Mathematical Money | October 13, 2026
The midterms are on Tuesday 3 November, twenty-three days away, and the seasonality charts are out in force. The headline number is a good one: since 1950, October in a midterm year has averaged +3.0% for the S&P 500 and finished higher 73.7% of the time. November is similar at +2.7% and 78.9%.
I like these posts and I read all of them. I also went and checked this one, because a 73.7% hit rate is the sort of number that should make you ask what it is being compared against.
It turns out the comparison almost nobody states is the one that decides whether the pattern means anything, and once you make it the edge mostly disappears.
First, the sample is bigger than people say
Most versions of this stat say there have been "about 17" midterm years since 1950. There have been nineteen completed ones: 1950, 1954, and every fourth year through to 2022.
You can prove it from the stat itself. 73.7% is exactly 14 out of 19. The November figure, 78.9%, is exactly 15 out of 19. The percentages have the sample size baked into them, and it is not 17.
That is a small thing and it does not change the conclusion. I mention it because if you are going to quote a statistic to strangers, the number of observations behind it is the first thing worth getting right.
Now the part that matters, in plain numbers
Forget probabilities for a second and just count. Fourteen of those nineteen Octobers finished higher. If October were a coin flip you would expect about nine or ten out of nineteen. Fourteen against nine or ten looks like a genuine pattern, and that is where all the excitement comes from, because the coin flip is the comparison almost every version of this post is implicitly making.
But October is not a coin flip. Stocks drift up. October finishes higher in 59.7% of all years since 1950, midterm or not, averaging +0.91%. So across an ordinary run of nineteen Octobers you would already expect about eleven of them to be up, with no election involved at all.
Eleven expected. Fourteen actual. That is the whole thing. The entire famous pattern is about two extra up months, spread across seventy-six years of history. November is the same shape: you would expect roughly thirteen up months from the base rate, and you got fifteen. Two extra.
Put that way it stops sounding like an edge and starts sounding like what it is, which is a small number of coin flips landing slightly one way in a very small sample.
For anyone who wants the formal version: against a 50% null, 14 of 19 gives a one-sided probability of 0.032, which is significant. Against October's own 59.7% base rate it gives 0.156, which is not. November goes from 0.010 to 0.273 the same way. Both months clear the wrong hurdle convincingly and fail the right one.
One month out of nineteen is carrying a quarter of the average
Having the year-by-year numbers rather than the summary lets you see something the aggregate hides completely.
October 1974 was +16.30%. It is by far the largest month in the set, and it lands in the middle of the worst bear market of the period rather than being anything to do with an election. Strip that single observation out and the famous +3.00% average falls to +2.26%. One month out of nineteen supplies a quarter of the headline number.
The median tells the same story more quietly. The mean is +3.00%; the median is +2.54%, so half the sample sits below that.
And here is the one I find most telling. October was up in 14 of 19. November was up in 15 of 19. But both months were up in only 10 of 19, which is 52.6%, essentially a coin flip. 1974 went +16.3% then −5.3%. 1994 went +2.1% then −4.0%. The two famous statistics are usually quoted side by side as if they compound, and about half the time they did not.
The average return has the same problem
The +3.0% is a real number. It is also an average of nineteen observations of something that moves about five percent a month.
Run the confidence interval on it and the true value sits somewhere between roughly +0.5% and +5.5%. The point estimate is positive and the interval is so wide it is almost useless for sizing anything. You would be making a decision on a range that contains both "barely worth the commission" and "a very good month."
And the sample cannot improve quickly, because midterms happen every four years. Nineteen observations took seventy-six years to collect. The next data point arrives in 2030. Any pattern built on a four-year cycle will be statistically thin for the whole of your investing life.
What does survive, and it is not the bit on the charts
I do not want to be the person who only ever says the number is noise, so here is the version of this that holds up properly. It is not about October at all.
Take the 31 October close of each midterm year and measure forward.
Over the next six months the S&P was positive in all nineteen cycles, averaging +14.8%. Over the next twelve months, positive in all nineteen, averaging +16.6%. Not one negative outcome in either column since 1950. The weakest was 2014, and it still returned +3.3% and +3.0%.
Now apply the same test that killed the October stat, because a run of nineteen means nothing until you know the baseline. Measure the identical windows from every October since 1950, not just midterm ones: positive 76% of the time over six months averaging +7.0%, and 80% over twelve months averaging +9.3%.
So the midterm version is roughly twice the return and a perfect record against a baseline that already fails one year in four or five. The probability of nineteen straight positives at those base rates is 0.005 for the six-month window and 0.014 for the twelve-month one. Both are significant, and both clear the hurdle the famous October statistic fails.
That is the finding. The pattern everybody quotes is the weak one, and the strong one sitting right next to it is about the following year rather than the next three weeks.
I still would not lever into it. Nineteen is nineteen, the windows overlap with the general upward drift of equities, and "it has never failed" is exactly the sentence that precedes a first failure. But if I had to carry one fact out of all of this, it would be that one, and it would not be the +3% October.
Why none of this changes what I do
Here is the part specific to how I actually trade, and it is the reason I find most seasonality content interesting but unusable. Every statistic above is about direction, and direction is not what I sell. I write puts and calls against positions I hold. What determines whether that works is not whether the market drifts up three percent, it is whether the options were priced above what the market subsequently delivered.
A 73.7% chance of an up month tells me nothing about that. If everyone knows October tends to be strong in midterm years, that expectation is already in the option prices I am selling into.
The genuinely useful fact in the whole seasonal picture is the one buried at the bottom of most of these posts, which is that September and October have been among the most volatile months since 1928. High realised volatility is what makes premium rich, and rich premium is what I am paid from regardless of which way the index ends up going. That one I can act on.
So the seasonal pattern I act on is not the drift. It is that this is a well-paid stretch of the year for selling options, and that the resolution of a binary political event is usually followed by volatility coming out of the market, which is good for anything short premium going into it.
What I'm doing into the vote
Nothing different, which is the honest answer. The cadence does not change for an event with a known date.
What I am watching is whether implied volatility is elevated going in relative to what the market actually does. If it is, premium selling is being paid for an uncertainty that resolves on a schedule. If it is not, the event is already priced and there is no extra compensation for holding through it.
That is a question about option prices in late October rather than about a seasonal chart built on data from 1950, and it is the only version of this I can actually trade.
Two things I'd like other views on.
Does anyone test seasonal stats against the base rate rather than against 50%? I had read the midterm October figure a dozen times before it occurred to me to ask what October does in a normal year, and the answer took the entire edge out of it.
And the harder one: if a pattern is well known enough to be on every chart in October, can it still be in the price? I lean towards no for anything directional and yes for volatility, which is why I pay attention to one and not the other.
Midterm October and November figures as widely reported since 1950. Base rates from S&P 500 monthly seasonality, 1950–2025. Significance tests are one-sided binomial, computed on 14/19 and 15/19.
#seasonality #midterms #SPX #options #volatility
Stop guessing. Start calculating.
Live to fight another day. 🤙
Comments