May 4, 2026

Government Bonds: Price, Yield, Paper Losses & The Yield Curve

macrofixed incomeyield curve
may 2026
government bonds · explained simply

Government Bonds, Price & Yield, Paper Losses, And What The Yield Curve Is Saying

Buy a 10-year US Treasury in 2020 at 0.6% interest - by 2023 the price has dropped about 30%, and that's what pushed SVB into insolvency in 48 hours. Buffett's Berkshire sits on a record $397 billion in cash, 90% of it short-term T-bills - holding more than the Fed itself. The US yield curve stayed inverted for 27 months straight, from 7/2022 to late 2024 - the longest in history. In early 2025 it flipped back to normal. Here's the irony: that exact moment of uninversion is the real recession signal, not a sign that the danger has passed. This piece explains the price-yield mechanism, paper losses, and how to read the curve - no formulas, no unnecessary jargon.

4.38%
us 10y yield, 5/2026
+50bp
10y - 2y spread, now normal
$397B
berkshire cash, q1 2026
2.525%
jp 10y, 29-year high
2.14%
sg ssb 10y average

1 · What A Government Bond Actually Is

Picture the US government needing $5,000 billion in 2026 to cover military salaries, social security, and interest on old debt. Tax revenue isn't enough. Whatever's left over, it has to borrow from whoever's willing to lend - ordinary citizens, pension funds, foreign banks, central banks. This borrowing mechanism has a name: issuing bonds.

A bond is a standardized IOU: "The US Treasury will pay you back $1,000 in 10 years, plus $43.80 in interest every year for those 10 years." You hand the Treasury $1,000 today in exchange for that promise. You can hold it to maturity, or sell it to someone else at any time on the secondary market.

Three core concepts need to be untangled right from the start:

Face value (par)
$1,000
The amount the government pays you back at maturity. Fixed, never changes.
Coupon
4.375%
The interest rate fixed on the bond at issuance. The government pays $43.75/year for every $1,000 of face value. Never changes.
Yield
4.38%
The actual return you get if you buy the bond today at the market price and hold to maturity. Changes every second along with price.

The third concept - yield - is the hardest for newcomers to grasp, and it's the one every piece of financial news mentions. Take the same bond with a 4.375% coupon: if the market price drops to $950, the new buyer's yield will be higher than 4.375% (because they paid less but still receive the full interest and face value). If the price jumps to $1,050, yield drops. This is the first see-saw the entire bond market runs on.

Classification by maturity

The US Treasury issues three types based on length of term, each playing a different role:

TypeMaturityCharacteristicsOutstanding (3/2026)
T-Bill 4 weeks - 1 year Pays no coupon. Sold at a discount, then matures at face value. The difference is the interest. $6.82T (22%)
T-Note 2 - 10 years Pays coupon semi-annually. The largest segment, treated as the "risk-free rate" the market references. $15.88T (50%)
T-Bond 20 - 30 years Pays coupon semi-annually. Most sensitive to rate changes. Suited to pension funds and insurers. $5.34T (17%)

Total outstanding US debt in early 4/2026 stood at $38.98 trillion - of which $31.41T is debt held by the public (bonds actually trading in the market), and $7.57T is intragovernmental debt (Treasury Fiscal Data). The average rate across all marketable debt is 3.365% - meaning the US is paying roughly $1,050 billion in interest every year, nearly equal to the entire defense budget.

Why the "risk-free rate" is considered risk-free. The US government has never defaulted on USD-denominated bonds since its founding (there was one technical glitch in 1979 that doesn't count). The reason is simple: US debt is paid in USD - and the Fed can print USD at any time. The real risk isn't failing to pay, it's paying back in a currency that's worth less. That's inflation risk, not credit risk. For the same reason, JGBs (Japan) and SGS (Singapore) are also considered risk-free in their own currency - because the BOJ and MAS both have the power to print their own money.

Why government bond yield is the "floor" under every other valuation

This is the single most important concept, and one of the least explained in mainstream finance. Every investment decision, every asset valuation in the world - Apple stock, a house in Hanoi, a solar project, a SaaS company raising a round - is measured, indirectly, against one single question: "Does this asset pay me more than what I could get for free from a government bond?"

The reason is simple: government bonds are considered risk-free. Every other asset carries risk - bankruptcy, devaluation, poor liquidity, inflation. Lenders/investors demand a reward for taking on that risk. Reward = (expected return of the asset) − (yield on a government bond of the same maturity). This is called the risk premium.

In other words, Treasury yield is the floor. Everything else needs to sit above it to have a reason to exist. When the floor shifts, everything above it shifts too. This is the core mechanism that explains why the Fed's 2022-2023 hiking cycle simultaneously crushed growth stocks, crypto, real estate, and M&A: not because those things "got worse" - but because the comparison floor moved.

Walking through each asset class, with the Apple 2021 vs 2023 example

Comparison table for when the 10Y yield rises from 2% to 4.5%:

AssetHow it links to Treasury yieldConsequence
Growth stocks (DCF) Discount rate = 10Y yield + equity risk premium (~5%). Future earnings get discounted more heavily. Theoretical drop of ~25-35%
Real estate (cap rate) Rent yield/price must exceed Treasury yield + liquidity premium (~2-3%). Prices drop ~30% or rents rise correspondingly
Corporate bonds Yield = Treasury + credit spread (Apple ~+0.6%, BB-rated ~+3%). The whole yield ladder shifts up 250bp
30Y mortgage ≈ 10Y yield + MBS spread (1.5-2%). Mortgage rates jump from 3% to 7% (2022-2023)
VC / tech startups Multiples move inversely with yield. Cheap money → 30x; expensive money → 5-8x. Valuations fall 60-80%, down rounds become common
Gold & Bitcoin Pay no coupon. Treasury at 4.5% = a 4.5%/year opportunity cost. Downward pressure unless offset by another factor

The Apple example: In December 2021, the US 10Y yield was 1.5%, Apple traded at a P/E of 30x with a $3,000 billion market cap - the market was saying, "Treasury only pays 1.5% (equivalent to a P/E of 67x), so Apple at 30x is relatively cheap." By October 2023, the 10Y yield had jumped to 5%, and the Treasury-equivalent P/E was just 20x. Apple at a P/E of 30x now looked expensive - it fell from $182 to $124, a loss of ~30% while the company itself hadn't changed at all. What changed was the comparison floor.

Three practical takeaways:

  1. High Treasury yield → "don't invest" becomes a rational decision. In 2026, at a yield of 4.4%, the only competing option has to clear 7-9% (floor + equity premium) - the logic behind Berkshire's $397B pile.
  2. Sharp drop in Treasury yield → a signal to re-rate everything. This is the mechanism behind "everything rallies" during QE periods.
  3. Whenever you read financial news, always ask "what's the 10Y yield?" It's the background variable that drives every other number - nobody mentions it, yet it decides everything.

2 · The Price-Yield Mechanism: The Most Basic See-Saw

This is the one rule you need to remember about the bond market: price and yield always move in opposite directions. When yield rises, price falls. When yield falls, price rises. One goes up, the other must go down. Like a see-saw.

Yield ↑
Old bond prices ↓
When market interest rates rise, newly issued bonds carry a higher coupon. Old bonds (with lower coupons) become less attractive → they must be sold at a discount to find a buyer.
Yield ↓
Old bond prices ↑
When market interest rates fall, new bonds carry a lower coupon. Old bonds (with higher coupons) become "gold" → their price jumps above face value.

A concrete example with a real bond

You buy a 10-year US Treasury in August 2020 - COVID era, the Fed had cut rates to zero. This bond has:

  • Face value: $1,000
  • Coupon: 0.625%/year (i.e., $6.25/year for every $1,000)
  • Maturity: 8/2030
  • Purchase price: $1,000

By October 2023, the Fed had raised rates from 0% to 5.5% to fight inflation. Newly issued 10-year Treasuries at that point yielded around 5%. With the same $1,000, someone could buy a new bond earning $50/year - instead of holding your old bond that only pays $6.25/year. The consequence?

To sell your old bond, you have to cut the price significantly. A buyer needs to earn a "yield equivalent" to the new bond (~5%). Since the coupon is fixed, the only way to push yield up is to lower the price. A 10-year bond with a 0.625% coupon, in a 5% yield environment, trades at roughly $680-700 - a loss of about ~30% versus your purchase price. This isn't theoretical: TLT (the 20+ year Treasury ETF) fell from $171 in early 2020 to $82 by late 2023, a 52% loss - the worst stretch in 40 years for long-term US bonds.

If you hold the bond all the way to 8/2030, you'll still get back the full $1,000 face value plus every interest payment along the way. You haven't "really" lost a dollar of cash flow. But on paper, every day of those intervening 7 years, the market value of your portfolio shows a loss. That's the next concept.

A short formula for readers who want to go deeper

Rule of thumb: price change ≈ −Duration × yield change.

Duration is the "weighted average" length of the cash flows. A 10-year Treasury has a duration of roughly 8-9. That means if yield rises by one percentage point (100 basis points), price falls roughly 8-9%.

A 30-year bond has a duration of roughly 18-20 → the same 1% yield move drops the price 18-20%. That's why long-term bonds are far more sensitive to rate changes, and riskier despite looking "safer" than short-term ones.

Three rules follow naturally:

  1. Longer maturity → higher duration → bigger price swings.
  2. Lower coupon → higher duration (because most of the cash flow is pushed to the end, at face value).
  3. Lower yield → higher duration (because the same absolute move is a bigger relative move).

That's why Japanese bonds during the 0.1%-yield era of 2022 were extremely risky - record-high duration. When the BOJ started raising rates in 2024-2025, long-dated JGBs saw brutal price declines.

3 · When A "Paper Loss" Stays On Paper - And When It Becomes Real

A paper loss (unrealized loss) is when the market value of a bond you hold is lower than what you paid, but you haven't sold. If you hold to maturity and the government doesn't default, you still get the full face value back - the paper loss "disappears." But if you're forced to sell in the meantime, the paper loss turns into a real one.

This concept sounds simple, but it's exactly what took down Silicon Valley Bank in 48 hours in March 2023.

The SVB story - the 2023 lesson every bond investor needs to know

In 2020-2021, SVB took in a massive wave of deposits from tech startups (thanks to Fed money-printing during COVID and a venture capital boom). Deposits grew from $60 billion in 2019 to $189 billion by the end of 2021. SVB couldn't lend all of it out - so it bought long-dated government bonds and MBS (mortgage-backed securities), with yields around 1.5% at the time.

By the end of 2022, after the Fed had raised rates from 0% to 4.5%, SVB's $91.3 billion HTM (held-to-maturity) bond portfolio had a market value of only $76.2 billion - a paper loss of $15.1 billion. SVB's book equity was only $16.3 billion (CFA Institute). In other words, if it had to sell everything to pay depositors, SVB would have nearly wiped out its entire equity.

The "hide till maturity" accounting trick. Under GAAP, banks are allowed to classify bonds into three buckets: trading (short-term, marked to market daily), available-for-sale (can be sold, marked to market but doesn't hit earnings), and held-to-maturity (HTM, carried at purchase cost, where paper losses don't show up on the main financial statements). SVB pushed the bulk of its portfolio into HTM. The $15 billion loss only appeared in a footnote at the back of the report - it didn't touch reported capital.

In March 2023, a few large startups pulled their cash as they burned through funding. SVB was forced to sell $21 billion of its AFS bonds - realizing a $1.8 billion loss. The news spread. Depositors tipped each other off over Twitter and Slack: "SVB is about to collapse." Within 24 hours, $42 billion in deposits left the bank - the fastest bank run in history. On the morning of March 10, the FDIC took over. A paper loss turned into a real one within 48 hours.

The lesson: a paper loss isn't actually harmless. It's harmless only if you have the right to hold to maturity. The moment you're forced to sell - because you need liquidity, because depositors are pulling out, because of a margin call - a paper loss turns into a real one instantly.

Why didn't SVB just buy short-term bonds to stay safe?

A natural question: if Berkshire can choose 4-week T-bills, why couldn't SVB do the same? Same environment, same excess cash, same credit-risk-free assets - why did SVB pick the harder road?

The answer lies in the banking business model, not in a bad bet on interest rates. There are three specific reasons:

Three reasons SVB chose long-term over T-bills - and why each reason seemed "reasonable" back in 2021

1. "Reach for yield" - the pressure to earn more in a 0%-rate world. In 2020-2021, the Fed held rates near 0%. 4-week T-bills paid only 0.05%/year at the time. SVB took in $189 billion in deposits - the excess over what it lent out alone was about $120 billion. Put $120B into 0.05% T-bills, and you earn $60 million/year. Put the same money into 10-year MBS yielding 1.5%, and you earn $1.8 billion/year - 30 times more. The board, shareholders, and analysts all pressured management to earn a return. SVB's CFO reasoned: "T-bills won't even cover the bank's operating costs."

2. "Maturity transformation" - literally the core business of banking. Every bank makes money by borrowing short, lending long: take in deposits at 0% (withdrawable anytime) → lend for 30-year mortgages at 4%. The spread is the net interest margin (NIM). T-bills, with duration near zero, generate no NIM. If SVB held only T-bills, the bank would have no reason to exist - customers could just buy T-bills directly through Treasury Direct without going through SVB. The bank was forced to accept a duration mismatch in order to earn a margin.

3. Tech clients didn't borrow the way traditional bank clients do. SVB's deposits grew from $60B to $189B (tripling) in two years - but its tech startup clients already had plenty of capital from VC and didn't need loans. SVB couldn't lend enough to absorb the enormous deposit inflow. The $120B excess had to be invested - and it went into long-dated MBS for a yield higher than the 10Y Treasury. This is a key difference versus JPMorgan/Wells Fargo (whose clients borrow heavily, leaving less that needs to go into bonds).

Comparing to Berkshire to see the core difference.
Berkshire is not a bank. Its money is shareholder capital and insurance premiums - not deposits that can be withdrawn at any moment. Berkshire has no depositor who can "pull" $42B in 24 hours. It has the right to earn nothing beyond T-bill interest when there's no good opportunity. Performance pressure only comes from shareholders at the annual meeting - and Buffett has enough credibility to ask them to wait.

SVB is a bank. Deposits can evaporate through a phone app within 6 hours. Every quarter it has to report NIM - if NIM is low, the stock falls, the board pushes back. SVB's management didn't have the luxury of saying "we'll just sit in T-bills for 5 years and wait for the right moment" the way Buffett can.

The real problem wasn't "SVB was wrong to buy long-term bonds" - it's that they lacked hedging. Banks have a standard tool to insure against interest rate risk: interest rate swaps. You hold a 10-year bond, buy a swap that "pays fixed, receives floating" - when rates rise, the bond loss is offset by the swap gain. JPMorgan, Wells Fargo, and Bank of America all do this with most of their HTM portfolios. SVB hedged none of it on its $91 billion HTM book - an almost unimaginable lapse in risk management. According to an MIT Sloan study, if SVB had hedged just 50% of the portfolio, the paper loss would have dropped from $15 billion to roughly $5 billion - and it wouldn't have collapsed.

The subtle conclusion: SVB didn't collapse because it bought long-term bonds. It collapsed because it bought long-term bonds against an extremely short-term deposit base, with no hedging, and with no cushion for when deposits evaporated.

One-sentence summary for the reader: banks are forced to accept a duration mismatch to survive - unlike individuals and non-bank companies. You, as an individual, have no obligation to earn a NIM; you just need to pick the duration that fits your own goal. SVB no longer had that choice. This is why every bank structurally carries interest rate risk, and why you should be cautious with mid-tier bank stocks (US community banks had $620 billion in unrealized losses at the end of 2022 - most of it still sitting there in 2026).

For individual investors: three paper-loss scenarios

Scenario 1: Hold to maturity
Paper loss disappears
If the government doesn't default and you don't sell, every intervening price swing is meaningless. You receive the full face value + coupons.
Scenario 2: Need cash mid-way
Real loss
Selling before maturity at a lower price = a real loss. This is the risk of long-dated bonds when you're not sure you can hold long enough.
Scenario 3: Bought via ETF
Real loss via NAV
ETFs like TLT and AGG have no fixed maturity date - they continuously roll old bonds into new ones. A falling NAV is a real drop in what you hold, with no "maturity" to wait out back to par.

Scenario 3 - buying through an ETF like TLT - is the biggest trap for individual investors. An ETF has no "maturity back to par" the way an individual bond does: its NAV moves permanently with yield. TLT went from $171 in early 2020 to $82 by late 2023 - anyone who bought at the peak officially lost ~52%. Section 9 will dig into why buying a bond ETF is actually a speculative position, not saving.

4 · Berkshire's $397 Billion Cash Mountain - A Living Lesson In Duration

At the end of Q1 2026, Berkshire Hathaway reported $397 billion in cash and equivalents - the highest in the company's history, up from $373 billion at the end of 2025 (Bloomberg). This was the first quarter with Greg Abel as CEO after Buffett stepped back. The interesting part: over 90% of that sum is not sitting in a bank account, but held as T-bills - short-term US government bonds (4 weeks to 1 year).

Total cash + T-bills
$397B
Q1 2026, an all-time high for Berkshire. Has been a net seller of stocks for 14 consecutive quarters since Q3 2022.
Of which, T-bills
~$360B
Equivalent to 5-6% of the entire US T-bill market ($6,150 billion). Roughly double the prior year.
Versus the Fed (SOMA)
> $195B
Berkshire is holding more T-bills than the Federal Reserve itself. A private company, bigger than the central bank, in the short-term debt segment of the US government's own debt.
T-bill interest/year
~$14B
At an average yield of ~3.9% on $360B, Berkshire earns roughly $14 billion a year in interest - "risk-free money" double Coca-Cola's operating profit.

Why did Buffett choose T-bills, not the 10Y or 30Y?

The answer lies in the very concept of duration introduced above. During 2022-2023, as the Fed raised rates from 0% to 5.5%, investors holding long-term bonds saw paper losses of 30-50%. SVB collapsed because of it. Pension funds lost trillions. Berkshire lost not a single cent - because 100% of its bond portfolio was T-bills with maturities under one year:

  • The duration of a 4-week T-bill ≈ 0.08. Even if yield jumps 1 percentage point in a week, the price falls 0.08% - barely noticeable.
  • The duration of a 1-year T-bill ≈ 1. A 1-point yield rise drops the price ~1%. Still small.
  • The duration of a 10-year Treasury ≈ 8-9. The same yield jump means an 8-9% loss.
  • The duration of a 30-year Treasury ≈ 18-20. An 18-20% loss.

Buffett didn't "correctly predict" that the Fed would raise rates - he didn't need to. He chose T-bills precisely because they let him not have to predict. Every few weeks a T-bill matures, he gets cash back at face value, and buys a new T-bill at the current yield. Yield rises → interest rises. Yield falls → interest falls. He never loses a dime of principal. This is the strategy of "I don't know the future - and I don't need to."

"T-bills are the safest investment there is. We'd happily put $100 billion into the right deal - but if there isn't one, sitting in T-bills earning 4-5% is a perfectly fine thing to do... compared to overpaying for something that isn't worth it." - Warren Buffett, Berkshire shareholder meeting, 5/2024

A hard-to-believe scale

To picture how big Berkshire has become in the T-bill market:

Share of the US T-bill market ($6,150 billion total), Q1 2026
Berkshire alone holds ~5.9% of the entire T-bill market - a single private company, bigger than the Fed (3.2%) and bigger than most large money market funds. This is unprecedented in the history of the US debt market.
1% 2% 3% 4% 5% 6% Berkshire 5.9% · ~$360B Money market funds (top 5) ~4.5% · ~$280B Fed (SOMA) 3.2% · ~$195B JPMorgan (largest bank) ~1.5% · ~$95B % of the total US T-bill market
Lessons for the rest of us + the real difficulty of this strategy

Berkshire's story isn't about "Buffett being a genius at predicting rates." It's a simple principle:

  1. Uncertain about the direction of rates → choose short duration. T-bills at 4% carry essentially free, negligible paper-loss risk.
  2. "Cash" is also an investment decision. When stocks are richly valued, holding T-bills at 4% is a deliberate, active choice.
  3. Individuals in Vietnam/Singapore can copy this. US T-bills via Interactive Brokers/Schwab; SGS via DBS/POSB/UOB; Vietnamese treasury bonds via TVSI. Different scale, identical principle.

The real difficulty isn't analysis - it's psychology. Berkshire has been a net seller of stocks for 14 consecutive quarters. Over the same period, the S&P 500 rose from 3,800 (Q3 2022) to ~6,200 (5/2026) - a 63% gain. Berkshire is mechanically "losing" to the index. But Buffett still isn't buying. Ask yourself: could you tolerate earning "just" 4% for 14 straight quarters while your friend brags about a 60% gain from Nvidia? Most people can't. That's the real difficulty of discipline.

5 · What The Yield Curve Is

Every day, the market prices dozens of US government bonds with maturities ranging from 1 month to 30 years. Plot the yield of each maturity on a chart - maturity on the x-axis, yield on the y-axis - and you get the yield curve. It's a snapshot of the market's mood about the future.

The "normal" shape slopes upward: the longer you lend, the higher the interest. The logic: lending for 30 years is riskier than lending for 3 months (inflation could rise, the government could spend more recklessly, the dollar could lose value), so lenders demand a premium for accepting that longer horizon. That extra charge is called the term premium.

A normal, upward-sloping yield curve
Illustrative example: US Treasury yields by maturity in a healthy economic-cycle environment. A gentle upward slope, with each longer maturity paid an extra term premium.
5.0% 4.5% 4.0% 3.5% 3M 2Y 5Y 10Y 20Y 30Y 3.62% 3.88% 4.10% 4.38% 4.72% 4.85% yield (%) maturity

Four basic shapes and what they mean

normal
Normal / Upward sloping
3M 2Y 10Y 30Y
Long yield > short yield. The market believes the economy is growing steadily, inflation is reasonably anticipated, and the Fed doesn't need emergency policy. This is the state of affairs about 80% of the time in US history.
steep
Steep curve
3M 2Y 10Y 30Y
Spread > 200bp. Usually appears early in a recovery cycle - the Fed is still cutting rates (pulling the short end down), while growth and inflation expectations push the long end up. It can also be a bear steepener - worry over fiscal sustainability.
flat
Flat curve
3M 2Y 10Y 30Y
Spread ~ 0. A transitional state - the market is uncertain about direction. The Fed may be nearing the top of its hiking cycle. Usually precedes inversion.
inverted
Inverted curve
3M 2Y 10Y 30Y
Short yield > long yield. Unnatural - borrowing for 3 months costs more than borrowing for 10 years. The classic signal: the market is betting the Fed will have to cut rates sharply in the near future to rescue a slowing economy.

The 10Y-2Y spread: the most closely watched indicator

Among countless ways to measure the curve's slope, one number gets cited most often by the Fed, the IMF, and every investment fund: the spread between the 10-year and 2-year Treasury yields. Why this one was chosen:

  • The 2-year fairly closely reflects expectations for the Fed funds rate over the next 2 years (the Fed can cut/hike multiple times within that window).
  • The 10-year reflects long-term growth and inflation expectations.
  • The spread (10Y - 2Y) is essentially "the reward for lending for 8 extra years." When it goes negative, it means the market believes long-term growth plus inflation will be lower than what the Fed is currently holding short-term rates at - a sign that policy is too tight and a recession is coming.

The historical record is remarkably consistent: every US recession since 1969 has been preceded by a 10Y-2Y inversion, with an average lag of 15 months (as short as 6 months, as long as 23) (PrimeRates). There hasn't been a major "false positive" in 56 years.

6 · The 2022-2024 Inversion And Today's Normalization Moment

In July 2022, the Fed had raised rates from 0.25% to 2.5% over four months to fight post-COVID inflation. The 2-year yield jumped from 0.7% in early 2022 to 3.0%. The 10-year yield only rose from 1.5% to 2.9%. The 10Y-2Y spread fell to −0.06% - the first inversion.

The deepest point of the inversion came on 7/4/2023: −1.08% - the deepest since 1981. For 27 months, from 7/2022 through the end of 12/2024, the curve stayed inverted - the longest stretch on record.

US 10Y − 2Y spread, 1/2022 → 5/2026
When the line falls below the 0% mark (dashed), the curve is inverted. The red zone marks the 27-month inversion - the longest in history (7/2022 → 12/2024). In early 2025 the curve uninverted - history says recession typically follows 3-12 months later.
+2.0% +1.5% +1.0% 0.0% -0.5% -1.0% 1/22 1/23 1/24 1/25 1/26 5/26 inverted 7/2022 trough −1.08% (7/2023) uninverted 12/2024 +0.50% ← 27-month inversion (record) → spread (%)
10Y − 2Y spread inverted zone (< 0) normal zone (> 0)

What happens during and right after an inversion? For many months, not much - inflation cools slowly, GDP keeps growing, the labor market stays solid. Plenty of 2023-2024 commentary declared: "This time is different. Inversion isn't a recession signal anymore." That same sentence has been written before every major recession of the past 50 years.

The uninversion moment - the truly dangerous signal

Inversion itself isn't the most dangerous signal. Normalizing back after an inversion (uninversion) is. History over the last four instances:

Inversion endedRecession beganLag
8/19897/199011 months
12/20003/20013 months
6/200712/20076 months
5/20192/20209 months (COVID, possibly coincidental)
12/2024?counting

The mechanism: an inversion is the market issuing a warning. Uninversion is when that warning starts becoming reality - the Fed sees the economy weakening, starts cutting short-term rates, pulling the 2Y yield down faster than the 10Y. This "bull steepening" of the curve typically accompanies or immediately precedes an official recession.

Important note - two different kinds of uninversion.
Bull steepener - short-term yield falls faster than long-term (the Fed is cutting). This is the classic recession signal.
Bear steepener - long-term yield rises faster than short-term. Usually driven by fiscal concerns: the government is issuing too much debt, and investors demand a higher term premium to bear the inflation and currency-debasement risk. This isn't a recession signal but a signal of eroding confidence in debt sustainability.

Right now, in early 2026, there are signs of both - the Fed cut rates 100bp in 2025 (pulling the 2Y down), yet the 30Y yield keeps rising due to an uncontrolled US budget deficit (pushing the 30Y up). This is an unusual state - showing both a classic recession element and a fiscal-confidence element at once (Reuters).

7 · Comparing The US, Singapore, And Japan

Three markets, three completely different macro stories. For the same $1,000 face value (or equivalent), you get very different outcomes:

Metric (5/2026)US (UST)Singapore (SGS)Japan (JGB)
10-year yield4.38%2.14%2.53%
2-year yield3.88%~1.50%~0.80%
10Y-2Y spread+0.50%+0.64%+1.73%
Central bank policy rate4.00%n/a (MAS uses the exchange rate)0.75%
Government debt outstanding$31.4TS$1.3T¥1,044T (~$7T)
Debt/GDP~120%~170% (special case)~260%
Main creditorsForeign 30%, Fed 17%, Public 53%Mostly domesticBOJ 50%, banks/insurers 31%
RatingAA+ (S&P)AAAA+ (S&P)

The US: the world's most liquid market

US Treasuries are the base measuring stick of global finance - every other asset (stocks, corporate bonds, real estate) is priced at a discount relative to this "risk-free rate." Buyers come from all over the world: Japan ($1.225T - the largest single creditor), China ($694B - down 9% since early 2025), the UK, the EU, US pension funds. Liquidity is enormous - over $700 billion trades every day (Reuters).

The main risks for 2026 are two unusual issues:

  1. The US government is issuing too much debt ($2,000 billion deficit in 2026), pushing up the long-term term premium - the bear steepener currently underway.
  2. China continues trimming its holdings - down to $694B (1/2026) from $760B in early 2025. A large sudden reduction could spike the 10Y yield.

Why a Chinese Treasury sell-off is dangerous

Ordinary readers often struggle to picture what "China sold $66 billion in Treasuries" actually means. Why is that a big deal? Why does it connect to a mortgage rate in Hanoi, Apple's stock price, or the US budget? Picture it through a produce-market example - the same supply-demand mechanism.

The vegetable-market example: You're a farmer selling tomatoes. Every day the market has 100 kilos of buyers. You stand in the market selling 100 kilos of tomatoes at 20,000 VND/kilo - you sell out. Then one day you need to sell 200 kilos, but buyers still only want 100 kilos. To sell it all, what do you do? Cut the price - down to 15,000, or 12,000, or lower. The more tomatoes relative to buyers, the more the price has to fall to clear the market.

Treasuries work exactly the same way. The US sells debt - China, Japan, pension funds, the Fed buy it. When China stops buying and starts selling what it already holds, the amount of "tomatoes" in the market spikes. To sell it all, the Treasury price has to fall. And a falling Treasury price = rising yield (the see-saw from Section 2). So the more China sells, the higher US interest rates have to rise to attract other buyers.

The scale in numbers: China has trimmed its holdings from ~$1,300 billion in 2014 to $694 billion in early 2026 - a net sale of $600 billion over 12 years. The sharpest selling wave was 2022-2023: roughly $200 billion sold in 18 months, coinciding exactly with the 10Y yield jumping from 1.5% to 5%. Not the sole cause (the Fed's rate hikes contributed heavily too), but not a trivial factor either.

The concrete impact of +0.5% yield + why China can't fully "weaponize" this

When the US 10Y yield rises 0.5%, the effect ripples through three layers:

Layer 1: US budget
+$155B/year
The US issues ~$31T in marketable debt. That's equal to the entire federal education budget.
Layer 2: Mortgage rates
+~0.5%
A 30Y mortgage ≈ 10Y yield + 1.5-2%. A homebuyer with a $400K mortgage pays an extra $120/month = $43,200 over 30 years.
Layer 3: Stocks & real estate
−5 to −10%
The S&P 500 loses ~$3,000 billion in market cap for every 50bp move in yield.

A hypothetical scenario: if China sold $300 billion over 6 months → the 10Y yield rises 50-80bp → US mortgage rates rise accordingly, the S&P 500 loses 8-12%, the dollar strengthens against the yuan, and the US pays an extra $200 billion a year in interest. This is why the monthly TIC (Treasury International Capital) data makes stocks react within seconds of release.

So why doesn't China just sell everything at once? Two reasons:

  1. Selling means hurting yourself. Selling $200B fast → prices fall 5-10% → China itself loses $20-40B on what it still holds. The more it sells, the more prices fall, the more it loses.
  2. There's no large-enough alternative. Gold is too small, the euro hasn't been trusted since 2010-2012, the yen's market is smaller than the US's, Bitcoin is too volatile. This is what the US calls its "exorbitant privilege."

So China has chosen to reduce gradually and quietly: selling $5-10B a month. Over 12 years, $600B has left without triggering a crisis. The dangerous scenario: several large creditors cutting back at the same time, or a tension event (Taiwan) forcing a fast sale regardless of loss.

The thing most worth watching in 2026. Not China's holdings level specifically - but the bid-to-cover ratio at the monthly Treasury auctions (total bids/total issued). Historically, 2.5-3.0 is normal. The 7Y note auction on 4/29/2026 came in at 2.51 - the lowest of the year. If it drops below 2.0 for several consecutive auctions, that's the real signal: demand for Treasuries isn't keeping pace with the rate of issuance. That would be the true "US debt crisis" moment - not a headline saying "China sold $X billion."

Singapore: the lowest yield, a rare AAA

Singapore is one of just 11 countries in the world rated AAA by all three agencies (S&P, Moody's, Fitch). The logic: near-constant budget surpluses, foreign reserves over $400 billion, and debt issued mainly for investment purposes (funding CPF, Temasek) - not because the government is short of cash. For that reason, SGS yields are low.

A distinctive structure:

  • SGS bonds - international standard, 2-50 year maturities, semi-annual coupons, purchasable via SGX or MAS auctions.
  • T-bills - 6-month and 1-year, auctioned every two weeks. At the 2023 peak, T-bill yields hit 4.2%, pulling in a wave of savings.
  • Singapore Savings Bond (SSB) - a unique retail product. 10 years, step-up coupon, withdrawable anytime with no penalty. The 5/2026 issue (SBMAY26) has a 10-year average rate of 2.14%, starting at 1.40% in year one and reaching 2.96% by year ten (TheFinance.sg).

The SSB is special: it's the only case in the world where an individual investor can "withdraw penalty-free" from a government bond. The reason it exists: MAS deliberately built a long-term savings tool for ordinary citizens, not an investment vehicle for funds. The $200,000 per-person cap is set right around the threshold of a middle-class Singaporean's savings.

Japan: the strangest market in the world

JGBs form the world's second-largest government bond market (roughly $7,000 billion equivalent) - but they operate under rules the US doesn't have:

BOJ holdings, 9/2025
50.0%
The BOJ alone holds half of all outstanding Japanese government bonds. This ratio first fell below 50% as of 3/2026.
Foreigners
6.6%
Extremely low compared to the US's 30%. JGBs are held mainly by the Japanese themselves.
10Y yield, 4/2026
2.53%
A 29-year high. In 2020-2021 the yield was near zero. This 250bp jump has caused trillions of yen in paper losses across Japanese banks and insurers.

JGBs are unusual because for 13 years Japan ran a "Yield Curve Control" policy - the BOJ directly pinned the 10Y yield at 0%, then 0.25%, then 0.5%, buying without limit. The curve was essentially hand-drawn by the BOJ, not priced by the market.

At the end of 2024 the BOJ officially scrapped YCC. By early 2026, the BOJ had raised its policy rate to 0.75%. The 10Y yield jumped from 0.6% in early 2024 to today's 2.53%. The consequence: Japanese banks and insurers (holding 31% of all JGBs) are sitting on a mountain of paper losses. This is exactly what the IMF has warned about as "hidden balance sheet stress in Japanese financial sector."

The mountain of paper losses: long-term JGBs 2020-2026, a slow-motion disaster (chart + numbers)

Section 2 introduced the rule of thumb: price change ≈ −Duration × yield change. Long-dated JGBs are the textbook illustration of applying this to an extreme situation: starting yields near zero, extremely high duration, and the sharpest yield move in six years since World War II.

JGB yield by maturity, 1/2020 → 5/2026
Four yield curves for Japanese bonds across long maturities. In the first half (2020-2022), the BOJ pinned yields at historic lows via YCC. In the second half (2024-2026), after the BOJ let go, yields spiked to 2-3 decade highs. Over the same period, the price of old bonds fell by varying amounts depending on duration.
5.0% 4.0% 3.0% 2.0% 1.0% 0.0% 1/20 1/22 1/24 1/26 5/26 YCC: BOJ pins yield down BOJ lets go → yield spikes 10Y · 2.53% 20Y · 3.45% 30Y · 3.63% peak 4.21% (1/2026) 40Y · 3.90% yield (%)
10Y 20Y 30Y 40Y

The 40Y line (red) is the most astonishing phenomenon: from ~0.7% in 2020 to a peak of 4.21% in January 2026 - the highest since the 40-year maturity was first issued in 2007 (Bloomberg). A jump of roughly 3.5 percentage points, multiplied by the ~30 duration of a 40-year bond, produces a shocking number:

10Y JGB held since 2020
−18%
Yield 0% → 2.53%, duration ~9. A paper loss of nearly a fifth. Recovery to par requires waiting until 2030.
30Y JGB held since 2020
−45%
Yield 0.5% → 3.63%, duration ~22. A $1,000 face-value bond now trades around $550. Recovery to par requires waiting until 2050.
40Y JGB at the 1/2026 peak
−65%
Yield 0.7% → 4.21%, duration ~28-30. A 40-year bond issued by the BOJ in 2020 with a ¥10,000 face value traded, at one point, at a market price of only ~¥3,500.
Picturing the scale of this loss. Say a Japanese life insurer holds ¥1,000 billion of 30Y JGBs bought in 2020 - by early 2026, the market value is down to ¥550 billion. A loss of ¥450 billion on paper. If forced to sell (policyholders redeeming) or forced to mark-to-market (new accounting rules), that ¥450 billion becomes a real loss. In 4/2026 the IMF warned that Japanese insurers carry over ¥30 trillion in total unrealized losses - roughly $200 billion USD - "hidden" via amortized cost accounting, the same trick SVB once used. If any "Twitter run" were to hit Japan (low probability, but not zero), this would be the SVB scenario at ten times the scale.

Notably, since late 2025-early 2026, Japanese insurers and pension funds have started actively selling long-dated JGBs to limit their losses - and that very selling pushes yields even higher. A self-reinforcing fire sale spiral: the more losses, the more selling, the more yield spikes, the more losses. The BOJ has had to step in with limited purchases a few times in 1-2/2026. This is also why the market treats JGBs as an early warning signal for global risk - not because of Japan on its own, but because the stress transmits to US Treasuries through the yen carry trade channel.

Why JGBs matter to the whole world. Japan is the largest foreign holder of US debt ($1.225 trillion in Treasuries). The "yen carry trade" cycle has run for 25 years: Japanese investors borrow yen at 0%, convert to USD, and buy Treasuries yielding 4-5%. When the BOJ raises domestic rates, Japanese investors sell Treasuries to buy JGBs instead - creating selling pressure on the US 10Y. Throughout the second half of 2024, there were repeated episodes where the US 10Y spiked every time news of a BOJ rate hike emerged. This is one of the few mechanisms that genuinely transmits from Tokyo back to Washington - most of the time, the cycle runs the other way.

8 · Reading The Market's Mood Through The Curve

Three practical signals to take away, if you can only remember three things:

Signal 1 - The 10Y-2Y spread hits zero or goes negative

The market is saying: "Short-term rates are too high relative to economic conditions over the next 2-10 years. The Fed will have to cut." Inversion doesn't mean recession tomorrow. It's usually 6-23 months out. During an inversion, the economy and stocks can still keep rising (as in 2023-2024). But anyone buying stocks at this point should understand they're in the late stage of the cycle.

Signal 2 - The curve normalizes back after an inversion (uninversion)

This is a more serious warning. When the spread moves from −1% back to +0.5%, it means the Fed has seen something worrying enough to start cutting rates to try to rescue the economy. An official recession usually begins 3-12 months after uninversion. The US uninverted in 12/2024. As of now, that's 16 months and counting.

Signal 3 - A bear steepener (long yield spikes on its own)

When the 30Y spikes while the 2Y stays flat or falls, this isn't a recovery signal - it's a signal of eroding fiscal confidence. The market is saying: "I still believe the Fed can control inflation in the short run. But I don't believe the government will spend responsibly over the long run." This is exactly what's playing out in the US and Japan between 2025-2026 - and it's the thing most worth watching over the coming year.

"The yield curve doesn't cause recessions. It's simply how the market expresses something ordinary analysis tends to overlook: monetary policy is tighter than the economy can bear. When that warning gets lifted - by the Fed cutting rates - it's usually because the warning was right." - Campbell Harvey, Duke University, the first to identify this relationship in 1986

9 · So What Should Individual Investors Do

This article isn't investment advice, but a few practical observations follow directly from the mechanics above:

  • Understand the maturity of the bond (ETF) you hold. AGG (US Aggregate) has a duration of ~6, TLT (20+ year) has a duration of ~17. For the same 1% yield move, AGG falls ~6%, TLT falls ~17%. "Bonds" are not one uniform asset class.
  • If you'll need liquidity within 1-3 years, T-bills or SSBs are safer than long bonds. Lower yield, but no large paper-loss risk if you need to withdraw.
  • Individual bonds are different from bond ETFs. An individual bond has a fixed maturity date - hold to that date and you get par back. A bond ETF has no "maturity back to par" - its NAV moves permanently. Buying a bond ETF is a bet on the direction of yield, not safe saving (details below).
  • The 10Y-2Y spread is a warning light, not a countdown timer. Inversion doesn't mean sell stocks tomorrow. Uninversion doesn't mean liquidate today. But they are signals to tilt more defensive - raise cash, trim cyclical stocks, extend bond duration if you think the Fed will cut.
  • For Vietnamese & Singaporean readers: US government bonds carry currency risk. When the dollar weakens (DXY falls), the VND/SGD-converted value of a Treasury drops - even though the USD yield is high. This is a risk Americans don't face. It makes the most sense for people who expect to spend USD in the future (children studying in the US, travel, export businesses).

Buying a bond ETF is a bet on the direction of yield - not safe saving

This is the most costly misunderstanding for individual investors, especially those newly shifting from stocks into "diversification" through bonds. They hear the pitch - "US government bonds are risk-free, diversify your portfolio" - and buy TLT, BND, AGG. When yield rises, the portfolio loses 20-50% - and they're baffled, thinking, "But I bought a safe asset, didn't I?"

The problem lies in a structural difference between two products that look identical on the surface:

Individual bond
Has a maturity date → returns to par
You buy a UST 10Y issued 5/2026 with a 4.375% coupon. Whether the market price falls to $700 or jumps to $1,200 over the next 10 years, in 5/2036 you receive exactly $1,000 face value back. Any loss in between is purely on paper. This is genuinely "interest-bearing savings."
Bond ETF
No maturity date → NAV drifts forever
TLT holds a basket of 20+ year bonds. When a bond in the basket ages below 20 years, the ETF sells it (at the market price, possibly at a loss) and buys a new 30-year bond. It never "matures." Your NAV is a continuous market value, with no recovery point.

The result: TLT 2020-2024, a disaster anyone can measure

TLT vs. an individual 30Y UST bought in 2020 - same underlying asset, two different fates
At the same moment in 8/2020, two investors each put $171 into two different 30-year Treasury products. Six years later, one is back at par; the other is still permanently down ~50%.
$200 $150 $100 $50 8/2020 2022 2023 2024 2025 5/2026 $171 (bought 8/2020) trough $82 (10/2023) Individual UST: back to par TLT: still down −45% price ($)
Individual 30Y UST (hypothetical) TLT (20+ year Treasury ETF)

These two lines move together from 2020-2023 because the market price of the individual bond and TLT track each other closely. The gap starts to open from 2024: the individual bond gradually returns to its $1,000 face value as maturity approaches - an unbreakable mathematical rule. TLT has no such rule. TLT's NAV only reflects the current market price of today's 20+ year bonds - and the bonds in the basket are continuously replaced at whatever the prevailing yield is.

This is the core point: when you buy TLT, you are not "lending the US government money for 30 years." You are betting that the 30-year yield will fall in the near future. If yield falls → the price of bonds in the basket rises → NAV rises → you profit. If yield rises → the opposite → you lose. This is the mechanism of a long-duration speculative position - not saving. The same mechanism as buying an option, buying a commodity, buying a stock - just a different underlying.
Four scenarios where a bond ETF makes sense vs. is a mistake + a duration reference table
Your goalIs a bond ETF right?Reason
"Safe 10-year savings" WRONG Buy an individual bond or a bond ladder instead. An ETF has no guarantee of returning to par. It can lose 30-50% permanently.
"Betting on aggressive Fed cuts" RIGHT TLT is exactly the tool for this. This is macro speculation, not saving.
"Balancing a stock portfolio" CONDITIONAL True when stock-bond correlation is negative. Since 2022 correlation has turned positive: the classic 60/40 has broken down.
"Earning 4% short-term interest" RIGHT (BIL/SGOV) Duration ≈ 0, NAV essentially flat. This is how Berkshire does it. Don't confuse this with TLT.

Quick rule - match duration to your goal (check on the ETF's page before buying):

  • Duration < 1 (BIL, SGOV, SHV) → interest-bearing cash. No real bet involved.
  • Duration 3-7 (AGG, BND) → a mix of income and partial duration speculation. Every 1% drop in yield = +3-7% NAV.
  • Duration 15-20 (TLT, EDV, ZROZ) → pure speculation. Not suited to "I want safety."

In 2022-2023, plenty of individual investors took heavy losses buying TLT because it sounded "safe" (Treasury + a convenient ETF wrapper) but was actually a duration-17 speculative position - a ~52% loss when the Fed hiked 525bp. Same underlying asset, two completely different experiences.

"The more convenient a financial product, the easier it hides the real risk underneath. An ETF lets you 'buy US bonds' with a single click - and conceals the fact that you've just opened a duration-17 speculative position. Convenient is not the same as safe. Two different concepts." - A general rule for any financial product with the word "ETF" in it

One final thought

Government bonds are the most boring asset there is, yet they're the foundation of the entire financial system. Every mortgaged house in Hanoi is priced with reference to the US 30Y Treasury. Every stock on the VN-Index has a discount rate that traces back to the 10Y. When you watch stocks tumble and ask "why" - the reason often traces back to some point on the curve that shifted overnight, unnoticed by almost anyone.

Learning to read the government bond yield curve is like learning to read a body's temperature. It doesn't diagnose the disease, but it tells you when something isn't normal - far earlier than when symptoms actually appear. For 56 years, every US recession has come with that warning. This time, the warning lasted 27 months and has just been lifted. The countdown is running.

Read next

More from the shelf

Jul 23, 2026When There's Too Much Welfare, the Nation Starts to WeakenApr 23, 2026A Singapore Resident's Map to US Markets 2026Apr 22, 2026Singapore 2026 - Mapping Savings & Investment as Rates SlideJul 13, 2026Global Public Debt Is Rising: Nobody Goes Bankrupt, So Who Pays?

Pass it on

If it found you, share it kindly

XEmail

03 Discussion

Leave a note

A considered space for questions, counterpoints, and useful additions. Civil, on-topic, signed.

Reader notes

...

Loading notes...