Everyone Knows the Classic — Almost Nobody Knows the Bond Version
Most investors know the classic Sell-in-May effect: from November through April, equities historically outperform the May-to-October stretch. The usual conclusion is simple — step out of the market for the summer and sit in cash.
What 2025 research shows: switching into bonds instead of cash over the summer makes the strategy meaningfully better. The reason is refreshingly simple. Government bonds barely show any seasonality of their own. They deliver roughly the same in the equity weak months as in the rest of the year. That is exactly what makes them the better parking position.
We ran the finding against our own data. The result is more nuanced than the headline suggests.
The Study: Halloween Effect Meets Bond ETFs
In "To outperform: Sell-in-May enhanced with bond investments" (Pacific-Basin Finance Journal, Vol. 90, 2025), Chih-Hsiang Hsu and Donald Lien ask one question: what happens if you park the summer capital in bond ETFs instead of leaving it in a deposit account?
Their three core findings, over roughly 20 years and a set of global equity ETFs:
- The Halloween effect is alive. The November–April window beats the May–October window across the breadth of equity markets studied — robustly over the full sample.
- Government bonds show no significant seasonality. Their returns spread fairly evenly across the calendar year. There is no "worst six months" for bonds.
- Long-term beats short-term in the summer window. ETFs on long-maturity treasuries deliver higher May-to-October returns than short-maturity ones.
From this the authors build the combined strategy: equities November through April, long-term bonds May through October. In their sample, that combination beats buy-and-hold on both return and risk, measured by the Sharpe ratio.
Cross-Check 1: The Halloween Pattern in the S&P 500
Point one is easy to verify in SeasonAlpha. The seasonal annual progression of the S&P 500 (SPY) shows the familiar shape: a strong climb from November into April, then a flat, choppy stretch through the summer.
In numbers, based on the last 22 half-year pairs (total return, dividends included):
| Window | Avg. return | Median | Win rate | Std. dev. |
|---|---|---|---|---|
| November–April | +7.34% | +7.68% | 77% | 10.05 |
| May–October | +4.30% | +4.21% | 77% | 10.50 |
Note something easy to miss: the hit rate is identical in both windows. Summer is not statistically more often negative — it simply delivers less return per unit of risk. That is precisely the gap a bond position is meant to close.
Cross-Check 2: Why TLT Works as a Parking Position
The more interesting part is point two. If bonds show no summer weakness, the rotation is essentially free — you give up nothing. The 20-year monthly cycle of TLT (iShares 20+ Year Treasury ETF, long-dated US government bonds) shows exactly that:
The distribution of strength stands out. Average monthly returns over the last 20 years:
| Month | TLT avg. return | Win rate |
|---|---|---|
| May | +0.33% | 50% |
| June | +0.76% | 65% |
| July | +1.34% | 65% |
| August | +1.65% | 60% |
| September | –0.37% | 50% |
| October | –1.58% | 30% |
July and August — traditionally thin trading months for equities — are TLT's two strongest summer months. That fits the familiar risk-off reflex: when equities wobble in the summer, capital flows into treasuries. TLT's weak spot is October, the classic volatility month.
Measured across the full half-year window, the difference is effectively zero:
| ETF | Nov–Apr avg. | May–Oct avg. | Difference |
|---|---|---|---|
| TLT (20+ year maturity) | +2.10% | +2.38% | +0.28 pp |
| IEF (7–10 years) | +1.80% | +1.82% | +0.02 pp |
| SHY (1–3 years) | +0.94% | +1.00% | +0.06 pp |
Our data confirms the study: bonds know no Sell in May. And long-dated TLT delivers more than twice as much in the summer window as short-dated SHY.
Cross-Check 3: The 20-Year Practical Test
Now the decisive question — does the combination actually deliver more? We ran four variants over 20 years (Oct 2005 to Oct 2025), two switches per year, excluding taxes and transaction costs:
| Strategy (SPY base) | CAGR | Volatility | Sharpe | Max drawdown |
|---|---|---|---|---|
| Buy & hold SPY | 11.11% | 19.37% | 0.64 | –55.2% |
| Nov–Apr SPY / summer cash | 6.69% | 14.19% | 0.53 | –36.5% |
| Nov–Apr SPY / summer SHY | 7.72% | 14.22% | 0.60 | –34.4% |
| Nov–Apr SPY / summer TLT | 8.58% | 17.55% | 0.56 | –39.5% |
There are two readings in this table.
First, the study's finding holds: within the Sell-in-May family, the ranking is TLT > SHY > cash. Rotating into long-dated treasuries instead of a deposit account improved the annual return over 20 years by roughly 1.9 percentage points per year — at a comparable drawdown profile. The bond version clearly beats the cash version.
Second, however: against plain buy-and-hold in the S&P 500, the rotation loses over this period — on return and Sharpe. The reason is obvious. The last 20 years were an exceptional bull run for US large caps. Leaving the market for six months a year inevitably means missing a lot of it.
Where the Rotation Actually Wins
The study appeared in the Pacific-Basin Finance Journal and covers a global ETF set — not just US mega caps. And that is where the picture flips. The same calculation using EEM (emerging markets ETF) over the identical 20 years:
| Strategy (EEM base) | CAGR | Volatility | Sharpe | Max drawdown |
|---|---|---|---|---|
| Buy & hold EEM | 5.89% | 27.97% | 0.34 | –66.4% |
| Nov–Apr EEM / summer cash | 6.34% | 19.51% | 0.41 | –38.0% |
| Nov–Apr EEM / summer SHY | 7.37% | 19.53% | 0.46 | –34.6% |
| Nov–Apr EEM / summer TLT | 8.23% | 22.07% | 0.47 | –38.5% |
Here the bond rotation beats buy-and-hold on all three dimensions: +2.3 percentage points of CAGR, a Sharpe of 0.47 instead of 0.34, and a maximum drawdown of –38.5% instead of –66.4%. So the effect is real — it simply works hardest where the equity market itself was weak or extremely volatile.
Limits: What These Numbers Do Not Say
Four caveats belong here.
Duration is a genuine risk. TLT carries a duration of more than 15 years. In 2022 the ETF lost over 30% — in a year when equities fell too. The assumption that bonds cushion equity losses does not always hold. In inflationary phases, both fall together.
The lookback contains a decades-long tailwind. A large share of TLT's history falls into a period of structurally declining interest rates, and falling rates mean rising bond prices. Whether long-dated treasuries can repeat those returns in the current rate environment is an open question.
Costs and taxes are missing. Two switches per year trigger capital gains events in most jurisdictions. Over 20 years that eats a noticeable part of the edge — buy-and-hold has a structural advantage here.
Sharpe ratios are computed without a risk-free rate (pure return-to-volatility). They are therefore comparable with each other, but not with figures from external sources.
Putting It to the Test Yourself
If you want to explore the idea further, you can recompute it in SeasonAlpha with the tickers you care about:
- Monthly cycle with ticker
TLT: shows the average return per calendar month including win rate and significance gauge. CompareTLT,IEFandSHYside by side — the duration difference becomes obvious immediately. - Annual cycle with
SPYorQQQ: the normalized annual path makes the Nov–Apr window visible at a glance. - Backtest engine: combine entry and exit rules with indicator filters and test them for significance.
One methodological note: every SeasonAlpha seasonality uses normalized returns — each year starts at 100 and daily returns compound on top. Absolute price differences are never summed across years, otherwise later years at higher price levels would dominate the picture.
Conclusion
The study's core claim survives our cross-check: bonds have no summer weakness, and parking Sell-in-May capital in long-dated treasuries instead of cash improved the 20-year outcome by roughly 1.9 percentage points per year. In the summer window, TLT delivered more than twice what short-dated SHY did.
The second, less comfortable finding: against US large caps in a bull market, even the improved version loses. The rotation's advantage shows up where equities ran weaker and more volatile — in our test, emerging markets, with +2.3 percentage points of CAGR and a nearly halved drawdown.
This is not investment advice, just a historical analysis. Check the TLT monthly cycle for yourself at seasonalpha.ai.
Frequently Asked Questions
What is the difference between Sell in May and the Halloween effect?
Both describe the same pattern: equities historically deliver higher returns from November through April than from May through October. "Sell in May and go away" is the colloquial rule of thumb; "Halloween effect" is the term used in the finance literature, named after the late-October entry point.
Why do long-dated bonds like TLT beat short-dated ones in summer?
Two reasons. First, long-dated bonds carry higher duration and benefit far more from falling yields — safe-haven flows have a much stronger leverage effect on them. Second, their running yield is usually higher. In our data TLT delivered +2.38% from May to October versus +1.00% for SHY.
Do bonds really have no seasonality at all?
Not quite — at the monthly level there are differences, such as TLT's weak October (–1.58% on average, 30% win rate). But at the half-year level it evens out: for TLT, IEF and SHY, the Nov–Apr and May–Oct windows sit less than 0.3 percentage points apart. There is no systematic "bad half year" for government bonds.
Does the strategy also work with a broad aggregate bond ETF instead of TLT?
A broad aggregate bond ETF blends short, intermediate and long maturities plus corporate credit. That lowers duration and therefore both the return potential and the interest rate risk — the outcome would land between our SHY and TLT figures. To replicate the study's finding precisely, you need the long end of the maturity curve.