The Lens · How Markets Actually Work

We Found a Real Edge. It Died in 2017. That Was Predictable.

Some of the largest investors on earth are forced to sell shares at the end of certain months. Not persuaded — forced, by rules they wrote down and cannot ignore. That forced selling leaves a footprint in prices, and we went looking for it. We found it: clearly, across four US stock indices and again in Europe, over almost a decade. Then we checked whether it still worked. It has been gone since 2017. The interesting part is not that we found a dead pattern. It is that why it died turns out to be the single most useful thing we know for judging whether any market pattern will survive being discovered.

Dragonfly Lens · 11 September 2026 · Test designed and its pass/fail conditions written down BEFORE it was run, so the result could not be argued into a success afterwards. Data: clean daily futures, 2008 to 2026.

The short version

The setup is real and public. A pension fund promising a 60/40 split between shares and bonds must rebalance when markets move it off target. Shares rise, the share slice grows too big, and the fund has to sell shares. This is written policy, not speculation.
The footprint was there. The bigger the drift during a month, the weaker stocks were in the final days — exactly as forced selling predicts — on all four US indices we tested, and strongly in Europe. It even rebounded afterwards, on all five.
And it is dead. Strong 2008–2011, present 2012–2016, gone 2017–2021, gone 2022–2026. That is nine years and roughly 120 months of nothing.
The flow never stopped. Pension funds still rebalance. The money still moves. What disappeared was its predictability in time.
The rule this gives you: a pattern survives being famous only if the people on the losing side cannot choose when or how to do the thing that loses them money. If they have any flexibility at all, they will use it, and the pattern will fade.
We were wrong about this in advance and wrote our reasoning down beforehand, so we could not pretend otherwise. That is included below.

1. Why anyone is forced to sell anything

Suppose a pension fund promises its members a portfolio that is 60% shares and 40% bonds. That is not a preference; it is written into the fund’s policy, and the people running it have a legal duty to stick to it.

Now shares have a good month and bonds do not. Left alone, the fund is no longer 60/40 — it might be 63/37. It is now carrying more risk than it promised. So it must sell shares and buy bonds to get back to target. This has nothing to do with anyone’s view of the market. They may be convinced shares will keep rising. They sell anyway, because the rule says so.

That is genuinely unusual and genuinely interesting. Most market participants can change their minds. These ones cannot. And because the drift is caused by returns everybody can observe, you can work out roughly what they will have to do before they do it.

Why this seemed so promising. Most supposed edges rely on being cleverer than the person on the other side. This one does not. It only requires that somebody else is obliged to trade at a moment not of their choosing — and that we are willing to take the other side. You are not out-thinking them. You are being paid to absorb an order they had no say in.

2. What we found

We measured how far shares had outrun bonds during each month, then looked at what shares did in the final few days — the window where rebalancing has to happen. The prediction is straightforward: the bigger the drift, the more they must sell, the weaker those last days should be.

It showed up on everything:

The relationship had the correct sign on all four US stock indices we tested.
It was statistically convincing on three of them.
It appeared independently in Europe, comparing European shares against German bonds — different funds, different regulators, same rule, and there it was the strongest of all.
Prices rebounded in the first days of the following month, on all five, once the forced selling cleared. That is the part that convinced us, because it is what a temporary price push caused by flow — rather than by news — should look like.

It was not a 2008 crisis artefact either: remove 2008 and 2009 entirely and it is still clearly present. By any normal standard this was a finding.

3. Then we looked at when

We split the eighteen years into four consecutive blocks and ran exactly the same measurement in each. The numbers below are a measure of statistical strength — roughly, how confident you can be that a relationship is real rather than luck. Anything past about 2 is usually taken seriously; near 0 means nothing is there.

Market	2008–2011	2012–2016	2017–2021	2022–2026
US small caps	2.3	1.7	1.3	1.2
US large caps	1.7	1.2	0.3	0.2
US blue chips	1.6	1.3	0.6	0.3 (wrong way)
US tech	0.8	1.2	0.1	0.9
Europe	2.6	3.5	0.1	0.2

The US tech row is included because it is inconvenient: it was never convincing in any period, and it does not decay tidily like the others. Showing only the four rows that tell a clean story would be the easiest kind of dishonesty, and the hardest to catch.

Read the Europe row across. It is the strongest result in the study and it falls off a cliff. Not a gentle decline — 3.5 to 0.1, and it has stayed there for nine years.

A pattern that only exists in the first half of your data is not a discovery, it is a history lesson. Had we tested only the full period, the overall number would have looked respectable and we would have concluded we had something. The full-sample figure is an average of a real effect and a decade of nothing — and the average is worthless, because you cannot trade the first half.

4. Why it died — and why the flow did not

Here is the thing worth taking away. Pension funds did not stop rebalancing. They are still bound by the same policies, still drift off target, still have to correct. The money still moves, in the same direction, in the same amounts.

What changed is when.

Think about rush hour. Everybody has to get to work. Knowing about rush hour does not excuse anyone from going. But once it is well understood, people adapt around it: leaving earlier, taking a different route, working from home on Tuesdays, staggering shifts. The commuting never stops. The predictable jam softens. The traffic is still there; it is just no longer reliably in one place at one time.

That is exactly what happened. As the pattern became well known, the funds and their brokers stopped doing it all at once on the same afternoon. They spread the trades across more days. They used execution algorithms designed to hide size. They moved into the closing auction, where the most buyers and sellers meet at once and a large order does least damage. They arranged trades privately rather than in the open market.

None of that changed the obligation. All of it destroyed the predictability — and the predictability was the entire edge.

5. Where our reasoning was wrong

Before running this we wrote down why we expected it to survive. Our argument was: knowing about a rule does not release a fiduciary from that rule; they still have to trade; therefore this cannot be arbitraged away.

Every clause of that is true. The conclusion does not follow, and we now have a reasonably crisp statement of why:

A mandate forces the TRADE. It does not force the SCHEDULE. We had quietly assumed that because someone must do a thing, they must do it at a predictable moment. They must not. Any freedom in when or how is enough for the predictable part to be competed away, while the obligation itself carries on untouched.

There was a second clue we could have taken more seriously, and it points the same way. If this were really about mandated rebalancing, the effect should be strongest at quarter-ends, when far more funds report and rebalance at once. We checked. Quarter-ends were weaker than ordinary months. That is the wrong shape for the story we were telling ourselves, and it was visible in the same output as the result we liked.

6. Zoom all the way out: which patterns survive being famous?

This gives a test you can apply to any market pattern, any strategy someone is selling you, and honestly to a lot of things outside markets. Do not ask whether the pattern is real. Ask:

Can the person on the losing side choose when or how?

If they have any flexibility — spread it out, do it quietly, route it elsewhere, wait a day — expect the pattern to fade as it becomes known. The obligation survives; the timing does not. If they have none — if the constraint binds at one moment with no alternative way to satisfy it — the pattern can persist indefinitely, no matter how many people know.

The difference is sharper than it sounds. Compare:

The obligation	Any way around the timing?	What happened
Restore a 60/40 portfolio	Yes — spread execution, algorithms, auctions, private trades	Faded to nothing by 2017
Maintain a fixed daily leverage ratio	Yes — swaps, staged execution, off-exchange	Also faded after 2017
Do not hold a position overnight	No — you hold the risk or you do not	Still present over 18 years

That last row is the shape that lasts. “Be flat by the close” is binary. There is no clever execution that lets you both hold a position overnight and not hold it. The only way to satisfy the constraint is to genuinely not own the thing — which means somebody else has to own it, and that somebody can be paid for the service.

Everything softer than that erodes. Not because the rule stops applying, but because people optimise around it — and people are very good at optimising around anything that costs them money in a predictable way.

The practical version. Next time someone shows you a pattern in markets, skip the chart. Ask one question: who is on the other side, and do they have a choice about when they trade? If they have a choice, you are looking at something that worked until people noticed. If they genuinely do not, you may be looking at something durable. Most patterns you will be shown are the first kind, and most will be presented using the full history, which is precisely where a decade of nothing hides behind a good first half.

Test pre-registered before running: the hypothesis, the measurement, and every pass/fail condition were written down in advance, including the condition that killed it. Clean daily futures data, 2008 to 2026, via an independent vendor. Statistical strength figures are t-statistics on the relationship between the equity-versus-bond drift within a month and the return over the final three trading days. Nothing here is financial advice.