval. Do not select only the window that produces your expected sign. Plot the surrounding weeks to see anticipation and reversals.
6. Annotate competing information. Rate guidance, employment releases, inflation news, issuance, fiscal developments and changes in risk sentiment can affect the curve. A QE announcement can also reveal the central bank’s assessment of a weak economy.
7. Separate description from estimation. A daily difference is an observed difference. Causal work needs a defensible policy-surprise measure, comparison or model, with stated assumptions. Adding a “QE on/off” column to a regression does not resolve policy endogeneity.
8. Test sensitivity and report exclusions. Show alternative windows chosen for a reason, missing dates and other events. With multiple episodes, preserve chronological validation and avoid choosing rules using the eventual outcome.

The downloadable checker uses Python 3 and its standard library. Save the Python file and CSV snapshot in the same folder, then run python qe-worked-example.py there. Its default calculation uses the fixed CSV. To check it against the online Treasury feed, run python qe-worked-example.py --refresh; that command retrieves the official XML and verifies the three observations. It does not backtest a strategy. A compact final research note should distinguish the observations, the hypothesized channel, the confounds and the claim the data cannot support.

For a broad macro explanation, the balance-sheet examples and policy comparison above are sufficient starting points. For a research project, begin with a chart whose axes, dates and units you can explain. Add a source log and an alternative interpretation before adding a model.

Read the yield-curve research guide for a fuller discussion of sources, baselines and validation. Carry the same unit and timing checks into your own research: a model of quoted yield changes answers a different question from bond performance or the causal effect of a policy.

To describe a central bank’s policy today, check its latest decision, implementation note and balance-sheet release directly. Record the announcement date and distinguish announced targets from realized holdings. This guide uses dated historical examples rather than extending an old runoff cap or purchase pace into an unverified current-policy claim.