Somewhere in a prospectus, somebody has written down that the question of whether a company counts as an infrastructure business may be settled, in part, by asking a chatbot.
The pitch is clean enough. American infrastructure needs rebuilding, certain companies will be paid to rebuild it, and an actively managed fund of roughly fifteen to twenty-five names will own them. Fifteen to twenty-five. Non-diversified, and the filing says so plainly, along with a note that as the fund grows it becomes more likely to hold over 10% of a single issuer with a small float.
Then you get to how a company qualifies. The manager may determine, using regulatory filings, analyst reports, trade publications, government data, and in the document's own words “outputs generated by large language models or other generative artificial intelligence tools”, whether a business derives half its revenue from infrastructure. The eligibility test for a fund about pouring concrete now includes a model that predicts the next word in a sentence.
There is more. One of the three selection factors is called Shareholder Concentration Level, and it scores who else owns the stock, marking down hedge fund, venture capital and private equity holders as less likely to hold and marking up large passive funds as more likely, so the fund can lean towards names about to be bought and away from names about to be sold. Estimated annual turnover is 100%, which is a brisk pace for assets whose defining characteristic is that they take a decade to build. And should the theme stop working, the fund may invest up to 100% of its assets in Treasuries, agency and government-backed positions, and cash, at which point Rebuild America becomes a money-market fund with an excellent name.
Sixty-five basis points, all in. I am not sure that is the expensive part.
Here is a more interesting piece of financial engineering, and almost nobody will read it as such because it arrives dressed as a fee cut.
A Berlin manager has taken a technology strategy running about one and a half billion euros and made it available as an exchange-traded share class. Note the wording. Not a new fund tracking the same strategy, and not a clone. The issuer is explicit that this is not a separate fund but a share class of the existing one, sharing the same portfolio, the same process and the same managers. The ETF and the old retail units are the same pot of money with different doors.
Now compare the doors. The traditional retail class carries a management fee of 1.87%, ongoing charges of about 1.9%, and a 3% entry charge on the way in. The exchange-traded class carries ongoing charges of 0.46% and no entry charge at all. Same portfolio. Roughly a quarter of the running cost. Anyone still sitting in the old class is paying a considerable annual premium for the privilege of having arrived early.
The catch, and it is a real one, is that the ETF class charges a performance fee of 15% of gains above a high-water mark, with no hurdle rate. The old retail class charged no performance fee at all. So this is not a fee cut so much as a fee restructuring: the fixed charge falls by well over a percentage point, and in exchange the manager takes a cut of the upside. On a strategy that has had a calendar year when it nearly trebled and another when it lost more than half, that trade is worth thinking about rather than skimming past. The manager's stated reason for the wrapper, incidentally, had nothing to do with fees. It was that the neo brokers would not stock the fund shares, so a successful strategy was gated behind plumbing rather than performance.
One more thing the documents will tell you if you look. The strategy's selection process leans on the firm's own AI platform to find the companies that benefit from AI adoption. Second product in this batch to outsource part of its judgment to a model. It is becoming a house style.
Somebody has built a fund about everywhere except America and given it a ticker that reads as an instruction to leave: X, US, M. Nobody at the firm explains the ticker anywhere, so that reading is mine, but I am confident enough.
To its credit, this one is unusually candid about what it is doing differently. Three deliberate deviations from standard momentum construction, all stated plainly: a six-month lookback instead of the conventional twelve, monthly rebalancing instead of semi-annual, and equal weighting instead of market capitalisation. Fifty to one hundred names, developed and emerging markets together, which is the actual point of difference, since most of the competition is one or the other but not both.
The claim I keep circling is that the approach identifies relative momentum even during market drawdowns.
Momentum's well-documented failure mode is precisely the sharp reversal, the moment when the leaders become the losers faster than a monthly rebalance can catch. A shorter lookback and a monthly cadence are a reasonable answer to that. They are not a solution to it, and there is no live track record yet to argue with. Seventy-five basis points is real money for an equal-weighted, rules-based portfolio.
There is a small coincidence in the same batch. A crypto momentum product is coming off the board in Zurich as this one arrives in New York. Momentum as a factor never really dies. It just changes what it is measuring.
This one is quietly the most interesting, for reasons that have nothing to do with the bonds.
At least 80% of assets go into impact bonds under the manager's proprietary framework, selected on a use-of-proceeds basis: affordable housing, community and economic development, renewable energy and climate change, natural resources. Crucially, the approach is inclusionary rather than exclusionary. The filing states flatly that the manager does not seek to exclude securities based on negative environmental or social characteristics. It screens in bonds whose proceeds do something specific and measurable, rather than screening out issuers for being disagreeable. That is a more honest construction than most of what gets labelled sustainable, and at thirty-seven basis points it is priced like a core bond fund rather than a conscience.
It also comes with one clause that does a great deal of quiet work. Government and agency securities are not restricted, and the manager considers investments in them to be consistent with the impact framework. Treasuries, in other words, qualify by being Treasuries. For mortgage-backed paper the framework looks through to the collateral pool and explicitly does not ask whether the sponsor itself measures up.
The detail I keep turning over is elsewhere in the same house. Sibling mutual funds recently had to insert a single word into their names, becoming the Short Duration and Impact Bond Fund and the Core and Impact Bond Fund. Not a strategy change. The manager's own notice says the changes do not affect strategy or portfolio management. It was a naming rule problem: the old name implied every bond had to be an impact bond, so the fix was to add a conjunction and redefine the 80% basket as a combination of two things, core bonds and impact bonds. An entire compliance exercise resolved by the word "and". Somewhere a lawyer billed for that, and was right to.
And then, as ever, the sensible one. Except this one turns out to contain the sharpest decision in the whole batch.
A sterling-hedged distributing share class giving British investors the MSCI USA, in the issuer's building-block range where ongoing charges start at three basis points. This class costs seven. No leverage, no theme, no language model deciding what belongs in it. The only exotic feature is that it is swap-based and Luxembourg-domiciled, with a single named counterparty, which sounds racy until you learn why.
The issuer documents the reasoning itself, and it is worth understanding. A physically replicating fund domiciled in Luxembourg or France surrenders 30% of its US dividends to withholding tax. An Irish physical fund surrenders 15%, thanks to the US–Ireland tax treaty. A synthetic structure can be paid the full index return, dividends included, because the swap counterparty collects the dividends instead. On a broad US index, the drag differential runs to tens of basis points a year, which is several multiples of the entire management fee.