How will AI investments drive up Treasury yields? In a new post on the blog, I look at the channels through which AI investments will increase Treasury Yields. Blog: The AI investment boom is expected to increase yields through the traditional channel of increased bond supply and the general equilibrium channel of intertemporal consumption smoothing. The supple channel depends critically on the amount and maturity of AI debt issuance, and the elasticity of demand for duration risk. Although the size of debt issuance needed to finance the AI infrastructure boom is massive, recent evidence suggests shifts in bond market participation has made the market less sensitive to supply shocks. Overall, the expected change in yield is highly uncertain but to-date the evidence suggests a notable amount of the AI boom is crowding out other investments and therefore muting its impact on equilibrium yields.

Today's the first day of my fall-term Financial Crisis class and the WSJ quotes me in their article about the bond market sell-off. The research I reference isn't mine - I just teach it to my class - but it's nice timing for my intro lecture claim that knowledge about financial crises is always a timely topic.

Should we worry that Ai agents will cause a bank run? Maybe! Torsten Slok has an interesting note worrying about increased competition for deposits potentially causing bank runs. Slok’s worry is households will rely on AI agents to allocate deposits to the highest yielding bank. The result will either be runs on the deposits of banks with uncompetitive rates or (more likely IMO) new equilibrium deposit rates that are uniform across banks and much closer to wholesale funding rates (close to SOFR). Obviously, the run scenario is bad but it's unlikely, in my opinion, that technology for moving deposits will advance faster than banks' ability to raise deposit rates. Nonetheless, we should worry about the run possibility. Even if runs don’t occur, the increase in competition for deposits will depress lending, investment and economic growth. Deposit beta – the change in bank deposit rates as policy rates change - is a key input into both banks' internal risk models and the Fed’s Stress Test models (the PPNR estimates). These deposit betas range from .40-.55 at all DFAST banks and are even lower at the G-SIBs. Should deposit betas increase to .75-.85 these banks would be required to hold materially more capital and a large literature shows increased stress test capital requirements are passed through to lower lending and less economic growth (see Shahhosseini (2022) or Berrospide & Edge (2024)). Bank deposits pay below market rates and are notoriously sticky. If competition suddenly forced banks to pay deposit rates 100 bps closer to SOFR, we would see bank ROE fall between 30-50% for the US G-SIBs and more for regional and smaller less diversified banks. Again, there is a large literature showing unexpected shocks to bank profits result in credit rationing and slower economic growth (for example see Chava & Purnanandam (2011), Repullo, & Suarez (2013), Gropp et. al. (2024)) In equilibrium, a move to close-to-perfect competition for deposits would result in less lending, higher bank capital ratios, higher credit spreads on bank loans and a small probability of a costly bank run as we transition from the current sticky deposit regime to a competitive wholesale market for deposits.

Should the Fed raise rates if demand is AI driven and Ai expected returns are potentially so high that Ai investment is interest rate insensitive? Yes! Let's ignore whether the current price pressure comes from AI demand or an energy supply shock and work through the general equilibrium effects of an AI investment demand shock alone with a thought experiment where a new technology creates inflation… Imagine a new miracle machine is invented. This machine is costly to build - we have to divert goods and services from consumption today to investment to build these machines. But if we invest a few percent of our output today and build these machines, the economy will produce much more output (100%+ increase) a few years in the future. How does the unexpected invention of these machines alter market prices and inflation today? First these new machines increase scarcity today because they consume resources to build but don’t deliver more output until the future. The return to building these machines is so high that firm's issue lots of stocks and bonds offering high expected returns to attract resources from other investment options or consumption. This moves real resources from consumption today into investment today. Second, we have a general equilibrium conflict if interest rates don't change. Humans are risk averse. We generally think about risk aversion as preferring a constant stream of consumption over a more variable stream with the same average (Prefer $1000 to a coin flip that pays $0 or $2000). But the human preference for smooth consumption works across time as well as across states of the world. Because of these new machines everyone now expects to be much richer in the future, and they prefer to smooth consumption by selling claims to their future consumption and using those sales to buy more consumption today. Think about what would happen if I told a worker who makes $100K and has a $1m stock portfolio that I was raising his income to $1m per year starting five years from now. That worker would not want to consume the same as before this news. He would want to move some of that future consumption to today so he could smooth consumption over time. i.e. he would sell his stock portfolio and consume more today. The worker can do this because he is one man and he can find others who will buy his stock (they consume less today to purchase stock and consume more in the future because they have more stock). But in the aggregate, we can’t all consume more today! Remember these machines don’t increase production until the future. Output today hasn't increased and some of our consumption today must now be invested in building these machines. To clear the market the real interest rate must go up sufficiently high that workers see such a high real return on savings that they are willing to work and consume less today even knowing that they will be very rich in the future. i.e. the return must be high enough that they are happy with a variable consumption stream. The general equilibrium effect is why interest rates must rise to finance the investment boom and the real increase in rates is made even worse by the desire of workers to smooth consumption by selling claims (bonds) to future consumption. Bonds aren’t time machines! They don’t move output across time. You and I could use a bond to smooth our consumption but in the aggregate we can’t all consume more today so rates have to rise until people are happy consuming much less today than in the future. So now we know real interest rates have to rise. How do we get there? We could just let rates rise. But if the short end is an administered rate (like Fed Funds) what would happen if the fed kept the rate fixed? The real rate has to increase but the nominal rate can’t move. Therefore, we would need expected deflation in the future or some other asset to offer much higher returns. Therefore, we need long bond prices to fall a lot, or we need the price of goods and services today to move up a lot relative to their prices in the future (or a combination of the two). If the Fed keeps the nominal short rate fixed and everyone wants to consume more today, the price of goods and services today must rise relative to the price of stocks and bonds or the price of goods and services in the future. i.e. inflation or long bond rout or a combination of both. We saw these dynamics during COVID when the government said don’t work and also gave sufficient income support that almost no one was poorer. People wanted to consume the same, but output was way down so the price of goods had to rise to clear the market. If the FED wants price stability they can arrest some of the price increase by increasing the nominal interest rate. This will make consumers that expect AI to make them much richer in the future happy to forgo more consumption today and prevent the rationing of today’s scarce resources via higher prices. This is a thought experiment to illustrate how interest rates increase can arrest inflation even when the demand for AI investment is so strong that it remains virtually unchanged in the face of the rate increase. In my opinion, there’s definitely some of these general equilibrium effect in play. We see it in security issuance to fund AI investment and the diversion of resources. But we also have a good old fashion energy shock as well. Whether we should raise interest rates because of that is another question.

Once again Treasury did not receive enough offers to fill their expanded buyback auction. Today was the first expanded size buyback auction for the ultra-long end of the curve (20y-30y). Treasury offered to buy back up to $6B – an increase from the $2B offer at previous reverse auctions - but did not receive enough offers to fill the $6B allotment. The Treasury received 10.5B offers but only deemed $4.1B of these bids acceptable. The $10.5B was a ~50% reduction in offers from the previous two buyback auctions for this maturity bucket. Remember these are reverse auctions to purchase off-the-run issues. The Treasury uses a non-disclosed model to determine what an acceptable liquidity premium is. Basically, they have the fitted yield curve from the on-the-run issues and a model of how much is a "fair" discount for off-the-run issues. If someone tenders a bid considered fair, they get filled (or put on the demand curve if the number of fair bids exceeds the auction cap). Treasury did not receive enough fair bids that they deemed “fair” today to buy back the entire $6B offered. This is becoming a pattern and a good sign that dealers do not fear allocating balance sheet capacity to off-the-run Treasuries. The first expanded buyback auction was held on 2 weeks ago when Treasury offered to buy back $6B in the 10-20y maturity bucket but did not receive enough acceptable bids to fill the entire $6B offer. Treasury also offered to buy back $4B last week in the 7-10year maturity bucket and only received $2.4B in acceptable bids. Although they didn’t fill the whole allotment, Treasury did buy back more than twice as much as previous 20-30y buyback auctions when the cap was only $2B. The previous two buybacks for this longer (20y+) maturity bucket both filled their $2B cap with $19.9B of offers in August and $21.9B in July. We don’t know how many of these previous auction offers were at acceptable (fair) bids beyond the fact that at least $2B were accepted at each auction.

Strong economic news and hawkish comments by Barr have pushed up the Fed Futures projected rate path. The futures market is now pricing a 71% chance of a hike in October and a 93% chance of at least one hike and 55% chance of two hikes by year end. The whole curve has shifted up. The futures implied odds of 4 or more hikes over the next year is now at 53%, up 13 percentage points since yesterday. Going into the inflation reports two weeks ago the market was forecasting the rate cycle would top out at 3 hikes. Now the most likely forecast from futures is 5 hikes (including last week) and the odds of 6 or more hikes in the cycle is 22%. This is a huge deviation from the FOMC's SEP released just a week ago. We did receive stronger economic data since then, but risk aversion is very likely driving a material amount of the increase in future break-even rates. The embedded tweet below explains why the market implied rates overstate true odds of a hike in cycles such as these...

Two things... If we go back over the past year, expectations of the policy path are certainly increasing the yield on the 10-year. However, regardless of model the change in the expected policy path only accounts for about half to change in the 10-year over that period. The policy path accounts for even less over the past 3 months and almost none of the change from late June to Jackson Hole when yields rose notably. It's because yields increased without a change in the forward path that so many commentators have looked for alternative explanations. Personally, I think the most likely explanation is just uncertainty about the future path was/is very high the past few month and those holding long duration demand more return because of this. Note this is not inconsistent with the findings that most of the movements in yield occur when information about the path arrives (data and speeches) nor is it inconsistent with arguments that large deficits or other supply worries matter. All of these explanations would be expected to generate most of the moves in yield around the arrival of news. second point is we can't even reject the worries about the deficit or inflation debt spirals with this data or the fact that the deficit has been known for a long time. It's an empirical fact that yield spikes due to sovereign debt runs resemble a Hemingway character going bankruptcy. Slowly for a long while and then all at once. The blog in the tweet embedded below has a review of these drivers and links to the literature on debt runs where everyone knows the debt is large and unsustainable, yet nothing happens to yield until some seemingly small tipping point. To be clear, I don't think a debt run is a worry. I think we just have a good old fashioned uncertain tightening cycle with a lot of duration risk driving investor risk aversion upwards.

Are we in "backwards land" where bonds will be poor protection during the next recession or stock market crash? In the blog I look at why bonds have traditionally acted as good portfolio insurance and what conditions are needed for that to continue to be true. Blog:

Does the Market believe the FOMC dot plot? Fed Fund's futures imply many more rate increases than the FOMC's SEP projections. In a new post on the blog, I break down what is driving the gap between market implied rates and FOMC communications. Blog: The difference is due to an abnormally high risk-premium. There is a large academic literature showing this premium varies over time and should overstate expected rate increases in the current economic climate. That said, the market prices carry important information about the risks of policy errors and should neither be ignored or rewarded by the FOMC.

When the FOMC announces I will be looking at the SEP projections. In June the committee forecast y/y headline PCE of 3.6 with a core PCE of 3.3 for 2026. Given the June expectations, the committee felt a 2026 year-end Fed Funds rate of 3.8 (median) 3.6-4.1 (central tendency) followed by a rate cut in 2027 was appropriate. Y/Y PCE is currently in-line with the committees' expectations but given the energy shock since last data headline PCE for 2026 will almost certainly exceed the June forecast and core PCE may as well. Staff forecasts of PCE in the remainer of the year will show up in the new SEP projections and the committee's new appropriate policy rate path. I doubt the SEP rate projections will rise as much as the SOFR curve has been projecting (others may agree as the SOFR gave back quite a bit so far today) and if the SEP projections for 2027 come in below the current SOFR curve I expect a rally in both equities and bonds.

It's FOMC eve and the market is calm. The Oct SOFR midcurve (dec27 contract) options have a risk reversal of -.05. Not a lot of fear that the SEP dots or the statement will move expectations about where the tightening cycle ends. The S&P500 ES options the expire tomorrow have a VIX of 19.1% which is fairly low for a 1-day VIX before a big economic news event. Likewise, tail risk is low with a risk-reversal of -2.8 which is around the 50th percentile over the past year but is notably low for the eve of a key economic report. Investors aren't paying up for tail risk ahead of the FOMC announcement.

Update on rate expectations... The FF futures probability of a rate hike this week is now at 93%. That is up from 59% a week ago. Not only have the odds of a September hike increased, but the expectations for the total number of hikes shifted as well. Both the FF futures and the SOFR options now imply the Fed will hike between 75-100 bps and the cycle will stop by June 27. This is a change from market prices in the hour after the CPi report (see below). Right after the report the timing of hikes moved forward but the terminal level of the hiking cycle was largely unchanged. That has changed as investors digested the report and we received negative oil news over the weekend. The odds of 2 or fewer hikes has now declined over 30% since a week ago and the right tail (5 or more hikes) has added probability. Interestingly, the difference between SOFR options implied June27 rate and the FF implied rate is now much more aligned suggesting there's been an expectations shift rather than an increase in risk premium.

We're starting to get a little bit of concern in the stock market. We're still a long way from fear but the ES 25-delta risk-reversal for 9/15 is -2.8 which is ~70th percentile. Option buyers aren't concerned but the risk reversal has been above the 90th percentile many days lately. Also, the post-FOMC 9/16 risk reversal is -4.2 which is ~30th percentile*. That is still low for an important FOMC data day, but traders are willing to pay up a little for downside protection around the announcement.

What's driving the recent increase in global bond yields? Sovereign yields have increased notably in the past 10 weeks. Investors are concerned about inflation and deficits but have either become notable worse since June? In a new post on the blog, I look at the theory of sovereign debt runs where equalibrium yield depends on investors' beliefs about other investors' behaviors. These models predict coordinated selling across many bond markets at once. See:

The CPI came in hot enough to move Fed Fund Future implied odds of a September hike to 85%. However, the inflation reading was cool enough that the market expectations for the end of the tightening cycle didn't move much. The FF future implied odds for year-end remained centered around 2 hikes and the odds of more than two hikes by year end increased by only 2% while the odds of 1 or fewer hikes declined by just under 5%. Future prices still see 3 hikes by mid-2027 as the likely stopping point and today's CPI report actually lowered uncertainty about that estimate. The FOMC rarely surprises the market at a regular scheduled meeting. If they were to leave rates unchanged next week this would be the biggest "surprise" at a schedule meeting since 1994 when the Fed held rates fixed despite the market pricing in a rate hike.

The Treasury did not receive enough acceptable bids to fill the entire $6B buyback this afternoon. These are reverse auctions to purchase off-the-run issues. The Treasury uses a non-disclosed model to determine what an acceptable liquidity premium is. Basically, they have the fitted yield curve from the on-the-run issues and a model of how much is a "fair" discount for off-the-run issues. If someone tenders a bid considered fair, they get filled (or put on the demand curve if the number of fair bids exceeds the auction cap). Treasury did not receive enough fair bids today to buy back the entire $6B offered. Higher yields are encouraging dealers to hold on to these bonds. Previous auctions for this maturity bucket had smaller $2B offers and were filled with tender offers of 16.3B and 15.7B in July and 7.4B in august. We don't know how many of these offers were at "fair" prices, but the Treasury knows and likely took this information into account when they set the cap at $6B. This is an encouraging auction with respect to dealer balance sheet capacity. At these yields dealers aren't willing to lower their price much to remove the more illiquid bonds from their balance sheets. If you think one reason bond yields are rising is bondholders are playing a global game of run risk, this is an encouraging public signal.