The bond this is all about
Oracle issued 6.700% senior notes due 2056 on 2 February 2026 at 99.820, yielding 6.714%. On 25 September 2026 they are quoted near 84.25, a yield of about 8.11%.
Buy $12,000 of face, which is twelve bonds, and you paid $11,978. Today that is $10,110 of market value plus $402 of coupon and about $101 of accrued interest, so $10,633 against $11,978, a loss of about 11% on an investment-grade credit in under eight months.
Two things about that are not what people assume.
It was not a slow bleed. Fund filings mark the bond at 94.13 on 30 June 2026, and at 96.31 a month before that. Roughly ten of the sixteen points of price loss happened in July, August and September.
The market did not do it. Over the same stretch the investment-grade index went from about 74bp to 79bp. The market widened about 5bp while this one bond widened about 83bp. Whatever this is, it is specific.
The move decomposes to roughly +57bp of 30-year Treasury (4.90% to 5.47%) and +83bp of credit spread (181bp to 264bp). So most of it is the market repricing Oracle, and a meaningful minority is duration.
So what happens next? Nobody knows. But four bonds from the past show the three shapes this can take, and unlike a scenario chart they actually happened.
Case 1. The rate trap, and it pays you first
Amazon 2.500% notes due 2050. Issued 3 June 2020 at 98.977 to yield 2.549%.
| Date | Clean price | Funds confirming |
|---|---|---|
| 3 Jun 2020, issue | 98.977 | prospectus |
| 31 Jul 2020 | 109.050 | 32 |
| 31 Oct 2020 | 99.106 | 24 |
| 31 Mar 2021 | 89.493 | 83 |
| 31 Dec 2021 | 95.173 | 2 |
| 30 Sep 2022 | 62.375 | 2 |
| 30 Sep 2023 | 58.496 | 2 |
| 30 Jun 2026 | 57.938 | 2 |
$10,000 of face cost $9,897.70. Twelve coupons of $125 have paid $1,500. The bonds are worth $5,793.80 plus $18.75 of accrued. Total $7,312.55, a return of −26.1% over six years. Price only, −41.5%.
Amazon's credit was never impaired. Its rating did not go down and its cash pile went up. Every dollar of that loss is duration.
Apple's 2.650% of 2050, issued three weeks earlier at 98.571, did −21.4% including coupons and −37.9% on price. Two different issuers, the same outcome, so nothing about this is Amazon-specific.
The part that matters, and the reason this is the most useful case on the page: the Amazon bond was up more than 10 points inside two months before it was down forty. It printed 109.05 on 31 July 2020. Anyone who bought the new issue was sitting on a 10% gain by the summer and a 40% loss by 2022.
That July spike is real and it reconciles: the 30-year Treasury went 1.41% on 30 June to 1.20% on 31 July to 1.49% on 31 August. On roughly twenty years of duration, that is the whole round trip.
The rate trap does not announce itself. It pays you first.
Case 2. The crisis coupon that paid twice
Carnival 11.500% first-priority secured notes due 2023. Issued 8 April 2020 at 99.000, yielding about 11.91%, secured on the ships.
| Date | Clean price | Funds confirming |
|---|---|---|
| 8 Apr 2020, issue | 99.000 | prospectus |
| 30 Apr 2020 | 104.484 | 60 |
| 30 Jun 2020 | 108.244 | 89 |
| 31 Dec 2020 | 115.632 | 82 |
| 31 Mar 2021 | 114.625 | 70 |
| 30 Sep 2021 | 111.644 | 115 |
Carnival tendered for them in July 2021 at 114.25 per 100, plus accrued. The rest were redeemed with a term loan that October.
$10,000 of face cost $9,900. Interest over the 15.6 months to settlement came to $1,495. The tender paid $11,425. Total $12,920, a return of +30.5%.
Even if you missed the new issue and bought three weeks later at 104.484, you still made +23.0%.
This is the case people have in mind when they buy a fat coupon in a frightening moment, and it is real. The distinction that matters: Carnival paid you for a risk that did not materialise. Oracle's 6.700% pays a fat coupon for a risk that is still in front of it.
Case 3. The floor, and what happens to a bond the day it stops being a bond
WorldCom 8.250% notes due 2031. Issued 9 May 2001 at 98.098. Investment grade. Unsecured. Maturities out to 2031, 99.8% of the company's debt fixed-rate. Structurally, the same seat as an Oracle long bond.
| Date | Price | Funds confirming | $10,000 face |
|---|---|---|---|
| 9 May 2001, issue | 98.098 | prospectus | $9,810 |
| 31 Dec 2001 | 105.71 | 66 | $10,571 |
| 31 Mar 2002, SEC inquiry | 79.52 to 82.21 | 11 | about $8,000 |
| 30 Apr 2002, CEO resigns | 44.00 | 17 | $4,400 |
| 30 Jun 2002, fraud disclosed | 15.14 | 28 | $1,514 |
| 30 Sep 2002, trough | 12.25 | 2 | $1,225 |
| 31 Dec 2002 | 24.00 | 23 | $2,400 |
| 31 Dec 2003 | 33.75 | 13 | $3,375 |
It went from +5.7% to −88% in nine months, and never missed a coupon on the way down.
Here is the finding that no scenario chart can give you. WorldCom issued a 7.500% tranche due 2011 in the same deal. Before default the two traded apart, as they should: at the end of 2001 the 2031s were 105.71 and the 2011s 102.65, a twenty-year maturity gap priced as duration.
From 30 June 2002 onward the two bonds print at exactly the same price at every single date: 15.00, 13.25, 24.00, 29.75, 33.75.
The day a bond defaults, its maturity and its coupon stop mattering. It is no longer a stream of payments, it is a claim in a class, and every claim in that class is worth the same. That is the whole lesson of default in two rows.
The plan of reorganisation, filed with the SEC, gave senior noteholders new MCI notes at .357 of the allowed claim. Through the plan, $10,000 of face returned about $4,449 including the two coupons received, a loss of 54.6%. A holder who sold at the September 2002 trough got 12.25 and gave up roughly two thirds of what the plan eventually paid.
One caution, because the number is often quoted carelessly: 35.7 cents was the notes election. Holders could instead take 14.28 shares per $1,000, which implied $25 a share, and MCI traded well below that. What the stock election was actually worth is not something we can verify, so we do not assert it.
What the three cases say together
A long bond can lose forty percent with the credit untouched, because that is duration. It can gain thirty percent in fifteen months because the crisis you were paid for did not arrive. And it can lose nearly everything while paying every coupon right up until it does not.
The 6.700% coupon on the Oracle bond is not a yield you collect while nothing happens. It is the price the market charges for the third outcome being possible.
Hold it to 2056 and the price stops mattering: 59 more coupons and the face back. Trade it and the price is the entire investment. Those are two different investments in the same piece of paper, and the coupon is what you are paid for the difference.
Educational only. Not advice, and not a forecast.
A note on where these prices come from
Historical corporate bond prices are supposed to be hard to get. They are not, if you look in the right place.
Every US mutual fund and ETF files its portfolio holdings with the SEC, and each holding carries both the fair value and the principal amount. Divide one by the other and you have that bond's price on that date, marked by a fund's own pricing service. Search EDGAR for a CUSIP and you get every fund that held it. The Carnival CUSIP alone appears in 1,598 filings.
Cross-confirm each date across every fund that held it and you get something better than a single quote: a distribution. On most dates here the funds agree inside half a point. Where they do not, this page says so.
The same trick works before Form N-PORT existed, on the N-30D filings of 2001 and 2002, which is how the WorldCom path above was built.