01 / The AI investment cycle
Why AI can raise costs before it lowers them.
Who pays for AI before the productivity gains arrive?
He separates the cost of building AI from its potential long-run benefits.
“There's a technology that has a promise of being deflationary in the long run, but the technology is being built and, as we already discussed, building is so expensive [...] that demand is creating incredible pressure on DRAM. [...] At the moment we're just experiencing those inflationary effects.”
Building comes before saving.
AI may eventually help workers produce more with less. First, someone must build data centres, connect power, and buy chips. If those orders arrive faster than factories and grids can expand, scarce inputs get dearer. Falling software costs can coexist with rising construction and electricity costs.
Capital expenditure, or capex, is this upfront investment. Companies can use operating cash and cash reserves; once that money is committed, extra spending needs outside funding, reduced payouts, or a smaller plan. Even a profitable technology company can become a large borrower.
More capacity is not yet more productivity.
New equipment only pays off when it is used well. Skills, business processes, and complementary infrastructure take time to develop. The productivity J-curve describes that delay; it does not date an AI payoff. The immediate question is whether financing and physical capacity can keep up while the eventual benefits remain uncertain.
Evidence & context [6]
Two clocks, one investment cycle
The bill arrives before the benefit.
Competition for scarce inputs and capital
Productivity can ease pressure, but its timing and size are uncertain.
Lab 01 Change one thing
When does more investment mean more borrowing?
Change investment spending. The gap is the cash the company still needs to find.
Each action starts from the Original example.
One hypothetical year; assumes every funding gap is borrowed, not funded by issuing shares.
Model details & full chart Equations, data, and sharing
Model output / hypothetical
Financing ($bn)Chart values Accessible data table
| Hypothetical annual capex ($bn) | Net new financing | Gross issuance |
|---|---|---|
| 0 | 0 | 50 |
| 50 | 0 | 50 |
| 100 | 0 | 50 |
| 150 | 0 | 50 |
| 160 | 0 | 50 |
| 200 | 40 | 90 |
| 250 | 90 | 140 |
| 300 | 140 | 190 |
| 350 | 190 | 240 |
| 400 | 240 | 290 |
The spending plan exceeds available cash by $40bn. Refinancing adds $50bn to gross issuance without funding extra investment.
Equation and limits
Net financing = max(0, capex + distributions - operating cash - cash draw). Gross issuance = net financing + refinancing.
- All quantities are hypothetical billions of U.S. dollars over one year. Cash draw uses existing reserves.
- The gap is entirely debt-financed. No equity issuance, asset sales, changed payouts, interest feedback, or deferred capex.
- The threshold creates a kink, not a marginal multiplier above one. This is not a hyperscaler issuance forecast.
The link contains this lab’s inputs only. No responses are saved in this browser. Chart CSV values are unrounded; displayed labels are rounded.
Evidence, limits, and deeper reasoning For the curious and the technically minded
Mechanism in one sentence: AI can raise demand for scarce materials and financing before it delivers cheaper production.
Keep the funding arithmetic honest.
Use operating cash before capex, not free cash flow after it. Subtracting capex twice invents a financing need. The lab holds other funding choices fixed: after internal resources are exhausted, one extra dollar of investment needs one extra dollar externally. A large percentage increase in borrowing may simply reflect a small starting amount.
Sources [1]
Separate evidence from scale claims.
The IEA estimates cover all data centres, not AI alone; their local grid effects need not match their global electricity share. Nordvig also compares hyperscaler and Treasury long-end issuance. This guide does not independently verify that comparison: it is a flow claim, not equal debt stocks. A test needs defined issuers, dates, currencies, maturities, and gross-versus-net issuance. Yields alone cannot establish it.
What to watch
As AI investment expands, do long borrowing costs and real yields rise alongside input prices, or does new capacity arrive quickly enough to absorb the extra demand?
Pair these series with company cash-flow statements and electricity evidence. Economy-wide yields and PCE prices reflect many forces; neither isolates AI investment or measures the local price of grid access.
What would change the story?
The near-term pressure thesis weakens if sustained investment meets rapidly expanding input supply and financing stays readily available. Rising share prices alone would not demonstrate the promised productivity gain.
Check your understanding One question, with an explanation
02 / The price of duration
Why long yields can rise when the Fed cuts.
Why can long yields rise while policy rates fall?
He links pressure on long yields to competition for lenders' money.
“All these years where we were used to, okay, we have a low inflation environment, right, and if you have a little bit of carry in your credit instruments you can always sell it, right—those days are just gone. Now we have such competition for capital that yields are being pushed higher.”
The Fed does not set every yield.
The Fed steers overnight borrowing costs. A 30-year Treasury yield reflects decades of expected short rates plus compensation for risks over that period. Stronger demand for capital or greater uncertainty can raise long yields even while the current policy rate falls.
Compare an ordinary Treasury's nominal yield with the real, inflation-adjusted yield on an inflation-protected Treasury, or TIPS, at the same maturity. The difference, called a breakeven, mixes expected Consumer Price Index (CPI) inflation with inflation-risk compensation and liquidity effects. It is not pure inflation expectations.
Evidence & context [7]
Fixed payments, changing prices.
An older bond promises fixed payments. If new bonds offer higher yields, buyers pay less for the old promise. Duration means rate sensitivity: longer, lower-coupon bonds usually fall more when yields rise. Stocks also discount future cash flows, but stronger expected profits can offset a higher discount rate. The cause and speed of the yield move matter.
Lab 02 Change one thing
Why might a bond yield rise?
Raise the real rate or expected inflation. Both can raise the headline nominal yield.
Each action starts from the Original example.
Assumed components at one maturity, not a live estimate of inflation expectations.
Model details & full chart Equations, data, and sharing
Model output / hypothetical
Percentage points- 1TIPS real
- 2Expected CPI
- 3Inflation risk
- 4Liquidity (-)
- 5Nominal
Chart values Accessible data table
| Components, then their sum | Yield components and total |
|---|---|
| TIPS real | 2 |
| Expected CPI | 2.3 |
| Inflation risk | 0.3 |
| Liquidity (-) | -0.1 |
| Nominal | 4.5 |
The 2.50% breakeven differs from 2.30% assumed expected inflation because risk compensation and liquidity also enter the spread.
Equation and limits
Breakeven = expected CPI inflation + inflation-risk compensation - TIPS liquidity premium. Nominal yield = TIPS real yield + breakeven.
- All inputs describe the same hypothetical maturity and instant. Simple additive yield decomposition.
- The risk and liquidity terms are assumptions, not independently observed live components. No second term premium is added on top.
- CPI-linked breakevens are not the Fed's PCE inflation target.
The link contains this lab’s inputs only. No responses are saved in this browser. Chart CSV values are unrounded; displayed labels are rounded.
Lab 03 Change one thing
What happens to a bond when rates change?
Change the market yield, then try a shorter bond. The promised payments stay fixed.
Each action starts from the Original example.
Immediate price change on $100 face value—not an investment return with earned interest.
Model details & full chart Equations, data, and sharing
Model output / hypothetical
Bond price change (%)Chart values Accessible data table
| Instantaneous yield change (basis points) | Exact discounting | Duration + convexity |
|---|---|---|
| -200 | 42.0346 | 40.4412 |
| -190 | 39.4092 | 38.0537 |
| -180 | 36.8483 | 35.7047 |
| -170 | 34.3501 | 33.3942 |
| -160 | 31.913 | 31.1222 |
| -150 | 29.5352 | 28.8886 |
| -140 | 27.2152 | 26.6934 |
| -130 | 24.9513 | 24.5368 |
| -120 | 22.7422 | 22.4186 |
| -110 | 20.5862 | 20.3388 |
| -100 | 18.482 | 18.2975 |
| -90 | 16.4282 | 16.2947 |
| -80 | 14.4234 | 14.3303 |
| -70 | 12.4663 | 12.4044 |
| -60 | 10.5557 | 10.517 |
| -50 | 8.6902 | 8.668 |
| -40 | 6.8688 | 6.8575 |
| -30 | 5.0901 | 5.0854 |
| -20 | 3.3532 | 3.3518 |
| -10 | 1.6569 | 1.6567 |
| 0 | 0 | 0 |
| 10 | -1.6184 | -1.6182 |
| 20 | -3.1993 | -3.198 |
| 30 | -4.7438 | -4.7393 |
| 40 | -6.2528 | -6.2421 |
| 50 | -7.7272 | -7.7065 |
| 60 | -9.1679 | -9.1324 |
| 70 | -10.5758 | -10.5198 |
| 80 | -11.9518 | -11.8688 |
| 90 | -13.2967 | -13.1794 |
| 100 | -14.6113 | -14.4514 |
| 110 | -15.8964 | -15.685 |
| 120 | -17.1528 | -16.8802 |
| 130 | -18.3811 | -18.0369 |
| 140 | -19.5822 | -19.1551 |
| 150 | -20.7567 | -20.2348 |
| 160 | -21.9053 | -21.2761 |
| 170 | -23.0286 | -22.279 |
| 180 | -24.1274 | -23.2434 |
| 190 | -25.2022 | -24.1693 |
| 200 | -26.2537 | -25.0567 |
A 35bp yield change moves this 30-year bond by -5.50%. Duration plus convexity estimates -5.50%. Neither number includes a holding period or earned coupon income.
Equation and limits
Price = sum of discounted semiannual coupons + discounted principal. Approximate change = -duration x yield change + 1/2 x convexity x yield change squared.
- Face value $100; fixed semiannual coupons; nominal annual yield with semiannual compounding. No default or embedded options.
- The shift is parallel and instantaneous. This is a price change, not a holding-period return; coupon income and the passage of time are excluded.
- A 35bp shift is 0.0035 in the discounting formula. Imported constant-maturity Treasury yields are reference inputs, not a tradable bond quote.
The link contains this lab’s inputs only. No responses are saved in this browser. Chart CSV values are unrounded; displayed labels are rounded.
Evidence, limits, and deeper reasoning For the curious and the technically minded
Mechanism in one sentence: The Fed steers overnight rates; long borrowing costs also depend on future policy, investor demand, and compensation for risk.
Put a number on rate sensitivity.
A basis point is one hundredth of a percentage point. A 35bp rise is 0.35 percentage points; modified duration of 15 implies roughly a 5.25% price decline. Convexity describes the curvature missing from that approximation. The lab also prices every cash flow. Both are instantaneous price changes under a parallel yield shift, not annual total returns.
Do not mistake a spread for an explanation.
Breakevens are not a poll of investors or the Fed's PCE inflation target. The watchlist's five-year breakeven and ten-year TIPS yield cannot be added to reconstruct a ten-year nominal yield. Nor does their movement identify issuance as the cause. Growth news, inflation risk, liquidity, and changing investor demand remain competing explanations; no universal 5% threshold determines when equities must fall.
What to watch
When long yields rise, is the move accompanied by higher real yields, inflation compensation, or short rates, and does its speed suggest pressure beyond today's overnight policy setting?
Use matched maturities and observation windows before decomposing changes. The five-year breakeven is not compatible with a ten-year real yield, and neither series directly identifies a cause or a term premium.
What would change the story?
An issuance-pressure explanation weakens if new borrowing is readily absorbed or changes in expected growth and inflation better explain the rise. Co-occurrence of bond sales and higher yields is not sufficient.
Check your understanding One question, with an explanation
03 / The sovereign balance sheet
A familiar interest rate on a much bigger debt.
Why does 5% mean something different when debt is much larger?
He compares sustained high yields against very different amounts of outstanding debt.
“The bottom line is it's a long time ago since we really been at these levels in a sustained way. If we look at the last time we were at this type of level, which was before the global financial crisis, the debt levels are totally different. So what we could handle back then with that level of yields, it's hard to imagine we can handle it now.”
Debt makes the interest bill larger.
A higher rate costs more when it applies to a larger debt balance. But today's Treasury yield does not instantly become the government's average borrowing cost: existing fixed-rate bonds retain their coupons until they mature. Refinancing gradually brings new rates into the interest bill.
The primary deficit is spending excluding interest minus revenue. Interest adds another financing need; a primary surplus can offset it. This distinction stops us from counting interest twice or assuming that higher rates alone determine the budget.
Evidence & context [9]
GDP growth changes the burden.
Debt is often compared with GDP, the economy's annual output. Nominal GDP growth enlarges that denominator. If it outpaces the average interest rate, it eases pressure from inherited debt; a large primary deficit can still raise the ratio. The useful question is how borrowing, growth, and refinancing interact, not which round-number yield guarantees a crisis.
A ratio has two moving parts
Interest adds to the bill. Growth changes its weight.
Lab 04 Change one thing
Can growth keep up with government debt?
Change new borrowing costs or economic growth. Follow debt relative to the size of the economy.
Each action starts from the Original example.
An accounting scenario, not a forecast or a measure of government profit.
Model details & full chart Equations, data, and sharing
Model output / hypothetical
Debt / GDP (%)Chart values Accessible data table
| Years from hypothetical starting point | Selected funding path | Starting rate held fixed |
|---|---|---|
| 0 | 100 | 100 |
| 1 | 102.1154 | 101.8269 |
| 2 | 104.4583 | 103.6415 |
| 3 | 106.9904 | 105.444 |
| 4 | 109.6814 | 107.2342 |
| 5 | 112.5067 | 109.0125 |
| 6 | 115.4468 | 110.7787 |
| 7 | 118.4856 | 112.5331 |
| 8 | 121.6103 | 114.2757 |
| 9 | 124.8104 | 116.0065 |
| 10 | 128.0772 | 117.7257 |
| 11 | 131.4037 | 119.4333 |
| 12 | 134.7842 | 121.1294 |
| 13 | 138.2142 | 122.8141 |
| 14 | 141.6897 | 124.4875 |
| 15 | 145.2077 | 126.1496 |
| 16 | 148.7657 | 127.8005 |
| 17 | 152.3618 | 129.4403 |
| 18 | 155.9942 | 131.0691 |
| 19 | 159.6617 | 132.6869 |
| 20 | 163.3633 | 134.2938 |
| 21 | 167.0983 | 135.8899 |
| 22 | 170.8658 | 137.4753 |
| 23 | 174.6657 | 139.05 |
| 24 | 178.4973 | 140.614 |
| 25 | 182.3607 | 142.1676 |
| 26 | 186.2555 | 143.7107 |
| 27 | 190.1817 | 145.2434 |
| 28 | 194.1394 | 146.7658 |
| 29 | 198.1284 | 148.278 |
| 30 | 202.1489 | 149.78 |
At the selected settings, debt moves from 100% to 202.1% of GDP. The comparison ends at 149.8% with the average rate held at its starting level.
Equation and limits
d(t) = [(1 + r(t)) / (1 + g)] x d(t-1) + p. r(t) = r(t-1) + rollover x [market rate - r(t-1)].
- Thirty annual periods; d is debt/GDP, r is the nominal effective rate, g is nominal GDP growth, and p is the primary deficit/current GDP. Rates enter the equation as decimals.
- No fiscal response, currency revaluation, inflation-linked principal adjustment, or maturity-level issuance simulation. The average-rate adjustment is stylized.
- If a scenario pays down all liabilities and the path goes negative, it becomes a net-financial-asset extension of the accounting, not literal negative gross public debt. This is not CBO's baseline.
The link contains this lab’s inputs only. No responses are saved in this browser. Chart CSV values are unrounded; displayed labels are rounded.
Evidence, limits, and deeper reasoning For the curious and the technically minded
Mechanism in one sentence: A larger debt load magnifies interest costs, but refinancing speed, nominal growth, and the primary balance determine how quickly that pressure builds.
Specify the balance sheet and the baseline.
Debt held by the public excludes intragovernmental holdings; gross federal debt includes them. Cross-country comparisons also require consistent government coverage, currencies, maturities, and public assets. CRFB reports a CBO baseline of public debt reaching 175% of GDP in 2056. The original CBO tables could not be retrieved during the September 14 review; this is a secondary report, not a reproduction of CBO data.
The refinancing path is an assumption.
The lab moves a chosen share of the effective interest rate toward a new market rate each year. It is not a security-by-security maturity schedule. Bills, floating-rate debt, and inflation-linked debt behave differently. Investor demand, taxes, spending, and growth can interrupt an adverse feedback; the dollar's reserve role neither removes the arithmetic nor fixes the timing of market reactions.
What to watch
As new Treasury financing becomes more expensive, how quickly will maturing debt pick up those rates, and can nominal growth or a stronger primary balance offset the added interest?
Combine market yields with Treasury debt composition and maturity data and a dated budget baseline. One long yield is neither the government's average funding rate nor a complete debt-sustainability measure.
What would change the story?
The adverse path weakens if the primary balance improves, nominal growth outpaces effective interest costs, or investors absorb refinancing without demanding persistently higher yields. Debt levels alone do not determine crisis timing.
Check your understanding One question, with an explanation
04 / Currencies and policy
What policy can change, and what it cannot erase.
Can policymakers lower yields, weaken the dollar, and restrain inflation at once?
Discussing Japan's intervention funding, he returns to the weak-dollar and inflation conflict.
“There's also been talk about them being able to use [...] lending facilities more aggressively so they can get the funding to do their [...] intervention without having to sell Treasuries. [...] It's pretty interesting that we have this kind of currency dimension where the administration is willing to endorse currency weakness and even actively pursue it while we have inflation problems at the same time.”
A Treasury buyback is not a money-printing machine.
Treasury pays for buybacks from its account at the Fed, the Treasury General Account (TGA). It retires selected bonds to support trading liquidity or manage cash. Replenishing the TGA with new debt transfers funding and maturity exposure; it does not cancel the financing need or create net reserves across the full cycle.
Spending down the TGA temporarily adds bank reserves, but that is not QE. Fed quantitative easing (QE) is a central-bank purchase financed by newly created reserves: it is money creation. If the seller is a nonbank, it receives a bank deposit; reserves are credited to its bank, not handed straight to the public. The seller exchanges an asset for money, not free wealth. More reserves guarantee neither more lending nor a jump in consumer prices.
A weaker currency has two sides.
Buying yen can support its price; stronger Japanese rate incentives can reinforce that move. Japan's Ministry of Finance decides intervention; the Bank of Japan acts as agent. A weaker dollar can help some U.S. exporters but also raise import prices.
For an investor borrowing yen to hold dollars, carry is the interest spread before currency moves. A stronger yen makes repayment costlier and can overwhelm that spread. Currency hedging is not a free way to keep the higher yield.
Follow one hypothetical $100 Treasury
Same bond. Different balance sheets.
Treasury buybacks use cash. Federal Reserve QE creates reserve balances.
Hypothetical $100 face value, bought at par ($100). No interest, fees, or price gains.
Treasury buyback
Start
Treasury buyback Start
A bond is an asset to its holder—and debt to its issuer.
The nonbank public holds an existing Treasury security. Treasury already has enough cash in its Treasury General Account (TGA) at the Fed. Zero means no change yet, not an empty account.
| Position | Before step | After step |
|---|---|---|
| Market-held Treasury securities | $0 | $0 |
| Bank reserve balances | $0 | $0 |
| Nonbank public’s bank deposits | $0 | $0 |
| Treasury cash (TGA) | $0 | $0 |
| Fed-held Treasury securities | $0 | $0 |
Inspect the balanced T-accounts
Federal Reserve
- Assets
- $0
- Liabilities
- $0
- Asset Treasury securities
- $0
- Liability Bank reserves
- $0
- Liability Treasury’s TGA
- $0
U.S. Treasury
- Assets
- $0
- Liabilities
- $0
- Asset Cash at the Fed
- $0
- Liability Old debt owed
- $0
- Liability New debt owed
- $0
Commercial bank
- Assets
- $0
- Liabilities
- $0
- Asset Reserves at the Fed
- $0
- Liability Deposits owed
- $0
Nonbank public
- Assets
- $0
- Liabilities
- $0
- Asset Old Treasury
- $0
- Asset New Treasury
- $0
- Asset Bank deposits
- $0
Changes since start; A = asset, L = liability. Each balance sheet reconciles.
Asset purchase
Treasury buyback Asset purchase
Treasury pays from its account. The old bond is retired.
Treasury cash falls; bank reserves and the seller’s deposit rise. At the Fed, one liability replaces another. Its total assets and liabilities do not grow. This buyback is not QE or a money-printing machine.
-
Deliver the existing Treasury$100
Nonbank seller to Treasury → retired -
Transfer Treasury cash$100
Treasury’s TGA at the Fed to Bank’s Fed account -
Credit the seller’s deposit$100
Bank’s deposit ledger to Nonbank seller
| Position | Before step | After step |
|---|---|---|
| Market-held Treasury securities | $0 | −$100 |
| Bank reserve balances | $0 | +$100 |
| Nonbank public’s bank deposits | $0 | +$100 |
| Treasury cash (TGA) | $0 | −$100 |
| Fed-held Treasury securities | $0 | $0 |
Inspect the balanced T-accounts
Federal Reserve
- Assets
- $0
- Liabilities
- $0
- Asset Treasury securities
- $0
- Liability Bank reserves
- +$100
- Liability Treasury’s TGA
- −$100
U.S. Treasury
- Assets
- −$100
- Liabilities
- −$100
- Asset Cash at the Fed
- −$100
- Liability Old debt owed
- −$100
- Liability New debt owed
- $0
Commercial bank
- Assets
- +$100
- Liabilities
- +$100
- Asset Reserves at the Fed
- +$100
- Liability Deposits owed
- +$100
Nonbank public
- Assets
- $0
- Liabilities
- $0
- Asset Old Treasury
- −$100
- Asset New Treasury
- $0
- Asset Bank deposits
- +$100
Changes since start; A = asset, L = liability. Each balance sheet reconciles.
- Deliver the existing Treasury
- The seller gives up an asset; Treasury extinguishes its matching debt.
- Transfer Treasury cash
- TGA down, bank reserves up. Total Fed liabilities do not grow.
- Credit the seller’s deposit
- The bank owes a new deposit; the seller holds it. This is not a transfer of reserves to the seller.
Funding cycle (optional, separate issuance)
Treasury buyback Funding cycle
Separate new borrowing reverses the cash and reserve changes.
Now a nonbank investor buys a new $100 Treasury at par. Deposits and reserves fall; the TGA refills. The equal-cash cycle replaces old debt with new debt, potentially changing maturity and liquidity—not creating free funding.
-
Issue a new Treasury$100
Treasury to Nonbank investor -
Debit the investor’s deposit$100
Nonbank investor to Bank’s deposit ledger -
Refill Treasury cash$100
Bank’s Fed account to Treasury’s TGA at the Fed
| Position | Before step | After step |
|---|---|---|
| Market-held Treasury securities | −$100 | $0 |
| Bank reserve balances | +$100 | $0 |
| Nonbank public’s bank deposits | +$100 | $0 |
| Treasury cash (TGA) | −$100 | $0 |
| Fed-held Treasury securities | $0 | $0 |
Inspect the balanced T-accounts
Federal Reserve
- Assets
- $0
- Liabilities
- $0
- Asset Treasury securities
- $0
- Liability Bank reserves
- $0
- Liability Treasury’s TGA
- $0
U.S. Treasury
- Assets
- $0
- Liabilities
- $0
- Asset Cash at the Fed
- $0
- Liability Old debt owed
- −$100
- Liability New debt owed
- +$100
Commercial bank
- Assets
- $0
- Liabilities
- $0
- Asset Reserves at the Fed
- $0
- Liability Deposits owed
- $0
Nonbank public
- Assets
- $0
- Liabilities
- $0
- Asset Old Treasury
- −$100
- Asset New Treasury
- +$100
- Asset Bank deposits
- $0
Changes since start; A = asset, L = liability. Each balance sheet reconciles.
- Issue a new Treasury
- New debt owed by Treasury; a new security held by the investor.
- Debit the investor’s deposit
- The investor’s deposit asset and the bank’s deposit liability both fall.
- Refill Treasury cash
- The Fed debits bank reserves and credits the TGA: its liabilities change composition.
Compare: Federal Reserve QE
Start
Federal Reserve QE Start
The Fed is the buyer—not the fiscal Treasury.
The nonbank public holds an existing Treasury security. A QE purchase exchanges that asset for a bank deposit. The Fed can create reserve balances; Treasury does not receive the purchase proceeds.
| Position | Before step | After step |
|---|---|---|
| Market-held Treasury securities | $0 | $0 |
| Bank reserve balances | $0 | $0 |
| Nonbank public’s bank deposits | $0 | $0 |
| Treasury cash (TGA) | $0 | $0 |
| Fed-held Treasury securities | $0 | $0 |
Inspect the balanced T-accounts
Federal Reserve
- Assets
- $0
- Liabilities
- $0
- Asset Treasury securities
- $0
- Liability Bank reserves
- $0
- Liability Treasury’s TGA
- $0
U.S. Treasury
- Assets
- $0
- Liabilities
- $0
- Asset Cash at the Fed
- $0
- Liability Old debt owed
- $0
- Liability New debt owed
- $0
Commercial bank
- Assets
- $0
- Liabilities
- $0
- Asset Reserves at the Fed
- $0
- Liability Deposits owed
- $0
Nonbank public
- Assets
- $0
- Liabilities
- $0
- Asset Old Treasury
- $0
- Asset New Treasury
- $0
- Asset Bank deposits
- $0
Changes since start; A = asset, L = liability. Each balance sheet reconciles.
Asset purchase
Federal Reserve QE Asset purchase
The Fed acquires the bond and creates reserve balances.
The Fed adds a Treasury asset and a reserve liability. The bank adds reserves and owes the seller a deposit. Treasury gets no new cash. Its existing debt remains outstanding, now held by the Fed.
-
Deliver the existing Treasury$100
Nonbank seller to Fed → still outstanding -
Create reserve balances$100
Fed’s reserve liabilities to Bank’s Fed account -
Credit the seller’s deposit$100
Bank’s deposit ledger to Nonbank seller
| Position | Before step | After step |
|---|---|---|
| Market-held Treasury securities | $0 | −$100 |
| Bank reserve balances | $0 | +$100 |
| Nonbank public’s bank deposits | $0 | +$100 |
| Treasury cash (TGA) | $0 | $0 |
| Fed-held Treasury securities | $0 | +$100 |
Inspect the balanced T-accounts
Federal Reserve
- Assets
- +$100
- Liabilities
- +$100
- Asset Treasury securities
- +$100
- Liability Bank reserves
- +$100
- Liability Treasury’s TGA
- $0
U.S. Treasury
- Assets
- $0
- Liabilities
- $0
- Asset Cash at the Fed
- $0
- Liability Old debt owed
- $0
- Liability New debt owed
- $0
Commercial bank
- Assets
- +$100
- Liabilities
- +$100
- Asset Reserves at the Fed
- +$100
- Liability Deposits owed
- +$100
Nonbank public
- Assets
- $0
- Liabilities
- $0
- Asset Old Treasury
- −$100
- Asset New Treasury
- $0
- Asset Bank deposits
- +$100
Changes since start; A = asset, L = liability. Each balance sheet reconciles.
- Deliver the existing Treasury
- The seller gives up an asset; the Fed acquires it. Treasury still owes the debt.
- Create reserve balances
- A new Fed liability matches the bank’s new reserve asset. The Fed’s balance sheet grows.
- Credit the seller’s deposit
- The bank owes a new deposit; the seller holds it. This is not a transfer of reserves to the seller.
No free financial wealth. At a fair price, the seller swaps a bond for a deposit.
Reserves stay at the Fed. Customers hold bank deposits—not automatic new loans.
“Market-held” excludes the Fed—not the official “debt held by the public” measure. Ledgers show changes, not starting balances.
Assumptions, bank sellers & limits
- Why Treasury buys back debt
- Its stated purposes are liquidity support and cash management, not an announced yield cap. The purchased security is retired. A temporary rise in reserves from spending Treasury cash is not a Fed asset purchase: TGA liabilities fall as reserve liabilities rise.
- The funding step is a separate transaction
- This example pairs the buyback with equal-value new issuance to a nonbank investor, who need not be the original seller. In practice, funding can precede the buyback or use other receipts and existing cash; auctions are not necessarily earmarked one-for-one. Timing and other transactions affect observed reserves.
- Why the par assumption matters
- Both securities have a hypothetical $100 face value and trade for $100 here. Equal cash funding offsets the cash and reserve changes. Equal face-value debt replacement follows only under this par assumption: market values, accrued interest, and face values differ in real buybacks. Maturity and liquidity can change even when the amount of debt does not.
- QE is not new fiscal revenue or debt retirement
- This QE example buys an existing Treasury from a nonbank seller. Treasury gets no new proceeds and still owes the security held by the Fed. Fed holdings are included in the official debt-held-by-the-public measure; our market-held asset metric deliberately excludes them. A fair-price asset swap creates no immediate financial-wealth windfall; subsequent market-price effects are outside the model.
- If the Fed buys a bank’s own inventory instead
- The bank exchanges its Treasury asset for a reserve asset. The Fed’s balance sheet still expands, but a new nonbank deposit need not appear on settlement. The main diagram intentionally uses a nonbank seller throughout.
- There is no mechanical lending or inflation switch
- Banks do not lend reserve balances directly to households. A bank loan creates a loan asset and a deposit liability; reserves help settle payments between banks. Lending depends on borrowers, risk, capital, prices, and other constraints. Neither these settlement entries nor an increase in reserves guarantees more loans or a particular CPI outcome.
Lab 05 Change one thing
Can a stronger yen wipe out the interest earned?
Borrow yen, earn dollar interest, then exchange back. Move the ending exchange rate to see what is left after repayment.
Each action starts from the Original example.
Gain or loss on borrowed yen, not return on your own money. Fees, margin calls, and credit losses are excluded.
Model details & full chart Equations, data, and sharing
Model output / hypothetical
JPY on 1,000,000 borrowed- 1USD interest
- 2FX translation
- 3Yen funding
- 4Net payoff
Chart values Accessible data table
| Contributions to payoff in yen | Unhedged payoff bridge |
|---|---|
| USD interest | 45,000 |
| FX translation | -104,500 |
| Yen funding | -10,000 |
| Net payoff | -69,500 |
Dollar assets return JPY 940,500 at the ending exchange rate, against JPY 1,010,000 owed. A lower USD/JPY hurts this unhedged position; parity hedging removes the free interest differential.
Equation and limits
Yen proceeds = borrowed yen x (1 + dollar rate)^T x ending spot / starting spot. Repayment = borrowed yen x (1 + yen rate)^T.
- Both spots are yen per dollar. A lower ending quote means a stronger yen. Returns are on borrowed notional, not on investor equity.
- Hypothetical deposits and borrowing with effective annual compounding. No credit loss, transaction cost, margin call, taxes, or changing funding rates.
- Parity forward = starting spot x [(1 + yen rate) / (1 + dollar rate)]^T. The frictionless hedge cancels the carry differential; actual forward basis and costs can differ.
The link contains this lab’s inputs only. No responses are saved in this browser. Chart CSV values are unrounded; displayed labels are rounded.
Evidence, limits, and deeper reasoning For the curious and the technically minded
Mechanism in one sentence: Separate Treasury debt management from Fed money creation, then follow the funding, currency exposure, and inflation tradeoffs that remain.
Follow the whole funding cycle.
The reserve-neutral full cycle assumes equal new borrowing and buyback spending, a restored TGA balance, and no other balance-sheet changes. Payment timing and whether the buyer or seller is a bank affect intermediate entries. Debt is reshaped, not costlessly removed. Treasury's FAQ establishes liquidity support and cash management, not an official yield cap; Nordvig's yield-control interpretation is separate.
Hedging and intervention still need financing.
USD/JPY measures yen per dollar: a fall from 150 to 135 means a stronger yen. Under simplified covered interest parity, forward currency pricing offsets the interest-rate advantage; real hedges also involve costs, collateral, and basis. The Fed's FIMA repo facility offers approved official institutions temporary dollars against Treasury collateral, subject to pricing and repayment. It is neither free funding nor an unlimited intervention promise.
What to watch
Do yen-dollar moves persist when interest-rate incentives change, and do official announcements confirm intervention rather than leaving us to infer a policy action from a market move alone?
Check Ministry of Finance, Bank of Japan, and Treasury releases for decisions and funding details. A spot-price change or a gross buyback headline does not establish intervention or reduced net duration supply.
What would change the story?
An intervention-signaling explanation weakens if the currency effect reverses without supportive rate expectations. A buyback-yield claim needs evidence about net maturity exposure and funding, not the size of gross purchases.
Check your understanding One question, with an explanation
05 / Gold and model risk
Why gold can resist higher interest rates.
Why might gold rise even when real yields rise?
He offers a demand hypothesis without claiming that real rates stopped mattering.
“But since 22 and especially middle of 23, gold has been doing something that's totally different. So, I'm not saying the real rates are not relevant, but there's been something else. And we think it's [...] central bank buying and speculators betting on more central bank buying.”
Interest is a headwind, not the whole story.
Gold pays no interest. When safe inflation-adjusted yields rise, holding gold means giving up a more attractive alternative. That can weigh on its price. A stronger dollar can also matter, but neither relationship fixes the outcome when demand from other buyers changes.
Nordvig thinks official buying and investors anticipating more of it can offset those headwinds. The World Gold Council's estimates support examining that possibility; they do not identify the motive behind every purchase or guarantee a floor under the price.
An unexplained gain is a question.
A residual is the observed return minus a model's predicted return. If a rates-and-dollar model predicts a fall while gold rises, the gap tells us the model missed something. It does not tell us what. Independently measured purchases, positioning, and alternative model specifications are evidence to investigate, not labels to paste onto the gap.
Lab 06 Change one thing
How much of a gold move is still unexplained?
Change real rates or the hypothetical gold return. Watch the gap left by an assumed two-factor explanation.
Each action starts from the Original example.
The residual is an unexplained gap—not identified buying, a fitted model, or an observed gold return.
Model details & full chart Equations, data, and sharing
Model output / hypothetical
Return / gap (percentage points)- 1Modeled return
- 2Hypothetical
- 3Residual
Chart values Accessible data table
| Same hypothetical return period | Assumed returns and residual |
|---|---|
| Modeled return | -6.9 |
| Hypothetical | 12 |
| Residual | 18.9 |
This assumed model leaves 18.90 percentage points unexplained. New buyers are one hypothesis; changing coefficients, omitted factors, and noise remain alternatives.
Equation and limits
Modeled return = rate sensitivity x real-yield change + dollar sensitivity x dollar return. Residual = hypothetical return - modeled return.
- Every return and sensitivity is hypothetical. There is no regression, estimated coefficient, live gold observation, or causal identification in this lab.
- Return units are percentage points. No intercept, interaction, nonlinear response, or changing coefficient is included.
- The residual can contain omitted variables, measurement errors, model instability, or noise. A smaller in-sample residual does not establish better out-of-sample performance.
The link contains this lab’s inputs only. No responses are saved in this browser. Chart CSV values are unrounded; displayed labels are rounded.
Evidence, limits, and deeper reasoning For the curious and the technically minded
Mechanism in one sentence: Unexpected gold strength is a reason to investigate demand and model limits, not proof that interest rates have stopped mattering.
Distinguish reported purchases from estimates.
WGC estimates central-bank and other-institution demand of 863.3 tonnes in 2025; it attributes 57% to unreported activity. Reported official reserve changes are narrower evidence. Estimated undisclosed buying is not direct observation of purchasers or motives. Nordvig's flow interpretation is dated to the episode, not a live measure of demand.
Test a model instead of renaming its error.
A positive residual does not mathematically equal central-bank purchases. Changing rate sensitivity, omitted variables, data errors, or an unusual shock can produce it. The lab uses assumed inputs, not an estimated regression. An empirical test would align frequency and dates, fit on a training window, evaluate unseen data, and test stability. Gold can still lose through higher real yields, liquidation, or reversing demand.
What to watch
When real yields and the dollar create headwinds for gold, does independent evidence show sustained buying, or could a changing model relationship explain why the expected price response is absent?
Compare WGC estimates with reported reserve changes and dated flow evidence; the categories are not interchangeable. This guide supplies no live gold quote, and its residual exercise does not estimate actual buying.
What would change the story?
A persistent new-demand explanation weakens if independently measured buying fades, holdings reverse, or a model tested on unseen data explains the apparent break. One exceptional period does not establish a lasting change.
Check your understanding One question, with an explanation
06 / Korea and capital flows
When a winning market makes investors sell.
How can foreign investors sell while a stock market rises?
He explains why selling can follow a rally, not predict a fall.
“It used to be the case that the stock market in Korea was driven by foreigners. And now we've had a period where the stock market has gone up so much and is driven by something else that actually we have the stock market going up and foreigners have to sell to rebalance their portfolio.”
An export boom can spread beyond chipmakers.
An AI buildout elsewhere can raise demand for Korean memory chips. More export receipts can support profits, investment, wages, and dividends at home. Korea's specialization makes that channel powerful, but the gains need not reach every industry or household.
Official export figures support the semiconductor channel. They mix prices and quantities, however: higher receipts do not by themselves establish faster production, stronger household income, or a more valuable currency.
Winners can become sales.
A global fund with a 10% Korea target becomes overweight after Korean assets outperform. Selling restores the target without requiring a bearish forecast. For an unhedged dollar investor, local equity gains and the won's dollar value compound; they are not simply added. FRED quotes won per dollar, so a falling DEXKOUS means the won strengthens. Read the price move before interpreting the flow.
Evidence & context [1]
Lab 07 Change one thing
Why might an investor sell a winning market?
Change Korean stocks or the won. See the trade needed to restore the same portfolio target.
Each action starts from the Original example.
A hypothetical $100m portfolio, no fees or new cash. This mechanism does not identify actual investor flows.
Model details & full chart Equations, data, and sharing
Model output / hypothetical
Share of portfolio (%)- 1Starting weight
- 2After returns
- 3After rebalance
Chart values Accessible data table
| Korea allocation through the scenario | Korea weight |
|---|---|
| Starting weight | 10 |
| After returns | 13.7931 |
| After rebalance | 10 |
The Korea position is worth $15.12m before trading. Restoring the original target requires selling $4.16m, without changing the manager's target allocation.
Equation and limits
Korea USD return = (1 + local equity return) x (1 + won return) - 1. Trade = target weight x new portfolio value - Korea value before rebalancing.
- Hypothetical $100m starting portfolio; constant target weight; unhedged exposure. No contributions, taxes, fees, price impact, or changed investment views.
- Won return means the return of the won in dollars, not the percentage change in a KRW-per-dollar quote.
- A negative trade is a sale. This example establishes a possible mechanism, not the actual cause of Korean investor flows.
The link contains this lab’s inputs only. No responses are saved in this browser. Chart CSV values are unrounded; displayed labels are rounded.
Evidence, limits, and deeper reasoning For the curious and the technically minded
Mechanism in one sentence: Strong Korean returns can cause portfolio selling; a sale may reveal an allocation rule rather than a negative economic forecast.
Separate dated evidence from recent claims.
Korea's ministry reports $173.4 billion in semiconductor exports in 2025, up 22.2%, within total exports of $709.7 billion. Those data and the Bank of Korea's November 2025 outlook do not verify the interview's later equity rally or foreign-selling account. Testing rebalancing requires investor classifications, target weights, and actual flows, not just a currency chart.
Translation is not a currency forecast.
For a dollar return, multiply one plus the local-equity return by one plus the won's return measured in dollars, then subtract one. Exporters may keep receipts abroad, hedge, or invest overseas; imported inputs also cost money. A current-account surplus therefore does not guarantee appreciation. Memory-cycle, valuation, governance, trade-policy, and geopolitical risks remain.
Sources [17]
What to watch
As the won moves, do semiconductor export receipts translate into broader income and investment, and do actual investor flows support rebalancing rather than a simple story that buying drives every rally?
Use ministry export releases and Bank of Korea data alongside the reverse-quoted currency series. A claim about foreign portfolio selling needs its own classified flow dataset and a defined observation window.
What would change the story?
The broadening thesis weakens if export gains remain temporary price effects without income or investment growth. The rebalancing explanation also requires evidence that investors were restoring targets rather than abandoning exposure.
Check your understanding One question, with an explanation
07 / The household transmission
Strong spending does not mean every household is fine.
Why can weak retail stocks coexist with strong spending?
On whether a temporary tax-refund boost can explain continued household spending.
“If it's just a couple of months, then there could be something coming after that. If it's spread over 6 months, we have another couple of months of it. [...] I think that's [...] one thing that worries me a little bit, that we had some extra juice from those tax refunds.”
Spending is not a retail share price.
Retail stocks reflect expected profits, competition, and discount rates, not just household purchases. People can spend more on services while particular shops struggle. Even rising total spending can disguise higher prices rather than more goods and services; BEA's real consumption series adjusts for that distinction.
Refunds can temporarily add spendable cash without creating a permanent paycheck. The same refund spent over two months creates a larger monthly boost than one spread over twelve. Its timing matters, and households may instead save it or repay debt.
Higher rates reach some borrowers first.
An existing U.S. fixed-rate mortgage does not reset because market yields rise. Its principal-and-interest payment stays fixed; taxes and insurance can change. New buyers, movers, refinancing borrowers, and adjustable-rate borrowers face different exposure. Keeping a cheap mortgage can discourage moving. Meanwhile, the tax provisions discussed are qualified deductions, not a blanket repeal of tax on Social Security or all overtime.
Lab 08 Change one thing
How long can a one-time refund lift spending?
Spread the same refund over more months. The monthly boost shrinks; it does not become permanent income.
Each action starts from the Original example.
One household, one refund, equal monthly spending. The boost ends when the chosen share is used up.
Model details & full chart Equations, data, and sharing
Model output / hypothetical
Additional spending ($ / month)- 12 months
- 26 months
- 312 months
Chart values Accessible data table
| Assumed refund spend-down horizon | Same total spending, different timing |
|---|---|
| 2 months | 1,500 |
| 6 months | 500 |
| 12 months | 250 |
4.00% nominal growth with 3.00% price inflation gives 0.97% real growth. Separately, spending 60% of the refund over 6 months adds $500.00 per month for that period only.
Equation and limits
Real growth = (1 + nominal growth) / (1 + inflation) - 1. Monthly refund impulse = refund x share spent / months.
- The real-growth comparison and refund scenario are separate exercises. Do not add their percentages as if they describe the same household or a causal national-account decomposition.
- A single household, equal spending each month, and no multiplier, borrowing, interest, or repeated refund. Propensity to spend is an assumption.
- The same total refund spending can be concentrated or spread out. Baseline spending is used only to express the monthly boost.
The link contains this lab’s inputs only. No responses are saved in this browser. Chart CSV values are unrounded; displayed labels are rounded.
Lab 09 Change one thing
What would a new mortgage cost each month?
Change the new-loan rate or amount borrowed. Compare monthly principal and interest, not total housing costs.
Each action starts from the Original example.
An unchanged old fixed-rate loan keeps its contractual payment. Taxes, insurance, and fees are excluded.
Model details & full chart Equations, data, and sharing
Model output / hypothetical
Principal + interest ($ / month)- 13.50% reference
- 27.00% new loan
Chart values Accessible data table
| Two hypothetical originations, same principal and term | Monthly principal and interest |
|---|---|
| 3.50% reference | 1,796.1788 |
| 7.00% new loan | 2,661.21 |
The new 30-year loan costs $865.03 more per month in principal and interest ($10,380.37 per year in absolute difference). An unchanged existing fixed-rate loan keeps its contractual payment.
Equation and limits
Monthly payment = principal x m / [1 - (1 + m)^(-n)], where m = annual rate / 12 and n = 12 x years. At zero interest, payment = principal / n.
- Two hypothetical fully amortizing, fixed-rate loans. Same original principal and term, paid monthly. Input is the note rate, not a fee-inclusive APR.
- Principal and interest only; no tax, insurance, mortgage insurance, closing fees, points, or prepayment.
- Not an automatic reset, an offer of credit, or a comparison of remaining balances. A current mortgage survey rate is only a reference, not a personalized quote.
The link contains this lab’s inputs only. No responses are saved in this browser. Chart CSV values are unrounded; displayed labels are rounded.
Evidence, limits, and deeper reasoning For the curious and the technically minded
Mechanism in one sentence: Follow real spending, temporary cash, and each borrower's loan contract before concluding that either households or the whole economy are doing well.
Check the refund rules and the timing.
IRS provisions are capped, income-limited deductions for qualified tips, overtime premiums, and eligible seniors. The senior deduction is not a blanket elimination of tax on Social Security; the overtime deduction does not cover all overtime wages. Compare cumulative refund counts and amounts over aligned filing windows: withholding, eligibility, processing, and statutory holds affect the sample. A larger early average does not prove a larger economy-wide transfer.
Match each calculation to its question.
Exact real spending growth divides one plus nominal growth by one plus inflation, then subtracts one. BEA figures are revised; the real PCE level is seasonally adjusted and annualized, not monthly dollars spent. The mortgage lab compares two new or refinanced loans, never an automatic reset of an unchanged fixed-rate loan. The refund lab assumes a spending share and horizon rather than estimating a causal multiplier.
What to watch
Is spending still growing after inflation and temporary refunds are accounted for, and which households actually face a new mortgage rate rather than retaining an existing fixed-rate loan?
PCE estimates are revised, seasonally adjusted, and aggregate; they do not describe every household. Freddie Mac's mortgage survey is not an individual loan quote or the rate on all outstanding mortgages.
What would change the story?
A temporary-refund explanation weakens if real spending outlasts the expected impulse and durable real income supports it. An economy-wide mortgage-reset story fails for existing fixed-rate contracts, regardless of prevailing market yields.
Check your understanding One question, with an explanation
08 / Building a research habit
Turn the argument into a testable view.
What would make you change your mind?
Asked about debt worries, he distinguishes fiscal arithmetic from investor behavior.
“I do think at some point you have to take it seriously, and I think we're starting to sniff that [...] it's starting to impact the asset allocation. [...] So what matters is when investors respond to it.”
The thesis is a race, not a trade tip.
AI investment demands resources and financing before its productivity payoff is known. Companies and indebted governments may compete for investors' money. If long yields rise, refinancing spreads the costs unevenly. Policy choices and changing capital flows can redirect the pressure without removing it.
There are several possible endings. Capacity and productivity could catch up, supporting earnings. Persistent bottlenecks could keep financing and input costs high. Or demand could weaken, lowering yields while hurting profits. None of those scenarios makes every asset move the same way.
Choose what would change your mind.
Pick one link in the argument and write down a measurable claim, a time window, a source, and a competing explanation. Decide beforehand what result would weaken it. Then revisit the evidence. A useful framework must allow revision; explaining every possible outcome after it happens is not the same as having anticipated it.
Evidence, limits, and deeper reasoning For the curious and the technically minded
Mechanism in one sentence: Treat the thesis as a set of testable links, not a promise that one forecast or trade must win.
Do not promote correlation into causation.
"Long yields rose" is an observation; "AI issuance caused the rise" is an interpretation; "stocks will fall next" is a forecast. Accounting identities and hypothetical labs cannot identify either of the latter. An empirical study needs dated inputs, plausible controls, specified event windows, tests on unseen data, and disclosure of how results depend on model choices.
Keep the clocks and the instrument separate.
The episode is dated September 11, 2026; the original source review was September 14. This September 16 edition adds two monetary-mechanics references, not a full refresh. Use each data point's observation date and historical release vintage. Even a sound macro thesis can disappoint as an investment if its price, instrument, or horizon already embeds the expected outcome.
Sources [1]
What to watch
Across financing costs, inflation compensation, energy, and real spending, which scenario fits the joint evidence best, and what observation would distinguish it from the alternative you currently find least persuasive?
Daily market prices and monthly economic releases answer different questions and arrive on different schedules. Align dates, preserve historical vintages where possible, and do not mistake a coherent story for causal identification.
What would change the story?
Revise the framework if its stated links repeatedly fail: financing pressure does not emerge, productivity arrives faster, or transmission differs. A story that survives every outcome without revision is not testable.
Check your understanding One question, with an explanation
Synthesis / not a forecast
Three paths through the same moment.
Compare what each story would lead you to look for. None has an assigned probability, price target, or recommended trade.
Productivity-led expansion
Capacity catches up
Investment becomes productive capacity, bottlenecks ease, and earnings can grow without a renewed inflation surge.
Look for
- Real activity stays resilient.
- Inflation compensation stabilizes or falls.
- Input and energy constraints ease.
Keep in mind. Real yields could still rise with stronger productive investment. "Good growth" does not require every yield to fall.
Capital and input pressure
Scarcity persists
Demand for financing, electricity, and scarce equipment stays ahead of supply. Higher required yields reach borrowers over time.
Look for
- Long yields rise faster than the overnight rate.
- Inflation compensation or real yields increase.
- Energy prices and financing needs stay elevated.
Keep in mind. Those observations do not uniquely identify AI or fiscal supply. Growth news and other shocks can produce similar moves.
Growth disappointment
Demand gives way
Tighter financing or weaker income reduces demand. Some inflation pressure eases, but earnings and credit can deteriorate.
Look for
- Real consumption weakens.
- Expected policy rates and yields may fall.
- Energy demand may soften.
Keep in mind. Lower yields are not automatically bullish for equities if the decline reflects a much worse earnings outlook.
The observation desk
Keep the argument close to the data.
Public releases, not real-time trading quotes. Every value keeps its own observation date.
Dated snapshot retrieved Sep 15, 2026. Live refresh is checked when this page opens.
4.96Percent
Observed Sep 11, 2026
5.35Percent
Observed Sep 11, 2026
2.60Percent
Observed Sep 11, 2026
2.40Percent
Observed Sep 14, 2026
97.26U.S. dollars per barrel
Observed Sep 9, 2026
153.71Japanese yen per U.S. dollar
Observed Sep 11, 2026
Sparklines: up to 60 published observations, each with its own vertical scale. Gaps are not filled. They are not comparable return charts.
Open the data explorer13 indicators · history · matched-date yield curve
Observed data
PercentChart values Accessible data table
| Observation date (not retrieval date) | 10-year Treasury yield |
|---|---|
| 2026-05-05 | 4.43 |
| 2026-05-06 | 4.36 |
| 2026-05-07 | 4.41 |
| 2026-05-08 | 4.38 |
| 2026-05-11 | 4.42 |
| 2026-05-12 | 4.46 |
| 2026-05-13 | 4.46 |
| 2026-05-14 | 4.47 |
| 2026-05-15 | 4.59 |
| 2026-05-18 | 4.61 |
| 2026-05-19 | 4.67 |
| 2026-05-20 | 4.57 |
| 2026-05-21 | 4.57 |
| 2026-05-22 | 4.56 |
| 2026-05-26 | 4.5 |
| 2026-05-27 | 4.48 |
| 2026-05-28 | 4.45 |
| 2026-05-29 | 4.45 |
| 2026-06-01 | 4.47 |
| 2026-06-02 | 4.46 |
| 2026-06-03 | 4.49 |
| 2026-06-04 | 4.47 |
| 2026-06-05 | 4.55 |
| 2026-06-08 | 4.56 |
| 2026-06-09 | 4.53 |
| 2026-06-10 | 4.55 |
| 2026-06-11 | 4.45 |
| 2026-06-12 | 4.48 |
| 2026-06-15 | 4.47 |
| 2026-06-16 | 4.43 |
| 2026-06-17 | 4.49 |
| 2026-06-18 | 4.46 |
| 2026-06-22 | 4.51 |
| 2026-06-23 | 4.5 |
| 2026-06-24 | 4.41 |
| 2026-06-25 | 4.4 |
| 2026-06-26 | 4.38 |
| 2026-06-29 | 4.38 |
| 2026-06-30 | 4.44 |
| 2026-07-01 | 4.48 |
| 2026-07-02 | 4.49 |
| 2026-07-06 | 4.48 |
| 2026-07-07 | 4.55 |
| 2026-07-08 | 4.56 |
| 2026-07-09 | 4.54 |
| 2026-07-10 | 4.56 |
| 2026-07-13 | 4.62 |
| 2026-07-14 | 4.58 |
| 2026-07-15 | 4.55 |
| 2026-07-16 | 4.57 |
| 2026-07-17 | 4.55 |
| 2026-07-20 | 4.6 |
| 2026-07-21 | 4.63 |
| 2026-07-22 | 4.67 |
| 2026-07-23 | 4.71 |
| 2026-07-24 | 4.69 |
| 2026-07-27 | 4.65 |
| 2026-07-28 | 4.61 |
| 2026-07-29 | 4.67 |
| 2026-07-30 | 4.68 |
| 2026-07-31 | 4.75 |
| 2026-08-03 | 4.7 |
| 2026-08-04 | 4.63 |
| 2026-08-05 | 4.63 |
| 2026-08-06 | 4.69 |
| 2026-08-07 | 4.65 |
| 2026-08-10 | 4.72 |
| 2026-08-11 | 4.7 |
| 2026-08-12 | 4.68 |
| 2026-08-13 | 4.63 |
| 2026-08-14 | 4.68 |
| 2026-08-17 | 4.72 |
| 2026-08-18 | 4.71 |
| 2026-08-19 | 4.65 |
| 2026-08-20 | 4.69 |
| 2026-08-21 | 4.74 |
| 2026-08-24 | 4.7 |
| 2026-08-25 | 4.64 |
| 2026-08-26 | 4.66 |
| 2026-08-27 | 4.67 |
| 2026-08-28 | 4.73 |
| 2026-08-31 | 4.75 |
| 2026-09-01 | 4.79 |
| 2026-09-02 | 4.79 |
| 2026-09-03 | 4.77 |
| 2026-09-04 | 4.78 |
| 2026-09-08 | 4.8 |
| 2026-09-09 | 4.83 |
| 2026-09-10 | 4.95 |
| 2026-09-11 | 4.96 |
History for this series is unavailable. Open the primary FRED source below.
Matched-date check (Sep 11, 2026): the 10y minus 2y slope is 33.0 basis points. A steeper curve is not itself a causal explanation.
Where these numbers come from, and what “latest” means
A same-origin endpoint requests a fixed list of public FRED series. It does not use a paid feed, an API key, a third-party proxy, or the podcast’s quoted prices. The site includes a dated, reproducible snapshot; a successful refresh replaces it with newly retrieved observations. If refresh fails, the retained snapshot stays explicitly labeled.
Daily series can lag business days; weekly mortgage data and monthly PCE releases follow different schedules. Retrieval time is not observation time. Unusually old values are marked stale. A successful request does not make an old release current. No interpolation, prediction, or silent zero fills are used.
Interest rates are annualized percentages; FRED’s FX quotes are local currency per U.S. dollar. WTI is a published spot price, not a futures quote. The real-PCE level is billions of chained dollars at a seasonally adjusted annual rate, not one month’s expenditure. Each series’s source and units appear in the explorer. Downloads preserve the displayed observations, not point-in-time historical release vintages.
Gold demand is linked in the evidence ledger, not presented as a made-up live gold feed. Yield-curve comparisons use a common observation date. A missing component means the comparison is unavailable.
Appendix A / research method
Rigorous about the boundaries.
The models teach accounting and valuation mechanics; they do not identify what caused a market move or reproduce proprietary Nordvig research. An identity, an observation, and a forecast are different kinds of claim.
Open the research notebook Derivations, evidence standards, and qualifications
Identity is not causality.
Bond valuation, debt-ratio accounting, currency translation, and real-growth conversion are identities under stated conventions. They cannot tell us which event caused a market move.
Assumptions are visible.
Inputs, units, horizons, equations, and omitted forces are disclosed beside each lab. Gold sensitivities and refinancing rates are assumptions, not estimated structural parameters.
Dates are part of the evidence.
The source review is dated September 14, 2026; the episode was published September 11. Live observations may be newer. A revised time series is not a historical information set.
One chart is not a conclusion.
Public rates and currency data provide context. They cannot independently verify investor flows, hyperscaler issuance comparisons, hidden official purchases, or private policy expectations.
Derivation notebook From cash flows to the lab equations
Debt and the denominator
Let D be nominal debt, Y nominal GDP, r the average interest rate, and P the primary deficit. Starting from D(t) = (1 + r)D(t-1) + P(t), divide by Y(t) = (1 + g)Y(t-1). This gives d(t) = [(1 + r)/(1 + g)]d(t-1) + p(t). The deficit term uses current-period GDP. At a constant starting debt ratio, the required primary surplus is [(r - g)/(1 + g)]d.
Bond duration and convexity
With face value 100, annual coupon c, semiannual period count N, and yield y, each period’s cash flow is discounted by (1 + y/2) to its period number. Modified duration is minus the first derivative of price with respect to y, divided by price. Convexity is the second derivative divided by price. The plotted approximation is a second-order Taylor expansion; the solid curve uses the exact cash-flow sum.
Currency compounding and hedging
For a dollar investor, the gross return of a foreign asset is its gross local return multiplied by the gross return of the foreign currency in dollars. For a yen-funded dollar position, reconvert with yen per dollar instead. Setting the reconversion price to S x (1 + rJPY)^T / (1 + rUSD)^T makes the frictionless asset proceeds match the yen repayment.
Purchasing power and amortization
A nominal spending ratio divided by the matching price-index ratio gives the real quantity ratio. For a fully amortizing loan, the monthly payment follows from setting principal equal to the present value of n equal monthly payments. At a zero note rate, the payment is simply principal/n. Rounding is for display; calculations retain full precision.
What a publishable empirical extension would require
Define the hypothesis before choosing the sample. Preserve release vintages; align trading times, currencies, horizons, and maturities. Identify the financing measure (stock or flow, gross or net, face amount or duration-equivalent supply). State an identification strategy, confounders, and alternative mechanisms. Evaluate stability and out-of-sample performance, report uncertainty, and disclose sensitivity to the sample and specification.
None of the lab curves is an estimated confidence interval, calibrated general-equilibrium response, or probability distribution. No Monte Carlo decoration substitutes for a justified data-generating process. Models omit transaction costs, taxes, market impact, liquidity, and feedback unless expressly included.
Important qualifications to the conversation
- Buybacks: officially liquidity and cash-management operations, not costless debt cancellation or an announced QE/yield-control program. [8]
- Mortgages: U.S. fixed-rate principal-and-interest payments do not automatically reset to market rates. [12]
- Tax refunds: qualified deductions and filing timing are not blanket tax repeal or permanent recurring income. [20] [21] [19]
- Issuance and flows: the interview’s long-end issuance comparison and recent market-flow interpretations are attributed claims, not independently reconstructed findings.
- CBO: the primary 2026 data locator is linked, but this review relies on an explicitly identified secondary summary for the numerical projection. [10] [11]
Appendix B / a shared vocabulary
A term shouldn’t stop the learning.
32 terms
Basis point (bp)
One hundredth of one percentage point. A yield move from 4.50% to 4.85% is 35bp.
See it in the argumentBreakeven inflation
The matched-maturity nominal-minus-TIPS yield spread. It contains expected CPI inflation plus inflation-risk compensation minus TIPS liquidity effects; it is not pure expected inflation.
See it in the argumentCapital expenditure (capex)
Spending on long-lived productive assets. Spending occurs now; the useful capacity and profits may arrive later.
See it in the argumentCarry
The interest or income difference earned while holding a position, before adverse price moves, funding changes, costs, and risks.
See it in the argumentConvexity
Curvature in the bond price-yield relationship. It explains why a straight-line duration estimate becomes less accurate for larger yield moves.
See it in the argumentCovered interest parity
In a frictionless setting, a currency forward offsets the difference in interest rates between two currencies, removing a free hedged interest-rate arbitrage.
See it in the argumentCurrent account
A country's trade in goods and services plus net primary income and transfers. A surplus is not a one-for-one forecast of currency appreciation.
See it in the argumentDebt held by the public
Federal debt held outside U.S. federal government accounts, including domestic and foreign investors and the Federal Reserve. It differs from gross debt.
See it in the argumentDiscount rate
The rate used to translate future cash flows into present value. Its meaning depends on the risk and timing of the cash flows.
See it in the argumentDuration
A measure of cash-flow timing and interest-rate sensitivity. Modified duration approximates the percentage price change for a small change in yield.
See it in the argumentEffective interest rate
The average rate paid on the debt in the model, not the yield on the latest 10-year or 30-year bond. Refinancing changes it gradually.
See it in the argumentFree cash flow
Commonly operating cash flow minus capital expenditure. Definitions vary. Subtracting capex from it again double-counts investment spending.
See it in the argumentGross versus net issuance
Gross issuance includes debt sold to refinance maturities. Net issuance subtracts retirements; neither measure is the outstanding stock of debt.
See it in the argumentHyperscaler
A very large technology company operating extensive computing infrastructure. Define the issuer universe before comparing its financing needs with sovereign issuance.
See it in the argumentLiquidity premium
Compensation for holding an asset that may be more difficult or costly to trade. TIPS liquidity effects can depress measured inflation breakevens.
See it in the argumentMarginal buyer
The investor whose incremental willingness to trade helps set the market price. That investor can change as flows, constraints, and participation change.
See it in the argumentNominal versus real
Nominal amounts use current prices. Real amounts remove a specified price change to express quantities or purchasing power. Match the price index and period.
See it in the argumentOff-the-run Treasury
A Treasury security issued before the most recent benchmark issue at a given tenor. It often trades less actively than the on-the-run benchmark.
See it in the argumentPrimary deficit
Government spending excluding interest, minus revenue. Positive means new primary borrowing; a primary surplus has the opposite sign.
See it in the argumentProductivity J-curve
The idea that complementary investment and adjustment can precede measured productivity gains from a general-purpose technology.
See it in the argumentReal yield
A yield expressed relative to inflation. A quoted TIPS yield is market-priced and may contain risk and liquidity effects, not a pure estimate of the economy's neutral real rate.
See it in the argumentRegime change
A change in the relationships or behavior generating data. An old statistical correlation can fail when institutions, constraints, or market participants change.
See it in the argumentResidual
An outcome minus the model's prediction. It measures what the model did not explain; it does not identify a cause.
See it in the argumentRollover / refinancing
Replacing maturing borrowing with new borrowing. New debt may carry a different rate; an unchanged fixed-rate contract does not reset by itself.
See it in the argumentSAAR
Seasonally adjusted annual rate. An annualized monthly spending level is not the amount spent in one month and is not a year-over-year growth rate.
See it in the argumentStock versus flow
A stock is measured at a point in time, such as debt outstanding. A flow is measured over a period, such as bonds issued in a month.
See it in the argumentTerm premium
The compensation for bearing long-horizon interest-rate risk beyond the expected path of short rates. It is estimated, not directly read off one yield.
See it in the argumentUSD/JPY and USD/KRW
Yen or won per one U.S. dollar. A falling quote means the yen or won strengthens against the dollar, not the other way around.
See it in the argumentBank reserve balances
Balances banks hold at the Fed to settle payments: assets for banks, liabilities for the Fed. These reserves are not deposits households can hold, and more reserves do not automatically create loans or consumer-price inflation.
See it in the argumentBank deposit
A household's or business's claim on a commercial bank: an asset for the customer and a liability for the bank. A nonbank bond seller receives this deposit, not reserve balances at the Fed.
See it in the argumentTreasury General Account (TGA)
Treasury's cash account at the Fed. Spending from it shifts Fed liabilities from TGA balances to bank reserves without expanding the Fed's balance sheet; funding receipts reverse that shift, all else equal.
See it in the argumentQuantitative easing (QE)
Central-bank asset purchases financed by newly created reserves. Fed assets and liabilities expand, all else equal; nonbank sellers receive bank deposits. Unlike a TGA transfer, this is money creation, but lending and consumer-price inflation are not automatic.
See it in the argumentNo matching terms. Try a broader word, or clear the filter to browse every definition.
Appendix C / evidence ledger
Follow every claim back.
Open a source note to see what it establishes and where its limits are. Current data definitions and original-provider links live in the data desk.
- [1]
The Most Interesting Macro Moment of My Lifetime with Jens Nordvig (episode 259) (opens in a new tab)
What this source supports
Used for: Identifies the episode, guest, hosts, title, and publication date.
Boundary: Quotations and approximate timestamp ranges come from a reader-supplied transcript, not independently verified audio. Words are preserved; punctuation is edited and [...] marks internal omissions. The transcript has no speaker labels, so attribution follows conversational context. End times use the next supplied timestamp, not a verified audio cutoff. Excerpts are not endorsements of this guide.
- [2]
Jens Nordvig: official biography (opens in a new tab)
What this source supports
Used for: Exante founding in 2016; Nomura, Goldman Sachs, and Bridgewater roles; economics PhD from the University of Southern Denmark; The Fall of the Euro (2013).
Boundary: Employer biography. Five consecutive first-place survey rankings are a biographical claim, not a record of investment performance.
- [3]
Exante Data to become Vanda Macro Intelligence (opens in a new tab)
What this source supports
Used for: Identifies Nordvig as Vanda President and announces the Exante brand transition planned for October 1, 2026.
Boundary: Corporate announcement, not independent evaluation. The October transition was still prospective at this guide's review date.
- [4]
First ranking for FX research in the Americas for the fourth consecutive year (opens in a new tab)
What this source supports
Used for: Contemporaneous confirmation of the 2011-2014 Institutional Investor All-America Fixed Income Research currency/FX survey rankings.
Boundary: A research-client survey, not a forecasting-accuracy competition. This release does not establish the fifth year.
- [5]
Energy and AI: executive summary (opens in a new tab)
What this source supports
Used for: Data-centre electricity demand, infrastructure investment, geographic concentration, and grid bottlenecks.
Boundary: Data centres are not exclusively AI. Forward demand paths are scenarios, not proof of a particular inflation outcome.
- [6]
The Productivity J-Curve: How Intangibles Complement General Purpose Technologies (opens in a new tab)
What this source supports
Used for: A model of complementary intangible investment and the lag between a general-purpose technology and measured productivity gains.
Boundary: A mechanism, not an estimated timetable for current generative AI or a prediction of net inflation.
- [7]
Tips from TIPS: Update and Discussions (opens in a new tab)
What this source supports
Used for: Decomposition of breakevens into expected inflation, inflation-risk compensation, and TIPS liquidity effects.
Boundary: Extracting expected inflation requires a model. Federal Reserve staff research is not an FOMC forecast.
- [8]
FAQs about Treasury securities buybacks (opens in a new tab)
What this source supports
Used for: Official liquidity-support and cash-management purposes, funding, and retirement of purchased securities.
Boundary: Not an announced yield-control target or a Federal Reserve QE program. Retirement of a bond requires payment and financing.
- [9]
Understanding the National Debt (opens in a new tab)
What this source supports
Used for: Distinguishes debt held by the public, intragovernmental holdings, and total national debt.
- [10]
The Long-Term Budget Outlook Data: 2026 to 2056 (opens in a new tab)
What this source supports
Used for: Primary locator for the 2026 long-term budget baseline and its assumptions.
Boundary: The original tables could not be retrieved during this guide's source review. Numerical discussion is explicitly attributed to the secondary summary below, not represented as independently reproduced CBO data.
- [11]
Debt Rises to 175% of GDP Under CBO's Long-Term Outlook (opens in a new tab)
What this source supports
Used for: Reports the CBO baseline of public debt reaching 175% of GDP in 2056.
Boundary: CRFB is a fiscal-policy advocacy organization. Separate its reported baseline numbers from its policy recommendations. Projections depend on the baseline vintage.
- [12]
Understand the different kinds of loans available (opens in a new tab)
What this source supports
Used for: Fixed-rate versus adjustable-rate loan mechanics and the distinction between principal-and-interest payments and total housing costs.
- [13]
Who decides and conducts foreign exchange intervention? (opens in a new tab)
What this source supports
Used for: The Minister of Finance directs Japanese intervention; the Bank of Japan acts as agent.
- [14]
FIMA Repo Facility FAQs (opens in a new tab)
What this source supports
Used for: Temporary collateralized dollar liquidity for approved foreign monetary authorities against Treasury securities.
Boundary: Eligibility, pricing, collateral, approval, and repayment matter. It is not free financing or an unlimited intervention promise.
- [15]
Gold Demand Trends: Full Year 2025 - Central banks (opens in a new tab)
What this source supports
Used for: Estimated central-bank and other-institution gold demand of 863.3 tonnes in 2025, including estimated unreported activity.
Boundary: Reported changes and estimated undisclosed purchases are different evidence. WGC attributes 57% of the annual total to unreported activity; motives are not directly observed.
- [16]
Korea's Annual Exports Reach New Highs in 2025 (opens in a new tab)
What this source supports
Used for: 2025 total exports of $709.7 billion and semiconductor exports of $173.4 billion, with chips up 22.2%.
Boundary: Nominal export values reflect prices as well as volumes. They do not prove the transcript's more recent equity, currency, or trade-surplus claims.
- [17]
Economic Outlook, November 2025 (opens in a new tab)
What this source supports
Used for: The semiconductor cycle as a driver of Korea's outlook, alongside trade-policy and growth uncertainties.
Boundary: A dated outlook, not a September 2026 forecast.
- [18]
Personal Income and Outlays, July 2026 (opens in a new tab)
What this source supports
Used for: Separately reports nominal spending, real spending, personal income, saving, and the PCE price indexes.
Boundary: Estimates are revised. Aggregate spending does not describe every household and does not identify a causal effect of refunds.
- [19]
Filing season statistics: week ending February 13, 2026 (opens in a new tab)
What this source supports
Used for: Explains filing-season timing and the PATH Act hold on certain refunds.
Boundary: Early averages are composition-sensitive. Compare aligned cumulative periods, counts, and amounts, not just a single average refund.
- [20]
Tax deductions for working Americans and seniors (opens in a new tab)
What this source supports
Used for: Qualified, capped, and income-limited deductions for tips, overtime premiums, and eligible seniors for 2025-2028.
Boundary: Not an exemption for all overtime income or a blanket repeal of Social Security taxation. This guide is not tax advice.
- [22]
Credit and Liquidity Programs and the Balance Sheet: Federal Reserve liabilities (opens in a new tab)
What this source supports
Used for: Securities purchases credit bank reserve accounts. Reserves and TGA balances are Fed liabilities; Treasury payments add bank reserves and receipts drain them, all else equal.
Boundary: No publication date verified. Describes accounting mechanics, not a lending or inflation forecast or the net effect of simultaneous operations. This addition does not refresh the original source review.
- [23]
How the Fed Changes the Size of Its Balance Sheet (opens in a new tab)
What this source supports
Used for: Fed asset purchases create reserves. Nonbank sellers receive bank deposits, while bank-inventory sales swap bank assets without necessarily creating nonbank deposits. Public sellers exchange equal-value assets.
Boundary: Staff explanation, not an FOMC announcement. Simplified four-sector balance sheets exclude subsequent portfolio adjustments and price effects; the initial asset exchange is not a claim that QE has no economic effects.