Half of This Trade Is Built to Last

Every allocator has now read that money rotated into short-dated bonds and out of credit as the interest-rate outlook hardened. Most of that is true. The part that is not true is the part most often repeated, and the part nobody mentions is the one that decides whether the trade is worth putting on.

This note does three things. It shows what actually moved, using money into and out of bond funds rather than movements in yields, because a yield can move on opinion while a flow requires somebody to act. It separates the half of the move that is chasing headlines from the half that is not. And it sets out what follows for a portfolio, in three questions that can be answered in an afternoon.

We are not the only ones whose attention has landed on the long end of the curve. Writing and broadcasting in late August, the economist and author David McWilliams arrived at the same place from an entirely different direction, arguing on his podcast that the strain in the American bond market sits at the long end and that the pressure there is structural rather than passing. He gets there through public finances and the supply of new debt. We get there by following where the money actually went. Two methods, one answer. Where his argument is useful to ours, later in this note, we say so and credit it.

Duration is simply how long you are lending for. A short-dated bond fund lends for roughly one to three years, a medium-dated one for about five to ten, and a long-dated one for twenty or more. When rates are expected to rise, the textbook move is to shorten, because longer loans lose more value when rates go up.

That is the story told about this summer, and the evidence usually offered is a comparison between short-dated funds and long-dated ones. Set those two side by side and the conclusion looks obvious. It also leaves out the middle of the curve, which is where most bond money actually sits.

Over the six months, short-dated funds took +$120bn across 318 funds. Medium-dated funds took +$138bn across 393. Long-dated funds took +$10bn across 50.

So the middle of the curve took more money than the front end, and it is the largest of the three by a wide margin. The only part of the market that genuinely failed to attract money is the very long end, which is fifty funds and about $10bn over half a year.

Why the very long end alone should go begging is not a question our data can answer. It measures where money went, not why. One hypothesis worth naming comes from David McWilliams, who argues that the sums American technology companies are now borrowing to build artificial-intelligence capacity have grown large enough to compete with governments for the same pool of long-dated money, and are crowding it out. We have not tested that and it is his argument rather than our finding. It is consistent with what we observe, and it is the most concrete explanation we have heard for a gap this specific.

The move away from risk is real, but it happened on a different axis. Not by how long you lend, but by who you lend to.

Corporate debt splits into two broad tiers. Investment grade is debt from companies rated as more likely to repay, so it pays less. High yield is debt from companies rated as less likely to repay, so it pays more and falls harder when confidence goes.

Higher-rated corporate funds took money in five of the six months. Lower-rated funds could not string two positive months together, and the swings are violent: minus $11.7bn in March, then plus $8.8bn in April as much of the same money came back.

That volatility is itself the point. Money that leaves in one month and returns the next is not repositioning. It is reacting. Which raises the question this note is really about: when a flow moves, how do you tell whether it is a decision or a reflex?

There is a way to test it. Take each group's month-to-month movement and ask how much of it lines up with what the news is discussing at the time. Whatever is left over is money moving for reasons that have nothing to do with headlines: cash arriving, liabilities being matched, a scheme rebalancing on its own calendar.

The answers are not close together. Around nine tenths of the movement in corporate debt, at both quality tiers, tracks the news. Under a fifth of the movement in government debt does.

That single contrast is the most useful thing in this note, because it tells you what each leg is made of. The credit move is a sentiment position. It arrived on a story and it will leave on one. The government move is not. It looks like the slow, unglamorous business of matching cash to obligations, which starts for its own reasons and stops for its own reasons.

If the credit leg is news-driven, it is worth looking at the news directly rather than assuming. We measure the tone of coverage mentioning the central bank: how positive or negative the language is, day by day, where zero would be neutral.

Two things stand out. Coverage got steadily more negative, reaching its harshest reading of the summer in the final stretch. And the volume of dedicated coverage fell across the same three stretches, from 639 to 469 articles a day.

So this was not a story getting louder. It was a story getting darker while getting quieter, and it darkened over precisely the weeks money was leaving lower-rated credit. That is a closer fit than a simple count of headlines would have given, and it is the sort of thing a count of headlines would have missed entirely.

The last question is the one that decides whether any of this is tradeable: how long does a reading stay true?

Government debt records its strongest week of the summer on 17 August and a below-average week seven days later. Nothing about the underlying case changed in between. On a longer view the same series still reads as gathering pace, and both readings are correct. They are simply measuring different lengths of time.

The economic readings underneath the story behave the same way. One widely used measure of how the labour market is doing, published at the end of August and pointing to a cooling job market, was updated three days later on the same month's data. The new reading was nearly three times the first and it pointed the other way. Nobody erred. That is what these measures do as information arrives.

The same pattern appears on the policy side of the market. David McWilliams reports that when the US Treasury announced it would double its buying back of long-dated debt, an intervention aimed squarely at the part of the curve nobody wants, it bought roughly two days of lower yields before they returned to where they had started. That is his account and not our measurement, and we have not tested it. But a two-day effect from a deliberate policy action is the same phenomenon we measure in flows, reached from the opposite end: the reading turns over faster than any position built on it.

Three questions. None needs a new mandate, a new manager or a new system, and all three can be answered against a book you already hold.

What this looks like in practice is less dramatic than a reallocation. It is mostly a change to how two existing positions are governed. The government leg gets a normal review cycle, because nothing about it is time-critical. The credit leg gets a shorter one and an explicit trigger, because the thing that would invalidate it, a turn in coverage, is observable and moves faster than a quarterly meeting.