Quantitative Tightening: Follow the Fed’s Balance Sheet

A weekly Fed balance-sheet release shows fewer assets, yet reserve balances rise. Both observations can be true. A useful dollar-market reading starts with the liability side of the balance sheet and the cash accounts that changed alongside the securities portfolio.

Start with dated Federal Reserve H.4.1 table. Look for securities holdings measured separately from total assets. The calculation you want is the change in reserves implied by a balanced ledger. A common reason for a discrepancy is Treasury balances, reverse repos and other liability changes. The first worked figures below are illustrative. The historical section identifies the actual observations used and lets you change one assumption at a time.

Balance both sides of the ledger

In a simplified illustration assets fall USD 20 billion and Treasury’s account at the Fed falls USD 35 billion. If all other nonreserve liabilities and capital stay flat, reserves rise USD 15 billion. A reduction in another liability can more than offset the effect of declining assets on reserves. Write assets, reserves, other liabilities and capital in the same units before assigning signs.

A runoff cap is a ceiling. Actual maturities and mortgage principal payments can be below it. Loans, repos and valuation or settlement timing can change total assets independently. Read the holdings lines and reserve liability directly in H.4.1. When using a series measured in millions, divide by 1,000 to describe billions.

Date the policy regime

The Fed announced on October 29, 2025 that it would conclude the reduction of its aggregate securities holdings beginning December 1, 2025. A past episode of declining assets should carry its historical date. Check the Fed’s normalization decisions for the policy in effect on the date you intend to analyze.

For a dollar study, funding pressures, policy expectations and the foreign central bank can matter at the same time. A weekly snapshot is too coarse to establish that a particular intraday liquidity sweep was caused by runoff. Use the release to check a balance-sheet claim and separate time-stamped dealing prices to investigate the chart.

Distinguish the dates of announcement, settlement and observation

An announced runoff policy can remain in force while a particular week shows an increase in securities holdings or total assets. Treasury security maturities follow their calendars, mortgage principal payments arrive unevenly, and unrelated facilities can grow or shrink. Check the instrument, the maturity schedule, settlement conventions and the date each weekly observation covers. Treat an apparent exception as a reconciliation problem before calling it a reversal of policy.

A useful second reading of H.4.1 starts with the liability categories. A reduction in a reverse repo balance and a comparable increase in reserves can change the composition of liabilities with little change in total assets. A note that describes this as fresh asset purchases would misstate the event. Pair any claim about monetary policy stance with the actual implementation announcement and name the balance-sheet line used to support the claim.

How to check the figures yourself

Begin by aligning the same two Wednesdays across all balance-sheet items. A weekly average from one line and a Wednesday level from another will not reconcile cleanly. Check the units before subtracting and treat the residual as a prompt to inspect omitted balance-sheet items.

When a claim links runoff to a dollar move, specify the channel that would transmit it. A change in funding spreads, expected rates or liquidity is testable in its own market. A simultaneous decrease in assets and increase in a dollar pair is insufficient to identify any one channel.

Historical cases and a reproducible dollar comparison

To see what changes the dollar decision, take one clearly defined exposure: JPY 4,648,000 purchase during balance-sheet changes. The historical column follows USD/JPY, using daily Federal Reserve exchange-rate observations distributed through FRED as DEXJPUS. The exercise uses 96 complete or selected calendar months from 2017-01 through 2024-12. The fixed research window runs from January 2017 through December 2024. Rates are publicly observed. Retail fees, contract terms and decision rules in the examples are assumptions.

Divide the invoice in yen by the contemporaneous yen-per-dollar reference. The dollar price of the same quantity can change when either country’s policy expectations move. Nothing in the arithmetic assumes the Fed’s total assets or reserve balances drive the rate.

The modeled output is measured in US dollars. Its sign and units are essential when you work backward from a chart. Write the initial reference, the final reference, their dates and the modeled result in separate columns. Calculate with unrounded inputs, then round for display. A dollar-cent display may conceal small changes in the reference, especially where the model floors an outcome at zero. A weekly decline in total assets and a change in the translated cost can coexist for many reasons. A narrower funding-market study requires data on the balance-sheet component and actual dollar funding rates around settlement.

Across these months the exchange quote rose 47 times, fell 48 times and was unchanged 1 times. The median absolute endpoint change was 1.38%. The 25th and 75th percentiles of signed endpoint returns, using the sorted observation at the floor of the fractional rank on a zero-based scale, were -1.14% and 1.91%. These are descriptive sample figures. A frequency of rising quotes is neither a rate of profitable trades nor the chance of a favorable outcome on a future payment.

The largest rise | 2022-04

Start with the largest signed monthly rise in this sample. At the last observation of 2022-03, the quote was 121.440. At 2022-04-29 it was 129.840, a signed change of 6.92%. Applying the exposure rule to the earlier price gives 38,274.04 US dollars. Applying it to the later price gives 35,797.91 US dollars. The resulting difference is -2,476.14 US dollars. The quote and the modeled amount need not move in the same direction. Read the numerator and denominator of the actual contract before interpreting a rising candle as a higher bill.

For a plausible contemporaneous alternative, use the first observation of the new month, 122.600 on 2022-04-01, as the budget anchor. The modeled change to the same ending date becomes -2,114.00 US dollars. You have altered only the starting timestamp. That small adjustment matters when the prior month ended before a weekend or holiday. A report that chooses the better anchor afterward mixes performance evaluation with hindsight. Use the timestamp of the actual invoice, forecast or mandate if one exists.

Link the weekly snapshots with a balancing item

The last available weekly total assets observations for 2022-03-30 and 2022-04-27 were USD 8,937.142 billion and USD 8,939.199 billion. Their signed difference is USD 2.057 billion. The measurement uses the Wednesday-level WALCL series, not an average of the month’s weeks. This total includes items besides securities. Labeling the entire difference as Treasury or mortgage runoff would assign a cause that the series cannot establish.

Impose an illustrative USD -15 billion signed change in other liabilities and capital combined. From the balance-sheet identity the associated change in reserves would be USD 17.057 billion. If the assumed other liabilities rose an additional USD 10 billion, implied reserve growth would fall by precisely USD 10 billion. This part is a balancing exercise. Retrieve actual reserve balances and the other H.4.1 lines to replace the assumption before reporting a historical reserve change.

The largest decline | 2024-07

The steepest fall among the remaining months went from 160.880 to 150.380. The cross rate expressed in reverse would instead move from 0.00622 to 0.00665. Its signed percentage change is 6.98%, whereas the original quote changed by -6.53%. Reversing a percentage sign is only an approximation. Arithmetic returns in reciprocal currencies have different denominators. Use full-precision quotes when translating a dollar budget in either direction.

In the working example the unchanged exposure generated a modeled 30,908.37 US dollars at the end, against an original 28,891.10 US dollars. At the arithmetic midpoint of the two observed rates the modeled result would be 29,865.71 US dollars. By comparison the mean of the two modeled endpoint amounts is 29,899.73 US dollars. A linear conversion can make these coincide; a reciprocal or an option-like floor can separate them. This is a practical unit test for a budget spreadsheet built by averaging rates before conversion.

The quietest finish | 2020-06

A quiet monthly finish concealed a sequence of 22 observations. The first budget mark and last mark differed by only 0.00% in absolute quote terms. Between them, the lowest sampled exchange rate was 106.440 on 2020-06-23 and the highest was 109.680 on 2020-06-05. Their ordering is observable at the daily reference frequency. It says nothing about whether a stop or limit was touched first inside an individual trading day.

Including the original budget mark, the lowest and highest modeled amounts were 42,377.83 US dollars and 43,667.79 US dollars. A payment falling due inside this interval would use the terms available on its own date. Picking the cheapest date after seeing the month grants an option the original decision maker may never have had. Ask whether the underlying agreement allowed early settlement, whether the counterparty would accept it, and what financing or fees would accompany changing the date. The end-to-end change alone leaves out all three questions.

The widest daily range | 2020-03

For the widest remaining sample range, compare the recorded low of 102.520 with the high of 111.440. That spread is 8.25% of the previous month-end reference. If the exposure were valued separately at those two observations, the modeled amounts would be 45,337.50 US dollars and 41,708.54 US dollars. A full intraday high-low range could be larger than this sampled range. Treat these as independent reference valuations, not executable bid and offer quotes sitting on a dealer screen at the same instant.

Now place an administrative review on the middle observation, 2020-03-16. At 105.640 the modeled amount was 43,998.49 US dollars. The remaining change to the final observation was -773.34 US dollars. This separates currency exposure before and after the person responsible could act. It is especially useful when a headline about the full month sounds alarming but the invoice was approved or the policy decision released much later. Use an actual approval timestamp for a real case.

Link the weekly snapshots with a balancing item

The last available weekly total assets observations for 2020-02-26 and 2020-03-25 were USD 4,158.637 billion and USD 5,254.278 billion. Their signed difference is USD 1,095.641 billion. The measurement uses the Wednesday-level WALCL series, not an average of the month’s weeks. This total includes items besides securities. Labeling the entire difference as Treasury or mortgage runoff would assign a cause that the series cannot establish.

Impose an illustrative USD 17 billion signed change in other liabilities and capital combined. From the balance-sheet identity the associated change in reserves would be USD 1,078.641 billion. If the assumed other liabilities rose an additional USD 10 billion, implied reserve growth would fall by precisely USD 10 billion. This part is a balancing exercise. Retrieve actual reserve balances and the other H.4.1 lines to replace the assumption before reporting a historical reserve change.

The busiest path | 2022-12

This month has the greatest remaining sum of absolute observation-to-observation log changes: 19.38 percentage points. Its net endpoint change in ordinary percentage terms was -5.38%. Offsetting moves can make the first number large while leaving the second small. Absolute log movement is a path statistic. It includes both directions and must never be booked as the return someone earned by holding the exposure. Collecting every move would require knowing each reversal and being able to execute it after costs.

The annualized sample standard deviation calculated from the 21 log changes was 22.23%, using 252 observations per trading year. The mean level across this month’s daily exchange marks was 134.915, whereas its median was 135.140. Valuing the exposure at those two central rates gives 34,451.38 US dollars and 34,393.96 US dollars. The mean answers a question about evenly weighted observations and the median is the middle observation or average middle pair. Neither is necessarily the benchmark named in a specific agreement.

A month near the middle | 2023-10

Extreme outcomes attract attention. This case was chosen for being close to the median absolute monthly move among the remaining candidates, with an absolute change of 1.36%. Its 21 daily marks give an average modeled value of 31,071.46 US dollars. Valuing the same exposure only at the last mark produces 30,687.97 US dollars. The final observation therefore differs from the typical daily valuation measured by that average by a signed -383.49 US dollars. The difference can go either way; the average has no privileged status for a one-date settlement.

Suppose a report allowed one person to choose a monthly average and another to use the last observation. They could disagree by that amount while applying identical formulas and the same published data. Settle the disagreement by identifying the decision the figure supports. A month of evenly spread expenses calls for transaction weights or a stated approximation. A single invoice calls for its conversion date. A policy announcement needs a timed quote. The right measurement rule follows from the economic event and should be fixed before examining the answer.

Link the weekly snapshots with a balancing item

The last available weekly total assets observations for 2023-09-27 and 2023-10-25 were USD 8,002.064 billion and USD 7,907.830 billion. Their signed difference is USD -94.234 billion. The measurement uses the Wednesday-level WALCL series, not an average of the month’s weeks. This total includes items besides securities. Labeling the entire difference as Treasury or mortgage runoff would assign a cause that the series cannot establish.

Impose an illustrative USD 51 billion signed change in other liabilities and capital combined. From the balance-sheet identity the associated change in reserves would be USD -145.234 billion. If the assumed other liabilities rose an additional USD 10 billion, implied reserve growth would fall by precisely USD 10 billion. This part is a balancing exercise. Retrieve actual reserve balances and the other H.4.1 lines to replace the assumption before reporting a historical reserve change.

A sharp single step | 2022-11

The largest remaining single-step move was from 146.150 on 2022-11-09 to 141.780 on 2022-11-10. In ordinary percentage terms that is -2.99%. Against the exposure rule, this step changes the modeled amount by 980.24 US dollars. Because the observations can be separated by nonbusiness days, describe this as an observation-to-observation change. The file cannot reveal which hour accounted for it or whether liquidity was available at either published level.

To check a claim about that step, identify the calendar of releases and local holidays for both currencies, then match each event with its precise published time. Look for an independent price series that actually traded in the proposed interval. A gap between daily reference marks invites a hypothesis; following the sequence of executable quotes is how you could test it. Changing an amount solely because of the two marks is an accounting exercise. Attributing the difference to a particular trader, institution or economic release needs additional evidence.

A split-date budget | 2022-03

Take the average modeled valuation on the first five available dates, 40,334.42 US dollars, and on the last five available dates, 38,021.43 US dollars. Their difference is -2,312.98 US dollars. This case has the largest remaining absolute gap by that definition. A plan fixed in advance to value half the exposure against each group would give a weighted amount of 39,177.93 US dollars if the exposure and its charges can be divided proportionately. The result is a comparison of timing policies. It does not reveal an optimal execution date.

A fixed charge can break that proportional assumption if two actual transactions each incur the full charge. An indivisible futures contract or option can prevent splitting the quantity at all. An announcement time cannot be averaged across ten dates. Before applying the exercise, identify the specific constraint: Treasury balances, reverse repos and other liability changes. Then identify which of the two windows an actual mandate would allow. Where the timing choice is unavailable, read the two amounts as a sensitivity range instead of an executable plan.

Why a decline in reverse repos can matter for reserves

As a bookkeeping exercise, hold total assets and capital constant and reduce a reverse repo liability by $12 billion. If no other liability changes except reserve balances, reserves must increase $12 billion. Now add a $9 billion decline in total assets. Under the same assumptions the reserve gain becomes $3 billion. The numbers are invented to show why the liability side can absorb some of the effect that is sometimes attributed mechanically to QT.

An actual release has more than those two moving parts. Treasury’s General Account, currency in circulation, foreign official deposits and other entries can change at the same time. Use the complete balance sheet to explain any residual. A reader evaluating the policy stance should pair that arithmetic with the implementation note and current policy regime. A small weekly increase in assets is a measurement result and does not by itself announce a new purchase program.

How to identify a usable USD/JPY setup on a chart

Open a five-minute USD/JPY replay and hide everything to the right of your intended decision. This is a hypothetical short case, constructed to show how Treasury balances, reverse repos and other liability changes fits into an execution record. Use New York local time for the chart, retaining the actual offset for the date. For a restricted or offshore currency, this drawing describes a price sequence, not the availability of a US retail contract. An economic reference period, a release time and a bar timestamp belong in separate fields.

In the first diagram the reference high is 151.531. The candle that trades above it reaches 151.604 before price returns into the range. To describe that move as a sweep in ICT terminology, mark the existing level first. If the level only becomes obvious after the reversal, save the observation as a retrospective pattern. An increase in volume or a financial headline cannot reveal whose orders traded there.

The displaced leg reaches beyond 151.351. Read bars five, six and seven as a sequence. A gap between the outer wicks brackets the interval from 151.364 to 151.408. It is a chart definition of an imbalance. Price could fill that interval, pass straight through it or never revisit it. Only the first return that satisfies a rule written in advance can provide a candidate entry for this particular exercise.

How to trade the idea on demo and define failure

For the rehearsal, a sell at 151.387 follows a return into the marked interval. If the order fails to fill, record an unfilled setup and retain the original timestamp. Moving the order toward the next favorable candle after that candle forms changes the test. Keep the macro or account question separate: The independent document check concerns Treasury balances, reverse repos and other liability changes. A convincing answer about that issue cannot make a missed entry executable.

The protective exit is 151.645, beyond the sweep, and the planned target is 150.871. For this short idea the favorable direction in the pair is lower. A live position can experience partial fills, a spread jump or a gap past the stop. For this drawing only we assume all levels execute exactly as stated. The second chart displays one favorable sequence so that the accounting can be checked. It supplies no evidence about how often such a sequence occurs.

Set an expiry for the entry and cancel the study if the invalidation trades before entry or if the spread makes the gross plan infeasible. If a coarse chart touches stop and target within one bar, lower-frequency OHLC cannot settle the order of events. Use properly timestamped executable quotes if available. Otherwise keep a separate ambiguous sample, and state that rule before counting wins. For this rehearsal the underlying document remains the source for the change in reserves implied by a balanced ledger.

Worked example in a US dollar account

Calculate the distance before calculating the money. The entry-to-stop distance is |151.387 minus 151.645| = 0.258, or 25.8 pips. The target distance is |150.871 minus 151.387| = 0.516, or 51.6 pips. Keeping absolute distances here avoids giving a short setup a negative risk amount. The sign of the actual result belongs on the fill record after closing.

At 23,000 base units, the distance to the stop represents 5,934.00 JPY and the distance to the target represents 11,868.00 JPY. Convert to a USD result separately at each assumed exit by dividing by 151.645 or 150.871 JPY per US dollar. The conversion is slightly different on the two outcomes. Gross reward divided by gross risk is USD 78.66 / USD 39.13 = 2.01. Written as risk:reward this is 1:2.01.

Assume a round-trip cost of USD 2.90 on either completed outcome. We use executable entry and exit prices, so a spread already inside those prices must not be charged again. The simplified net reward is USD 78.66 minus USD 2.90 = USD 75.76. Net planned loss is USD 39.13 plus USD 2.90 = USD 42.03. Net R:R is 1:1.80 because USD 75.76 / USD 42.03 = 1.80. Financing, slippage and tax are excluded from this short rehearsal.

Check the losing branch with the same care as the winning one. If the exit slips by one additional pip against you, the quote-currency impact on this size is 230.00 JPY before any conversion. A plan denominated in dollars should reserve for an adverse fill or skip a setup that cannot fit the intended budget. Keep the original quote and the executed price alongside the size and currency of the exposure.

To annotate the rehearsal, mark the existing swing first, then identify displacement and the interval between the outer candle wicks. For definitions and session context, see the school’s lessons on fair value gaps, order blocks, the London session, market structure shifts, and the Judas swing.

Sources and calculation notes

Federal Reserve policy normalization history. Federal Reserve H.4.1 statistical release.

Historical series and original records. DEXJPUS observations and units · WALCL observations and units. The calculations use observations available when checked on November 14, 2026. Subsequent revisions can change retrospective comparisons.

Frequently asked questions

To take the release from the data table to a dated trading record, use the news-release filter, bond-market fundamentals, liquidity sweeps, and hindsight-free replay.