On 13 of 16 days in 2026 when the Nasdaq Composite fell at least 1.5%, the 10-year Treasury yield rose. George Saravelos says an AI blowup would send money into bonds. This year's pattern would have to reverse.
Deutsche Bank's George Saravelos says investors have turned too gloomy on bonds. Bloomberg reported that his Friday note says the swing to deeply negative sentiment on fixed income "has gone too far". He argues the market underprices an artificial intelligence (AI) blowup that triggers a rush into bonds.
Saravelos is right that investors are deeply negative on bonds. A rush into Treasuries still needs stocks and bonds to move in opposite directions again. In 2026 equity losses and higher yields have mostly arrived together. His call therefore depends on a change in how the two markets behave. This year's record does not yet support that change.
The selloff-day test
The Nasdaq Composite, a broad stock index, is a rough proxy for AI equity risk. The calculation checks the 10 year Treasury yield, the return on a government loan lasting ten years, on every day the index fell by at least 1.5%. A yield rise means bonds fell alongside stocks and gave no cover. The count runs through Friday 9 October.
Using Nasdaq Composite data from the Federal Reserve Bank of St. Louis and Treasury's daily par yield data, there were 16 such days this year. The yield rose on 13 of them and fell on three. The latest session lifted the Nasdaq 0.64% to 27,366.17, so it adds no sell off day.
The comparison with ordinary trading days matters. The yield rose on 102 of the 195 trading days of this year, which is 52%. That is close to a coin flip. Higher yields occurred on selloff days about 29 percentage points more often than usual.
The worst Nasdaq day of the year shows the pattern. On 5 June the index lost 4.2%. The yield rose 8 basis points, where a basis point is one hundredth of a percentage point. Across all 16 days the average rise was about 4 basis points.
Last year looked different. On 10 October 2025 the Nasdaq fell 3.6% and the yield dropped 9 basis points. Investors who held Treasuries as insurance were paid while stocks fell. The hedge did its job that day.
Applying the same calculation to the whole of last year gives a different result. The index lost 1.5% or more on 27 days. Yields fell on 15 of them, or 56%. This year only 19% of such days brought a fall, so the old cover has faded.
The pattern repeats across the year. All five sell off days in March came with higher yields. Since 5 June the yield has risen on five of six such days. The one fall was a single basis point on 23 June.
Treasuries therefore failed as insurance on most large down days this year. Daily Nasdaq returns and yield changes have a correlation, a measure of how closely two series move together, of about minus 0.38 in 2026. Stock losses and yield rises tend to travel together, so Saravelos's call needs that link to reverse.
Why the hedge has failed
Bonds hedge stocks when a shock hits growth. Investors then expect lower interest rates, so they buy bonds and yields fall. Bonds fail when a shock hits the price of long term money. Stocks and bonds are then marked down together. A stretch in which the real yield, the return on a bond after expected inflation, climbed 100 basis points fits the second case better than the first.
Stocks kept rising through the dips. The Nasdaq itself ended last year at 23,242 and sits about 18% higher. Bonds kept losing over the same stretch, as the 10 year yield rose from 4.18% to 5.24%. The selloffs were dips inside a rising market, which is the opposite of a flight to safety.
The cause appears to be the price of holding long bonds: the term premium, or extra yield investors demand for lending for ten years. The comparisons that follow run from 2 September 2025 to 8 October 2026. The term premium series ends a few days earlier.
The New York Fed's term premium estimate rose from 0.49% to 1.08% over that window. The series is published with a lag, so 2 October 2026 is its latest reading. The premium has more than doubled in a year, which is what a bond market that no longer hedges would look like.
The inflation protected yield on 10 year bonds measures that real return. Over the same window it rose from 1.87% to 2.87%. That is a gain of 100 basis points. The series runs to 8 October, the latest day available.
Breakeven inflation, the market's inflation forecast, went from 2.41% to 2.35%, so it contributed nothing to the rise. Clients told Saravelos that AI is a "big driver of higher yields and the US Treasury has lost control of the long-end". The long end means bonds with the longest maturities.
Real yield plus breakeven inflation gives the nominal yield. On 8 October that is 5.22%. The nominal yield was 4.28% at the start of the window. So the whole 94 basis point rise is real yield, with breakeven inflation 6 basis points lower.
The case for a reversal
The strongest case for Saravelos's call is positioning: how investors are already placed. If most investors already expect bond losses, the next shock that sends them out of stocks can pull yields down fast. 2026 has such days. On 5 February the yield fell 8 basis points, and on 12 February it fell 9.
Yet the pattern can flip back within weeks. On 20 March the Nasdaq fell 2.0% and the yield rose 14 basis points. That was the biggest rise on any of the 16 sell off days. A hedge that paid in February had failed by March.
Supply and the November test
More bond supply is the main alternative explanation for rising yields. Treasury has not signalled an increase. Its August refunding statement, which sets out its quarterly borrowing plans, said it "anticipates maintaining nominal coupon and FRN auction sizes for at least the next several quarters". Coupon auctions sell bonds that pay fixed interest, and FRNs are floating rate notes.
The 20 year auction is $13 billion in most months and $16 billion in refunding months. Monthly coupon auctions total about $315 billion. That auction is therefore under 5% of the total, a small slice of the supply Treasury sells each month.
Treasury will most likely repeat that stance at its next quarterly refunding on 4 November, with auction sizes unchanged. That would leave real yields and the term premium as the explanation, rather than new issuance. On the evidence, the next Nasdaq drop of 2% or more is more likely to lift the 10 year yield than lower it. Saravelos's hedge stays unproven until such a day brings a fall of more than 5 basis points.