The Friday Five: Issue #14

I hope you all had a great week. This is Friday Five, Issue #14, and this was the week the bond market stopped being background noise.

On Wednesday a flash survey of purchasing managers came in far hotter than anyone forecast, with manufacturing at 57.0 against a 53.6 estimate and services at 58.7 against 55.8. The 10 year Treasury yield jumped 13 basis points that day. By Thursday's close the 30 year yield was at 5.461 percent, its highest close since June 2004, and the 10 year at 5.162 percent, the highest since July 2007. The Treasury bought back 6 billion dollars of long bonds on Wednesday trying to slow it down, and it did not work.

Here is the part worth your attention. A move like that is supposed to hit everything. It did not. It sorted the market into two piles and hit only one of them.

Anything owned for its yield was sold: Utilities down 4.23 percent, Real Estate down 2.07 percent, Financials down 2.38 percent, gold down 2.36 percent, long Treasuries down 2.25 percent. And Technology rose 2.70 percent while semiconductors rose 4.80 percent, in the same four sessions, against the same rising discount rate.

The S&P 500 finished up 0.70 percent. The equal weight version of the same index finished down 0.96 percent. The Dow fell, the Russell 2000 fell, and the gap between the index and the median stock was the widest in five weeks. Alright, let's get into it.

The Week in 60 Seconds

This Week in One Minute

The entire Treasury curve repriced to multiyear highs, and instead of hitting everything it sorted the market into two piles.

The 30 year Treasury yield closed at 5.461 percent on September 24, its highest close since June 2004, and the 10 year at 5.162 percent, its highest since July 2007. The whole curve moved: the 5 year rose 16.9 basis points, the 10 year 16.4 and the 30 year 13.0. What happened next is the interesting part. Every yield substitute was sold. Utilities fell 4.23 percent, Real Estate 2.07 percent, Financials 2.38 percent, gold 2.36 percent and long Treasuries 2.25 percent. And technology went up anyway, gaining 2.70 percent, with semiconductors up 4.80 percent. The S&P 500 rose 0.70 percent while the equal weight version fell 0.96 percent, the Dow fell 0.64 percent and the Russell 2000 fell 0.87 percent. The index went up and the median stock went down, by the widest margin in five weeks. None of this was panic: the VIX closed at 15.67.

The single thing to watch next week is the PCE price index on September 30 and the September employment report on October 2. Those two readings decide whether the long end keeps rising, and everything in this issue follows from that.

The Week Ahead

WHAT I EXPECT · 68% · MODERATE

I expect the market to keep paying for a small number of companies and selling everything else. From the September 25 close to the October 2 close I expect the cap weight S&P 500 to keep beating the equal weight version of the same index, because the thing driving the gap is arithmetic rather than mood. The whole Treasury curve repriced this week: the 30 year closed at 5.461 percent on September 24, its highest close since June 2004, and the 10 year at 5.162 percent, its highest since July 2007. A rising discount rate takes the most from businesses whose cash flows are fixed and the least from businesses whose cash flows are growing fast enough to outrun it, and that sorting does not stop while the rate is still rising. The gap reached 1.68 points this week against 0.92, 0.87, 1.06 and 0.92 in the four weeks before it. I put this at 68 percent rather than higher because the two readings that decide the rate path both land inside the window, the PCE price index on September 30 and the September employment report on October 2, and either one can end the move in a single session.

What would prove me wrong. The equal weight S&P 500 fund RSP beats the cap weight fund SPY by more than 1.0 percentage point from the September 25 close to the October 2 close, measured as the percentage change in RSP minus the percentage change in SPY.

Last week's call, graded. In Issue No. 13 I said the long duration bid would hold, and that Technology would keep beating the equal weight index from the September 18 close to the September 25 close, because the long end was staying contained. Technology beat the equal weight index by 3.65 percentage points, so the call survived comfortably. The reason I gave was wrong. I said it would happen because the long end was contained; instead the 30 year yield rose 13 basis points to a 22 year high and Technology won anyway. That tells me the long duration bid was never a discount rate bid at all. Separately, the risk I named did fire on its trigger and did not produce the damage I attached to it: I said a back up in long yields would take Generac, Zscaler and Natera down together, the 30 year closed at 5.461 percent against my 5.45 percent trigger, and then Generac fell 4.53 percent while Zscaler rose 8.78 percent and Natera rose 10.67 percent. They were never one bet. I grouped them because they looked alike, not because they trade alike.

By how much. I said I would be wrong if Technology trailed the equal weight index by more than 2 points. It beat it by 3.65, so the call cleared its own kill line by 5.65 points and the market never moved toward the trigger. The ledger tagged this a near miss, which is an artefact: it measures the closest approach including the window open reading, and for a spread that reading is zero by definition. This was a clean survival.

The biggest risk. The rate move stops being selective. So far a rising long end has only sorted the market, taking utilities, real estate, financials and gold while technology and the agent names went up anyway. Every decoupling like this ends at some level of rates, and nothing this week told me where that level is. The 30 year yield rose 13 basis points to a 22 year high while the S&P 500 still gained 0.70 percent and semiconductors gained 4.80 percent. A 6 billion dollar Treasury buyback on September 23 failed to slow the selling, and the market moved to price roughly a 70 percent chance of an October rate increase. I put the odds at 35 percent. The 30 year Treasury yield closing above 5.55 percent on or before October 9, 2026. That is about 9 basis points above the September 24 close of 5.461 percent, roughly one standard deviation of a week in this observable.

Market Risk Score

Market Risk Score: 62 / 100
Neutral, with the risk sitting in the bond market rather than in equities.

What supports the market. Equity volatility stayed low through the whole repricing, with the VIX closing at 15.67. Credit showed no stress, with high yield down only 0.81 percent against a 2.25 percent fall in long Treasuries. Technology and Communication Services, together a large share of the index, both rose. The Trump and Xi meeting in Washington on Thursday extended the existing trade truce rather than breaking it.

What works against it. The 30 year Treasury yield is at a 22 year high and was still rising on the last day of the week. A 6 billion dollar Treasury buyback on Wednesday failed to slow it. The median stock fell while the index rose, for a fifth consecutive week. And the market has moved to price roughly a 70 percent chance of another rate increase in October, one month after the September rise.

The read: the equity market is calm because the damage is being absorbed by the bond market and by the parts of the equity market that behave like bonds. That works until the level of rates starts reaching the winners too, and nothing this week told me where that level is.

Breadth. Cap weight beat equal weight by 1.68 percentage points, after 0.92, 1.06, 0.87 and 0.92 in the four weeks before it. That is the widest gap in five weeks and 0.76 points above the recent average, so it clears the bar where this becomes the story rather than a footnote. One company did most of it: Meta rose 16.80 percent and by itself carried Communication Services to the top of the sector table while Alphabet, Comcast, T-Mobile and Verizon all fell inside that same sector. The second contributor was semiconductors, and even there the weight was uneven, with AMD up 12.40 percent against Nvidia up 1.04 percent and Broadcom down 2.03 percent.

What makes this narrowing different from the usual kind is that it has a mechanical cause rather than a crowding one. The market did not decide it liked a few companies more this week. It raised the rate it applies to everything, and only a few companies could absorb it. That also tells you how it ends: when the long end stops rising, the equal weight index closes the gap quickly, because the names that lagged fell on rates rather than on results.

Sector Rotation Update

Into: Communication Services (+2.87%), Technology (+2.70%), Health Care (+0.88%).

Out of: Utilities (-4.23%), Energy (-2.66%), Financials (-2.38%), Real Estate (-2.07%).

The mechanism is a discount rate, and it is worth being precise about what that means. When the rate used to convert future profits into present value rises, it subtracts most from the businesses whose profits are fixed and furthest away in character, and least from the businesses growing fast enough to outrun it. Utilities and Real Estate are the former. That is the whole sector table this week.

The single most useful thing in the data is a split the sector average hides. Utilities were the worst sector at 4.23 percent down, but not uniformly. The regulated names were sold as yield substitutes: NextEra fell 6.03 percent, Eversource 5.79 percent, DTE 5.71 percent, Dominion 5.00 percent. The independent power producers tied to data centre electricity demand went the other way: Constellation rose 2.71 percent and Talen 2.64 percent. One sector label, two completely different assets, and the rate move told you which was which.

Energy is the other case where the label misleads. The sector fell 2.66 percent, and the obvious explanation would be falling crude. Brent actually rose 2.63 percent over the window, from 103.87 to 106.60, after dipping to 99.25 on Tuesday. The commodity went up and the equities went down, because on September 22 the administration said it was studying a ban on diesel exports and the refiners took it hardest.

Does it last? This rotation runs off the rate path, which makes it durable while the rate is rising and reversible the moment it stops. The honest caveat is my own record: across measured weeks, sector calls at one month have been right 59 percent of the time, close to a coin flip, and the Improving label specifically has been right 17 percent of the time. The labels here describe the week that happened. They are not forecasts.

The Five

Three of these rose and two fell. The two that fell are the ones that show the mechanism, and one of them is the stock of the week.

01 · COMMUNICATION SERVICES / INTERACTIVE MEDIA

Meta Platforms (META) +16.80%

Meta launched Muse, a personal AI agent, on September 8. On September 21 third party download data put it first among free apps on both the United States App Store and Google Play, ahead of the pace the incumbent assistant app set at its own launch, and the shares rose 11.34 percent that session. Meta then built its September 23 Connect keynote around the product and the shares added another 4.50 percent on September 24. The move is adoption data, not revenue: the paid tiers exist but the company has disclosed nothing about what they convert at.

Why it made the five. Of everything that moved this week, this is the clearest case of a large company being repriced on evidence rather than on atmosphere. A download ranking is a weak form of proof, but it is a measured one, and it arrived two weeks after a launch rather than being promised at a conference. The reason it survived the screen while Advanced Micro Devices did not is that the competing explanation fails here: if this were the whole AI complex being bid, Alphabet would not have fallen 2.05 percent and Nvidia would not have risen only 1.04 percent in the same four sessions. The two step path, 11.34 percent on the adoption data and 4.50 percent the day after Connect, tracks the two dated company events.

The case against it. Free downloads are the cheapest number a consumer product can produce and the least predictive one. The company has disclosed no conversion rate, no retention curve and no revenue from the paid tiers, so a 16.8 percent move in a company of this size rests on a ranking that can reverse in a month. An agent that fills in forms and makes purchases on a user behalf carries privacy and consumer protection exposure that no regulator has tested yet. And the cost side is not free: agent products are expensive to run, and the spending shows up in results before the subscriptions do.

What to watch next. Third quarter results, expected in late October though the company has not announced a date, the first report that can put a revenue number against the agent product Confirms: Management discloses paid tier subscribers or agent revenue, turning an adoption ranking into a line item Breaks it: The company declines to quantify agent revenue while raising the spending outlook to support it

02 · TECHNOLOGY / IDENTITY SECURITY

Okta, Inc. (OKTA) +13.31%

At its Oktane conference on September 22, Okta convened eleven other companies, including AWS, CrowdStrike, Databricks, Docker, Google Cloud, Proofpoint, Salesforce, ServiceNow, Wiz and Zscaler, to form the Blueprint Alliance and publish a shared reference architecture for securing AI agents. Okta also announced a runtime gateway of its own. The shares rose 13.31 percent over the window. This is the second consecutive week security software has repriced, and it is the first with a dated event behind it.

Why it made the five. The interesting thing about this catalyst is who benefits from it and who does not. Every co-signatory rose this week, but Okta convened the group, which is a different position from joining it. If agents are going to run inside companies, something has to decide what each one is allowed to do, and the company that gets the other eleven to agree on the shape of that layer is arguing for its own place in it. That is also why the other Alliance members were screened out of this issue rather than added: they share the catalyst and the date, so holding several of them would be one position wearing five names.

The case against it. A reference architecture is a document. It creates no contracted revenue and no obligation on any signatory, and the list of co-signatories includes the largest cloud providers, each of which can build this layer itself and has reason to. The stock has now risen two weeks running on the same theme without a results event, and the broker targets raised on September 23 into a 200 to 230 dollar range sit barely above the 206.64 dollar close, so the consensus upside has largely been used up. Identity is also a competitive market where Okta has lost share before.

What to watch next. Third quarter fiscal 2027 results, expected in early December though the company has not announced a date, and before that any Alliance member shipping against the published architecture Confirms: Management ties new bookings to agent identity products rather than to the core workforce identity business Breaks it: A major cloud co-signatory ships its own agent governance layer that does not need an independent identity vendor

03 · HEALTH CARE / LIFE SCIENCE TOOLS

Illumina, Inc. (ILMN) +14.31%

Illumina rejoined the S&P 500 before the open on September 21, about 27 months after being removed in June 2024, and rose 14.31 percent over the window. The honest complication is that the shares did not move on the day the index change took effect. The gain came on September 23 and 24, alongside a whole peer group with no index event: Pacific Biosciences rose 15.33 percent, Agilent 10.46 percent, 10x Genomics 9.80 percent, Mettler-Toledo 9.11 percent and Thermo Fisher 4.14 percent, against a Health Care sector up only 0.88 percent.

Why it made the five. This one is in the issue with a caveat attached, and the caveat is the lesson. The index inclusion is the only dated, primary sourced event available, and it is genuine: S&P Dow Jones Indices announced it on September 4 and it took effect on September 21. But it cannot be what moved the stock, because the stock did not move on that day and because seven companies with no index event moved with it. So what this week actually shows is a re-rating of the whole sequencing and lab tools group, and I could not find a single dated catalyst behind it. I am reporting that gap rather than filling it, because the alternative is inventing a reason, and a plausible reason is worse than an acknowledged gap.

The case against it. Index inclusion buying is mechanical, finishes in days, and adds nothing to earnings. If this week move was a group re-rating rather than a company event, it can unwind with the same group and for reasons that have nothing to do with Illumina: research budgets, China demand, or simply the trade getting crowded. The stock has already re-rated sharply over the past year from a depressed base, so the recovery is no longer the surprise it was, and the price now assumes the sequencing consumables recovery continues at the pace of the last three quarters.

What to watch next. Third quarter results, expected in late October though the company has not announced a date and two data providers disagree about it, so no date is published here Confirms: A fourth consecutive beat with the consumables recovery intact, which would make the re-rating about the business Breaks it: Sequencing consumables growth slows while the rest of the tools group keeps re-rating, separating the stock from its peers

04 · ENERGY / REFINING

Valero Energy Corporation (VLO) -7.36%

On September 22 the President said the United States should stop sending out its diesel, and the Treasury secretary said the administration was studying a full or partial export ban. Valero, the most export levered of the large refiners, fell 7.36 percent over the window. Marathon Petroleum fell 7.99 percent and Phillips 66 fell 6.32 percent. Nothing has been signed or scheduled, and the White House later denied a report that a 90 day restriction was being prepared.

Why it made the five. This is in the issue because of what it rules out. Brent crude rose 2.63 percent over the same window, so this was not the commodity. The Energy sector fell 2.66 percent while Valero fell 7.36 percent, so the gap between them is refining specifically. Three refiners with the same export exposure fell together on the same date. That is the signature of a policy landing on a shared exposure, and it is the cleanest illustration of a risk that no earnings model contains: the margin is not being competed away, it is being legislated about.

The case against it. The bear case here is the bull case inverted, and both are live. A refiner trading at a low earnings multiple on a record margin is the oldest trap in the sector, because the multiple is cheap precisely because the margin is peaking. But the reverse risk is just as real: nothing has been enacted, the White House has already denied part of the report, and a formal decision not to act would take the shares back up as fast as they came down. This is a binary with no scheduled resolution date, which is the hardest kind of exposure to hold.

What to watch next. Oct 22: Third quarter results and conference call, which the company has announced for October 22, 2026, the only company confirmed date among this week five Confirms: Management quantifies export volumes and what a restriction would cost, turning a headline into a number Breaks it: A formal decision not to restrict exports, which would remove the discount without any change in the business

05 · UTILITIES / REGULATED ELECTRIC

NextEra Energy, Inc. (NEE) -6.03%

NextEra fell 6.03 percent, the worst decline among large regulated utilities, in the week the 30 year Treasury yield closed at a 22 year high. Eversource fell 5.79 percent, DTE 5.71 percent and Dominion 5.00 percent alongside it. A Morgan Stanley target cut to 111 dollars and the 67 billion dollar Dominion Energy merger still before four state regulators are real overhangs, but they do not explain why the whole regulated group fell together.

Why it made the five. This pick exists to show a split that the sector average hides. Utilities fell 4.23 percent as a group, but not uniformly: while the regulated names were sold, the independent power producers tied to data centre demand rose. Constellation gained 2.71 percent and Talen 2.64 percent in the same four sessions. One sector label covered a bond substitute and a growth asset, and the market treated them as completely different things. If you own a utility for the dividend, this week told you which of the two you actually own.

The case against it. The uncomfortable part is what happens to the yield as the price falls. A utility that drops 6 percent now shows a higher dividend yield, which makes it look cheaper at exactly the moment the rate it competes against is rising, and that is how a bond proxy keeps looking attractive all the way down. Utilities also fund large capital programmes with debt and equity, so a sustained rise in long rates raises the cost of the growth plan itself, not just the discount applied to it. The Dominion merger adds regulatory risk on top, and it has four state approvals still outstanding.

What to watch next. Sep 30: The PCE price index on September 30, which is the reading that decides whether the long end keeps rising, and third quarter results expected in late October with no announced date Confirms: A hot PCE print followed by another leg up in the 30 year yield, which would keep the whole regulated group under pressure Breaks it: A cool PCE print that stalls the long end, at which point the bond substitutes are the fastest thing in the market to bounce

Stock of the Week

Valero Energy Corporation (VLO). This is the pick that fell, and it is the one worth the most attention. Valero did not miss a number or lose a customer. On September 22 the administration said out loud that it was thinking about restricting diesel exports, the Treasury secretary confirmed a study was under way, and the most export levered refiner in the country lost 7.36 percent in three sessions. Nothing has been signed. The White House has since denied a report that a 90 day restriction was being prepared. What makes it the stock of the week is the shape of the risk: the United States diesel crack spread hit an all time high of 102.20 dollars a barrel this month, and a company earning a record margin on a product a government is considering keeping at home has a specific exposure that no earnings model contains. Crude actually rose 2.63 percent over the same week. The commodity was not the problem. The policy was.

What Could Change My Mind?

A cool PCE print. My entire read rests on the long end continuing to rise. The PCE price index lands on September 30, and it is the hard price reading that either confirms or contradicts the survey that started this. If it comes in soft, the 30 year stalls, and the bond substitutes I just described as broken are the fastest thing in the market to bounce.

Two of my five are one bet. Meta and Okta both rose on the same belief, that autonomous agents are arriving in volume. Neither catalyst is revenue: one is a download ranking, the other a document published by a group of companies. If enthusiasm for agents cools, both give it back together, and they are the two largest gainers on the list.

I got the reason wrong last week and the answer right. That is not a comfortable thing to publish, but it is the most useful thing I learned. I said Technology would beat the equal weight index because the long end was falling. The long end rose 13 basis points to a 22 year high and Technology beat it anyway, by more than I asked for. So whatever is driving the technology bid, it is not the discount rate, and I had the mechanism wrong even while the call worked.

And I grouped three names that were not a group. In Issue No. 13 I named Generac, Zscaler and Natera as one bet on a falling long end. The long end rose past my own trigger. Generac fell 4.53 percent. Zscaler rose 8.78 percent and Natera rose 10.67 percent. They were never one bet, and I had said they were because they looked similar rather than because they trade together. That is the part of last week I would most like back.

One caveat on my own scoring. Across matured picks, a score above 85 has not produced better one month returns than a score below 75, and the rank correlation between my score and forward relative return is slightly negative. Read the reasoning rather than the number.

Scoreboard

This is the running record, losers included. Two numbers matter and they say different things.

The issue week column is not a forecast that came true. Names are screened into the issue partly because of how they moved that week, so they beat the index that week by construction. It cannot come out any other way. It is shown because it is the number readers saw, not because it measures skill.

The number that measures anything is what happened after publication, from the Friday close a reader could actually have acted on. Across 65 picks over 13 issues, one week after publication the average pick returned 0.38 percent against 0.44 percent for the S&P 500, so 0.06 points behind, with 60 percent finishing higher and 52 percent beating the market. At one month, where 45 picks across nine issues have matured, the average pick is up 4.19 percent against 1.41 percent, 2.78 points ahead, with 64 percent beating it.

And a new number that does not flatter me. The first picks have now reached three months, and there are only four of them from a single issue, which is far too small to conclude anything. For what it is worth at that size: they are down 6.30 percent against the index up 5.24 percent, so 11.54 points behind, and none of the four beat the market. That points the opposite way from the one month figure. I am publishing it at n equals four rather than waiting until it is either comfortable or large, because waiting until a number is flattering is how a scoreboard stops being one.

The "since published" column is measured from the publication Friday close to one week later, so its benchmark differs from the issue week figure quoted elsewhere. Marathon Petroleum is the loss worth naming. I picked it in Issue No. 13 for widening refining margins, and the thing that took 7.99 percent off it was not the margin at all: it was a government saying out loud that it might keep the diesel at home. I named the risk to that pick as a crude reversal. The risk that actually arrived was policy, and it is the same exposure that put Valero in this week's five.

Terms in This Issue

A Possible Next Step

I'm testing interest in a future Edge Report with entry zones, invalidation levels, sizing tiers, and exit frameworks. It does not exist yet and nothing is for sale. If that is something you would read, tell me and I'll build it.

What's going on, everybody?