Backtesting Arena

Backtesting Arena

Back to blog

Sell Before the Midterm Year? In Midterm Years the Stock Market Performs Worse Than Usual — but Rarely in the Autumn.

Across 18 midterm years since 1954 the S&P fell 17.7 percent from its high on average — but only 5 of them had a 19-percent drop inside the second half. What the number measures, how much of it reaches Bitcoin (through the S&P, not through yields), and why Bitcoin's own midterm-year bears come from the halving.

Backtesting Arena·September 17, 2026·9 min read·0 views
Sell Before the Midterm Year? In Midterm Years the Stock Market Performs Worse Than Usual — but Rarely in the Autumn.

The formula is being passed around again in 2026: "Every midterm year has had a stock-market correction in the second half, averaging 19 percent." Anyone who believes it spends September with a finger over the sell button, for stocks and for Bitcoin.

The 19 percent exists. It just measures something other than what the sentence claims. And the part that reaches Bitcoin travels through a different channel than most people assume.

What the number measures: the whole year, not the autumn

Computed from S&P 500 daily closes for all 18 midterm years from 1954 to 2022:

Metric (n = 18 midterm years)Value
Largest decline from the high within the calendar year, average−17.7 % (median −17.9 %)
Years with a decline ≥ 10 %12 of 18
Years with a decline ≥ 19 %9 of 18
Low in the second half13 of 18
Largest decline inside the second half, average−12.3 % (median −10.0 %)
Second halves with a decline ≥ 19 %5 of 18
Second-half return, median+4.2 % (12 of 18 positive)

The 19 percent is the average pain across the whole calendar year, measured from the highest close to the lowest. A drop of that size inside the second half happened in 5 of 18 years: 1974, 1990, 1998, 2002, 2018. In the other 13 the low was already in by spring, or the autumn stayed under ten percent. The median second half is plus four.

The pattern the number really carries is a different one: midterm years hurt more than other years. In the 57 non-midterm years from 1951 to 2025 the largest annual decline averaged −12.5 percent; in midterm years, −17.7. Full-year return: +11.2 percent against +3.7. The midterm year is the weakest of the four-year cycle. What is not fixed is the calendar of when it hurts.

The full list, so nobody has to hunt for the five:

YearDecline in yearLow onDecline in H2H2 return
1954−4.4 %31 Aug−4.4 %+23.2 %
1958−4.4 %25 Feb−4.2 %+22.0 %
1962−26.9 %26 Jun−10.5 %+15.3 %
1966−22.2 %7 Oct−16.4 %−5.2 %
1970−25.9 %26 May−4.7 %+26.7 %
1974−37.6 %3 Oct−27.6 %−20.3 %
1978−13.6 %14 Nov−13.6 %+0.6 %
1982−16.6 %12 Aug−8.2 %+28.3 %
1986−9.4 %29 Sep−9.4 %−3.5 %
1990−19.9 %11 Oct−19.9 %−7.8 %
1994−8.9 %4 Apr−6.4 %+3.4 %
1998−19.3 %31 Aug−19.3 %+8.4 %
2002−33.8 %9 Oct−21.5 %−11.1 %
2006−7.7 %13 Jun−3.6 %+11.7 %
2010−16.0 %2 Jul−7.1 %+22.0 %
2014−7.4 %15 Oct−7.4 %+5.0 %
2018−19.8 %24 Dec−19.8 %−7.8 %
2022−25.4 %12 Oct−16.9 %+1.4 %

Five bold rows in eighteen. That is the formula's hit rate when taken literally.

What 2026 makes of it

The S&P closed the first half of 2026 at +9.6 percent and stood at +11.6 percent year to date on 9 September; the largest decline so far was −9.1 percent, low on 30 March. A strong first half is no protection, but it is a different sample: in the 25 years since 1951 in which the S&P ended the first half up at least 9 percent, the second half was positive 21 times, median +9.6 percent; the one large exception was 1987 at −18.7 percent. A year can sit on both lists. Neither list is a reason to sell.

The tradeable part of the midterm pattern is not the autumn but what follows it. From 30 September of a midterm year to 30 September of the next, the S&P was up in 16 of 18 cases, median +24.1 percent; the two exceptions, 2010 and 2014, were −0.9 and −2.6 percent. That is not hindsight from a low nobody knew in advance; it is a fixed calendar date. The midterm pattern is not a sell signal for the autumn. It is a buy statistic for the winter.

What reaches Bitcoin: the S&P, not yields

The second half of the timeline formula says: when stocks correct, Bitcoin falls with them, and rising long-end yields are to blame. Both can be measured, on daily moves.

Correlation of Bitcoin daily returns with …90 days1 year5 yearssince 2015
S&P 5000.300.480.410.23
Change in the 10-year yield−0.12−0.05−0.020.00
Change in the 30-year yield−0.04−0.030.000.01

Bitcoin is tied to the S&P, clearly so since 2020: 0.43 in 2020, 0.56 in 2022, 0.45 this year. It is not tied to long yields. The 10-year yield rose from 4.16 to 4.84 percent in 2026, the 30-year stands at 5.29; the correlation of daily moves with Bitcoin is zero in every window. The platform measures the same thing on a weekly basis: 10-year real yield against Bitcoin over five years 0.15, S&P proxy over 90 days 0.34. Anyone selling Bitcoin because of yields is selling because of a number that does not move Bitcoin.

What moves Bitcoin is the stock market. A correlation of 0.48 over one year means the S&P explains about a quarter of Bitcoin's daily move (r² 0.23), and because Bitcoin swings 3.4 times as hard, an S&P day of −1 percent arrived at Bitcoin as −1.65 percent on average (beta 1.65; 2026: 1.53). A 19-percent S&P correction, which happened in 5 of 18 midterm autumns, is not a separate event for Bitcoin. It is leverage on somebody else's.

Bitcoin's own midterm bears come from the halving

And yet midterm years were Bitcoin's worst years of all:

YearBitcoinlargest declineS&P
2018−73.6 %−81.5 %−6.2 %
2022−64.3 %−66.9 %−19.4 %
2026 (to 10 Sep)−12.5 %−39.6 %+11.6 %

In 2018 Bitcoin lost three quarters while the S&P lost six percent and the correlation of daily returns sat at 0.07, at 0.00 in the second half. Bitcoin fell on its own. The reason is in the platform's halving table: cycle highs on 4 December 2013 and 16 December 2017 (371 and 525 days after the halving), the bull-market high of November 2021, and the high of 6 October 2025 (535 days). Each was followed by the bear year, and the bear year was a midterm year every time: 2014, 2018, 2022 and now 2026, 39 percent below the October high.

That is not an election cycle. It is the halving calendar, which since 2012 has landed on the election calendar every four years. Bitcoin falls in the midterm year because the halving was two years earlier, not because there is a vote in November. Three cases with a number, listed one by one, and the fourth in progress. Too few for a rule; enough to explain why the timeline formula looks so good on Bitcoin.

The objection: so the formula holds, just for a different reason?

"Halving or election, Bitcoin was deep in the red in every midterm year. So selling in the midterm autumn is right anyway."

By autumn it is too late. By the end of June Bitcoin had already lost 55 percent in 2018, 57 in 2022 and 33 in 2026; anyone selling in September sold after the larger part of the move (lows: 15 December 2018, 21 November 2022, so far 30 June 2026). And for the S&P, the 30 September calendar says the opposite of sell. The formula takes the right number, attaches it to the wrong half-year, and explains it with the wrong market.

What to do with this

  1. Ask for the window. "19 percent" is an annual maximum. Anyone turning it into an autumn wants attention, not accuracy.
  2. Measure the channel. Bitcoin takes equity corrections with leverage (beta 1.5 to 1.65) and yield moves not at all. Anyone hedging hedges against the S&P, not against the yield.
  3. Take the calendar that holds. 30 September of a midterm year to 30 September of the next: 16 of 18 positive. That is the part of the pattern that could be traded without hindsight, and even that is not a promise.

Not claimed: that 2026 will have no autumn decline; 5 of 18 is not zero. Still open: whether Bitcoin's correlation with the S&P is higher in midterm autumns than in the rest of the year, or whether 2022 at 0.55 was the outlier and 2018 at 0.00 the rule.

Method

S&P 500 price index, daily closes 1950 to 9 September 2026 (Yahoo Finance, ^GSPC). Decline = lowest close relative to the highest prior close, with the prior year's close as the starting point. Second half from 1 July. Correlations: Pearson of daily returns on S&P trading days; Bitcoin return over the same calendar gap; yield changes as differences in percentage points. Bitcoin-USD from 17 September 2014 (Yahoo), hence no 2014 figure. Platform values from the connector on 16 September 2026: macro correlations (weekly, FRED), halving table (cycle highs on daily-close basis), cycle snapshot (distance from the high).

FAQ

Why the price index and not SPY with dividends? Because the formula "19 % SPY" talks about a decline, and declines are measured on price. With dividends the first half of 2026 would be slightly above +9.6 percent; the drawdown figures barely change.

Is a correlation of 0.45 a lot? For Bitcoin, yes: from 2015 to 2019 it was under 0.1 every year. For a stock against its own index it would be low. The number means roughly a quarter of Bitcoin's daily move can be explained by the S&P; three quarters come from elsewhere.

Why the 30-year yield and not the Fed? Because the timeline formula names the long end. The platform also measures the policy rate: correlation over five years −0.06. Nothing there either.

Not investment advice, not a recommendation, not a forecast — historical patterns are not a promise.

Study the Past — Improve your Future 🥋

Try it yourself

Run the backtest with your own parameters and time ranges.

Run backtest →

More on this topic

Market Analysis

Asset Price Inflation — The Question Is Right, the Chart Is Not

Backtesting Arenatradingstrategies.work

If money grows faster than output, where does it go if not into consumption? The question has a good answer. It just isn't the one in the chart everybody posts.

MacroMethodology
Aug 9, 20261 min
Market Analysis

Popular Trading Myths, Tested Against the Actual Numbers

Backtesting Arenatradingstrategies.work

Max pain, the yen carry trade, rate cuts, the four-year cycle: seven trading myths, each held against the numbers. The answer is almost never "myth" — usually it is something less comfortable.

BacktestingMarket structureMethodology+2
Aug 6, 20261 min
Market Analysis

Bitcoin Yield: Where It Comes From — and What You Sell for It

Backtesting Arenatradingstrategies.work

Bitcoin has no native yield. Every advertised BTC yield is a sold option, a loan, or a token emission. And with the income ETFs, the window decides whether the answer flatters or destroys.

BitcoinDerivativesMethodology
Jul 14, 20261 min
Market Analysis

Did Bitcoin Front-Run Its Bear Market? What the Broken 4-Year Cycle Means for 2026

Backtesting Arenatradingstrategies.work

For the first time since 2012, Bitcoin hit a new all-time high *before* the halving. 2025 didn't look like a post-halving bull market — it looked like a front-run bear market. If everyone knows what's coming, everyone trades it earlier — and the classic October-2026 bottom around $40k might already be history. What this thesis implies for the months ahead.

BitcoinMacroETF
May 23, 20261 min
📬

Don't miss new blog posts

One short email per new post — strategies, backtests, market analysis. No spam, unsubscribe with one click anytime.

By subscribing you accept our privacy policy. We use Resend for delivery. Double opt-in confirmation required.

Comments (0)

Join free to post comments.

Sign up →

No comments yet. Be the first!