Last month I wrote that every midterm year since 1950 has had a decline before the vote, and that we’d assume it rather than predict it. Two readers sent back the only fair question: fine — when? This is my honest attempt at an answer: the case for the drop, the case against it from the fellow who’s been right all year, the eight-week window where every ingredient sits on the calendar, and what I’d do at each date. With a number on it, so December can grade me.
My window is September 11 to November 3. My center of gravity is the fortnight after the Fed meets. My odds are 60/40. Here is why.
Left is what the market and the weekend Journal said. Right is what I make of it. None of it is a forecast; all of it is a budget.
Six Reasons the Bill Comes Due This Fall
I won’t re-argue the season; it’s the floor under this letter and Part I has the whole thing. Nineteen midterm years since 1950, nineteen declines before the vote, an average of about −16%, the pre-election quarter the weakest of the cycle at +1.1%, and volatility peaking October through December. That is reason one, and it was true in August.
Reason two is what 2026 has not done. The S&P 500 closed Friday at 7,718.60, one percent off the record it set on Aug. 13. By one count it has closed at a record 27 times this year. It is up 12.8%, the Russell 2000 is up 19.9%, the semiconductor index is up 65.7% — and by my count the index has not had a 10% pullback at any point in 2026. The average calendar year since 1980 gives back about 14% from peak to trough somewhere along the way and still finishes up more often than not. A year that hasn’t paid that toll yet isn’t a year that got excused. The bill is unpaid, not forgiven.
Reason three is the one I’d keep if I had to throw the other five away. The Fed is not coming to help. Friday’s report showed 162,000 jobs in August against roughly 53,000 expected; July, reported as a loss, was revised to a gain; six-month average hiring is the highest in more than two years. The Journal’s own line is the right one — removing an objection to a hike is not the same as building the case for one — but the objection is gone. Futures put the September increase somewhere between a coin flip and 65%; at least one strategist wrote clients Friday to expect hikes in September and December; the chairman said last week he was hard pressed to call financial conditions restrictive; and the prediction markets put 93% odds on zero cuts for the rest of 2026. Every dip since 2022 got bought for one reason: the next move from the Fed was down. This is the first autumn in four years where the next move is up. Everything else in this section is decoration around that sentence.
Every dip since 2022 got bought for one reason: the next move from the Fed was down. This is the first autumn in four years where the next move is up.
The one thing that is different this year
Reason four is the bond market, which is already voting. The 10-year closed at 4.783% and the 2-year at 4.379%, both higher on the day stocks fell — stocks and bonds down together, the diversification promise failing in public again. Norway’s $2.4 trillion sovereign fund asked permission this week to cut government bonds from 70% of its bond book to 50%, roughly $80 billion less in Treasurys, and it gave the honest reason: safety has a cost and it had been buying too much of it. Prediction markets put 42% odds on the 10-year touching 5% before 2027. A stock market at roughly 24 times trailing earnings has to justify that multiple against a risk-free rate that keeps rising, and the arithmetic gets harder every week the 2-year sits above 4.3%.
Reason five is the mood, and the weekend paper was a catalogue of it. SpaceX raised $86 billion in June in the largest IPO ever and is down 7% since; the two big private AI labs are described as preparing to list “as early as this fall,” which is a great deal of new stock arriving into thin October books; Elon Musk’s tunneling company, valued around $20 billion, is telling investors they have to help recruit staff or lose their allocation; Nvidia paid about $13 billion for a company named after an emoji; a diagnostics firm is trading above the price it agreed to be bought at; and 612 of the 772 leveraged funds in America now track a single stock or coin. Jason Zweig’s column has the compressed version: South Korean stocks doubled in five months, single-stock leveraged funds launched at the end of May, and the market fell 40% in the six weeks after. None of these is a timing tool. Together they are a temperature, and the reading is “nothing can go wrong,” which is historically when something does.
Reason six is the fuel and the consumer underneath the index. Diesel set an all-time record Friday at $5.85 a gallon, up 56% since the war began, with crude sitting at $91 — the constraint has moved from the barrel to the pump, and it feeds every shelf price on a lag. Wages are up 3.1% and still behind prices. Lululemon fell 17% on Friday on an outlook cut; Campbell’s cut its dividend 36% the day before. The economically sensitive sectors — consumer discretionary, real estate — underperformed Friday while utilities are up 1.4% for the year and have gone nowhere. And here is the number that stuck with me from the trading diary: with the index one percent from a record, the New York Stock Exchange logged 90 new 52-week lows against 47 new highs. The average is at the top. The average stock isn’t.
None of the six is a date. Together they say the market is priced for a Fed that isn’t coming and for a consumer that’s thinner than the index. My weighting: the Fed reason carries more than the other five combined, because it is the only one that is genuinely new this year.
The last nine dips were bought by people who knew the central bank was on their side. The next one gets bought, or not, by people who know it isn’t.
The Fellow Who’s Been Right All Year Gets the Floor
Let me argue against myself properly, because the devil’s advocate has beaten me on timing since the spring and has earned an honest hearing. Here is his case, and it is a good one.
First, the volatility market says no. The VIX closed at 14.53 on Friday and rose a grand total of 1.5% on a day stocks fell. It has spent 25 straight sessions in a three-point band, one day short of a record that has stood since 1992. The three-month VIX sits at 17.61, so the curve is in contango — the calm, normal shape — and has been for 105 consecutive days; over the last sixteen years the curve has been inverted only about 7.7% of the time. Vol-of-vol is 83.8, well under the 100-plus reading that says tail hedges are being bought in size. It has been 111 trading days since the last volatility spike. A market that is about to fall does not usually look this bored. Volatility regimes persist far longer than the people betting against them expect: 2017, the calmest year on record, kept its VIX pinned for twelve months and never fell 3% from a high, and everyone who called the top that March was still calling it in December.
Second, the earnings are real. HP grew personal-systems revenue 18% last quarter while selling 16% fewer machines; Dell’s client business grew 20%; Lenovo nearly 30%. The semiconductor index is up 65.7% this year on a physical memory shortage that the people who track it do not expect to ease before 2028. Biotech is up 31.9%. Amazon is financing the largest gas-fired power plant in American history; OpenAI wants 3.2 gigawatts in one Georgia county. This is gas turbines and fabrication plants, not vapor. A real bear market needs an earnings recession, and I cannot find one anywhere in this weekend’s paper.
Third, the labor market is strong and the recession odds are single digits. The same 162,000 that argues for a hike argues against a crash. Unemployment is 4.1%; prediction markets put 6% odds on a U.S. recession by the end of this year. The deep drawdowns — 2001, 2008, 2020, 2022 — needed a recession or a central bank at war with inflation from a standing start. A quarter-point from 3.50–3.75% into 4.1% unemployment is the 1994 template, not the 2022 one: seven hikes that year, a bond-market massacre, and the stock market’s worst stretch was under 10%.
Fourth, the midterm record cuts both ways. Nineteen of nineteen includes a −4.4%. A decline that satisfies the pattern can be one you barely notice on a statement. The −16% average is dragged around by 1974 and 2008; the typical midterm dip is gentler than the average, and one of them might already have happened in a week you have forgotten.
Fifth, the market is already hedged. The crash-protection index near 151 means the people who wanted insurance bought it months ago; a hedged market doesn’t gap the way a naked one does, because the dealers on the other side of those puts buy the dips mechanically. The index put/call ratio is above 1.0 while the equity ratio is 0.67 — institutions insured, individuals long. That combination has cushioned every dip this year.
Sixth, the politics run one way for eight weeks. An administration heading into a vote with record fuel prices will pull every lever it has — reserves, export talk, public pressure on the Fed to cut — and every one of those is bad policy and good for stocks in the short run. The Journal reports the White House has no plan to restrict exports. It also reports the president posted Friday that the Fed “must get smart” and lower rates. That pressure does not get quieter between now and Nov. 3.
The devil’s advocate wins on timing more often than not, and I have written his best line at the top of my playbook: do not fight a volatility regime; wait for it to break on its own. Where he loses is the Fed. He is arguing that this dip gets bought like the last nine. The last nine had a central bank cutting or about to. This one has a chairman who cannot call conditions restrictive and a market that has stopped pricing relief for the year.
So I keep the 60/40. Sixty because of the Fed. Forty because he has been right, and I have watched him be right, since March.
Eight Weeks, Nine Dates, and One Fortnight That Has Everything
Here is the calendar the way I’d game it. Every line is a real date. The odds are mine, they are rough, and they are written here so they can be wrong in public.
Monday, Sept. 7 — Labor Day. Markets closed. Tuesday reopens with Friday’s jobs number still the only new fact in the room, and a bond market that had the weekend to think about it.
Friday, Sept. 11, 8:30 a.m. — August inflation. The report the Fed said would decide it. Hot, and hike odds go to 80%-plus, the 2-year goes through 4.5%, and a VIX at 14 gets its first real test. Cool, and the odds collapse, stocks rally, and my 60 drops toward 45. Diesel up 56% argues for a hot headline; the core number is the actual question.
Tuesday–Wednesday, Sept. 15–16 — the Fed. A hike would be the first this year after holding at 3.50–3.75%, into a market a percent from a record at 24 times earnings; that is the trigger the bear has waited on since March. A hold is a relief rally — and, I’d argue, one that gets sold by the middle of October, because a hold in September only moves the same question to Oct. 28.
Friday, Sept. 18 — quarterly options expiration. The biggest expiry of the quarter, and the moment the dealer positioning that has pinned the VIX inside three points rolls off the board. The two weeks after the September expiry are, on average across the last few decades, the weakest fortnight of the trading year. This is the one date on the list that has nothing to do with news and everything to do with plumbing.
Wednesday, Sept. 30 — Micron, after the close. The memory shortage goes on the record with a number attached, from the company at the center of it. Also the end of the government’s fiscal year; prediction markets put about 1% on an appropriations lapse Oct. 1, so I’m not budgeting for it.
Friday, Oct. 2 — September jobs. If August was a fluke, this is where it shows. If it wasn’t, the December hike gets priced.
Wednesday, Oct. 14 — September inflation, and the banks open third-quarter earnings the same week. Guidance, not beats: Lululemon lost 17% on Friday with a perfectly ordinary quarter behind it and a cut outlook in front of it. October is when companies that carried record diesel through the summer tell you what it cost them.
Tuesday–Wednesday, Oct. 27–28 — the Fed again, the same week the largest technology companies report. Two live meetings inside six weeks is not something this market has had to price in four years.
Tuesday, Nov. 3 — the vote. Then Nov. 10 inflation, Nov. 17 Nvidia, and Dec. 8–9 a Fed meeting with fresh projections. The window closes on election night either way.
Now the scenarios, with my rough odds attached. Base case, about 45%: a 6–10% pullback from the Aug. 13 high, the low landing somewhere between the last week of September and the middle of October, bought into the election. The dividend book and the utilities that have gone nowhere this year outperform on the way down; long bonds, as they did on Friday, do not help. Bear case, about 15%: 12–18% — a hot print, a hike, October guidance cuts in the consumer names, an AI-lab IPO wave draining liquidity into thin books, and the 10-year at 5%. Midterm-average territory, the kind Part I was built for. No drop, about 40%: the 2017 script. Dips of 3–5% get bought, the VIX stays pinned, the inflation report cooperates, the Fed holds in September, and the record streak runs straight into the vote; the drop, if there is one, becomes 2027’s problem.
Add the first two and you get my 60. Look at how small the second one is and you see why the dividend book, and not a short position or a pile of puts, is how I choose to express any of this.
The question isn’t whether I can see the drop. It’s whether the market can, and at a VIX of 14 it has decided not to look. My window is the six weeks in which it will be made to. My center of gravity is the fortnight after the Fed meets, because that fortnight holds the reason and the plumbing at the same time.
If I’m wrong, it will most likely be because the inflation report on the 11th comes in cool and the whole thing gets postponed to Oct. 28. That is the specific way I expect to be wrong, and I have written it down so you can check.
What I’d Do at Each Date — Written Down Before Any of Them Arrive
Now through Sept. 11: nothing new. The Midterm Dividend book stays the flagship — forty names at the $100,000 tier, a measured beta of 0.39, every payout coverage-tested, and a T-bill-and-gold reserve earning its keep while it waits. The Aggressive Tactical book stays held through, as it was in Part I; you don’t swap the destination vehicle because the road has a rough section coming. No new positions into a binary. The one addition I’d like to make — Micron, on the memory shortage, at no more than 1% — waits for the print. Valero, the refiner added Sept. 4, keeps its 1.5% with a tripwire written down: any White House move toward restricting diesel exports is the trim signal. Not the diesel price. The policy.
Sept. 11: a hot report confirms what the income sleeve already assumes — no long nominal bonds, the cash in floating-rate paper and short bills that do not need the meeting called correctly. A cool report and the tactical additions come back on, starting with the one above.
Sept. 16: a hike, and the dividend book earns its keep the same afternoon. A hold, and there is no victory lap, because the fortnight after the meeting is where the drop would live regardless, and the same question just moved to Oct. 28.
The fortnight, Sept. 18 to Oct. 14: the shopping list is written now, on a Sunday, so nobody has to write it on a red Tuesday. On an 8% index drawdown from the Aug. 13 high, one-third of the reserve goes to work — the equal-weight index first, then the power builders and the memory names. Another third at 12%. The last third stays for November no matter what. The reserve was always November’s shopping money; the only question this letter asks is whether it gets to shop early.
Oct. 14 and the earnings that follow: guidance, not beats. A name in the tactical book that cuts its outlook gets sized down that day, on a rule, not debated in a meeting. Lululemon is the template for what that looks like from the outside.
Oct. 28: if two hikes are priced by then, a 5% 10-year is the base case and I want no duration in any account that the owner did not choose on purpose. That is the single most common thing I find in a target-date fund, and it is the one that would hurt.
Nov. 3: the other nineteen-for-nineteen. The twelve months after every midterm vote since 1950 have been positive, averaging about +18.8%. The reserve starts spending the week of the vote whether or not the drop ever came. If it never comes, I will have been wrong about the date and right about the plan — which is the only way I am willing to be wrong.
The devil’s advocate’s veto, honored: while the volatility curve stays in contango and vol-of-vol stays under 90, I buy no hedges. A vol regime that has held for 105 days is not something I fight with a put. I wait for it to break, and I have the reserve to act when it does.
For a household, none of this requires my window to be right. It is four checks, and they are the same four whether the drop comes in October or never: how much duration is in the bond fund you did not choose, inside the target-date account; twelve months of expenses in cash if you are over fifty, in an economy adding jobs but where only a third of people say it’s a good time to find one; the “growth” fund on the menu that turned out to be a momentum fund, the trade having its worst stretch against the market in 25 years; and the beneficiary pages, which override the will and which nobody has looked at since the last move.
A prediction needs the world to cooperate. A dated plan only needs me to follow it on the day it stops being fun. I have put the dates in this letter so you can hold me to them, and so that when one of them arrives, the decision was made on a Sunday in September by someone who was not frightened.
Sixty-forty. Sept. 16 to Oct. 14. A third, a third, a third. December can grade the rest.
Which side of the window is your plan on?
Bring the statement. We’ll find the duration you didn’t choose, the growth fund that is a momentum fund in disguise, and the cash line — before Sept. 11, not after it. Fifteen minutes, and the calendar does the rest.