Capital efficiency and accessible diversifiers

Question. How does the funding rule change the value of a diversifier, and which economically distinct return sources can a retail investor currently access at plausible cost?

Current answer. Funding is load-bearing. Selling the core to fund a sleeve and financing an overlay are different portfolio constructions, and the difference can be larger than the premium being studied. Trend is the best-supported accessible financed diversifier in the current evidence; gold is easy to finance but has weaker expected-return evidence; several other mechanisms appear distinct on paper but are inaccessible, expensive, short-lived, or poorly measured. Among the stacked trend wrappers a retail investor can actually buy, neither the funding rule nor the fee table separates the candidates — all three whose holdings are filed keep the whole gap, and the one advertising a ten-to-one cost advantage has none once its swap layer is counted — so age, size, survival, counterparty and tax placement decide. This is a finding about the present vehicle and evidence shelf, not about the number of return engines in markets.

Funding algebra

For a sleeve added by pro-rata sale of a base portfolio, the hurdle differs from a financed overlay by:

a_p - sigma_p²

where a_p is the base portfolio’s arithmetic return and sigma_p its volatility. In the 100%-equity reference case previously tested, that gap was about 2.44 percentage points a year—larger than the effects most sleeve experiments attempted to resolve. The expression contains no sleeve property. It explains why a pro-rata test can reject candidates that may still improve a leverage-matched financed construction.

This does not mean financing is free. A wrapper retains only the part of the gap left after financing, fee, collateral, rebalance, tax, and implementation costs. It also adds leverage, liquidity, counterparty, path, and behavioral failure modes. Funding must therefore be part of the hypothesis and comparator, not a default chosen for prudence or optimism.

A wrapper’s structure enters exactly once

Real wrappers are not a dichotomy between pro-rata sale and financed overlay. They are a continuum, and one number places a fund on it. For a wrapper delivering b of base and d of diversifier notional per dollar of capital,

dg/dw at w=0  =  (a_net - rho sigma_p sigma_d) - delta (a_p - sigma_p**2),
delta = (1 - b) / d

so the wrapper’s structure enters exactly once, as a multiplier on the gap above. delta is the base sold per unit of diversifier notional obtained, and 1 - delta is the share of the funding-rule benefit the wrapper keeps. The identity is implemented and tested in studies/wrapper_economics.py.

Two consequences bite on the real shelf.

• Gross notional per dollar decides nothing. A 50/50 equity-and-trend fund and a standalone trend fund both show 1.0x gross and both pay the gap in full; a 90/60 fund shows 1.5x and pays a seventh of it. A wrapper at 40% equity and 30% trend has delta = 2.0 and is worse than selling equity outright — a case the gross-notional figure cannot distinguish from the good one, and one with no name in the marketing vocabulary.
• An expense ratio is quoted in the wrong units. The hurdle is stated per unit of notional; a fee is charged on capital. The conversion is fee / d, so a wrapper charging 20 bp for 0.10 of notional is dearer than one charging 100 bp for 1.00.

Read the base leg from the whole filing, not from its largest line

The error this page has actually produced is not a bad model. It is reading one holding and stopping. MATE’s 2026-02-28 Form N-PORT shows an S&P 500 ETF at 49.8% of net assets, which is squarely inside the range above where a wrapper is worse than selling equity; the same filing shows a long E-mini S&P 500 future at 61.8%, which takes the base leg to 111.6% and delta to -0.116. At 2026-05-31 the two lines read 50.30% and 65.57%, a base leg of 115.87% and delta = -0.159. A stacked fund’s base leg is the sum of every instrument that delivers the base, and on these products at least one of them is a future.

CTAP repeats the error and the fix. Its 2026-03-31 filing shows an S&P 500 ETF at 70.41% of net assets; reading that line alone gives delta = +0.299, deep in the range where a wrapper is worse than selling equity. The same filing shows a long E-mini S&P 500 future at 32.23%, taking the base leg to 102.64% and delta to −0.027 — the tightest on the shelf. Two funds, two quarters apart, same mistake available and the same instrument fixing it.

Two caveats travel with any such reading. The index future is not separable into base completion and the trend book’s own equity position, because both sleeves trade equity-index futures and no filing tags a contract by sleeve; the contractual floor is the prospectus target, and for MATE that is 100% and delta = 0.00. And gross notional on a volatility-targeted trend book is an artefact of the risk target rather than an exposure: MATE’s whole derivative book ran 339% of net assets in February and 405% in May while its stated targets did not move.

A leg that has not been filed is not a missing measurement

JPFP commenced operations on 2026-05-27 and has filed no Form N-PORT. Its series is S000101300 in the SEC’s own ticker map, and that series appears in none of the 24 N-PORT filings its trust made for the 2026-05-31 period. Its first holdings filing belongs to the quarter ending 2026-06-30 or 2026-07-31 and is due 2026-08-29 or 2026-09-29, so delta is not computable before then. not filed, with a date attached, is the finding.

A cheap fee on the wrong category is not a cheap wrapper

SDMF was reported to this repository as a stacked equity-plus-managed-futures wrapper at 0.35%. Its 2026-03-31 Form N-PORT holds no equity ETF, no equity index future and no equity of any kind — 89.5% of net assets in Treasury bills, 4.4% in a money fund, and four total return swaps on DBi managed-futures indices. b is zero, so delta = 1.000 and it keeps none of the funding-rule gap: it is a standalone trend fund, arithmetically identical at the margin to selling equity to buy DBMF. Its 0.35% is cheap for that category and irrelevant to this one.

This is the same trap in a new place. Gross notional cannot tell a good wrapper from a bad one, and a headline fee cannot tell a wrapper from a trend fund. Only b and d, read off the holdings, say which object is on the table.

The fee table is the wrong instrument, in two different ways

First, financing. RSST files 0.00% of interest expense. MATE’s unitary fee excludes interest expenses and its Other Expenses line is 0.00%, estimated. JPFP’s unitary fee excludes them likewise. Futures financing is not an expense item: it sits inside the futures basis and is paid out of the contract’s return, so those zeros are accurate and uninformative. The one exception is CTAP, whose diversifier is a swap rather than a futures book: every one of its swaps files its financing leg as SOFR plus a spread of 0.00000000, which makes it the only wrapper here whose financing spread is stated at all.

Second, and larger, the diversifier’s own fee. A total return swap pays the reference fund’s return net of that fund’s fees, and Acquired Fund Fees and Expenses reaches direct holdings rather than a swap reference. So when a wrapper buys its diversifier as a swap on another fund, that fund’s expense ratio is charged to the investor and appears nowhere in the wrapper’s fee table. CTAP is the worked case: 0.28% gross and 0.10% net after a waiver, with 95.17% of net assets in a swap on CTA — an affiliated fund charging 0.75% with no waiver — plus 3.71% of CTA held outright, which is the part the 0.03% AFFE line does reach. All-in for the trend dollar is therefore about 0.81%/yr today and about 0.99% once the waiver lapses on 2026-12-04, against RSST’s 0.99% and MATE’s 0.97%. The two fact sheets report that near-tie as a ten-to-one gap.

The general rule: read what the diversifier leg is before reading what the wrapper charges. A fee table prices the wrapper, not the exposure, and the gap between the two is a whole fund’s expense ratio when the leg is a swap on a fund. It also travels with three costs no fee table carries — CTAP’s trend exposure is bilateral counterparty exposure to Bank of America at 82.48% of net assets and to Citibank at 12.70%, rather than to a clearing house; the reference fund is affiliated and the prospectus concedes the conflict; and a swap is not a Section 1256 contract, so the 60/40 treatment below does not reach 95% of its diversifier.

Where a stacked wrapper belongs, and why it is not the taxable account

The deferral argument that makes a financed overlay attractive does not survive the filings. Three mechanisms, all from primary sources, all annual, none optional.

• Section 1256 marks the futures book to market every year. Gain on regulated futures contracts is recognised at year end whether or not a contract is closed, and is 60% long-term and 40% short-term. MATE’s registration statement states this plainly; RSST’s 2026-04-27 statutory prospectus does not contain the string 1256 at all, and JPFP’s registration statement mentions it only inside the Section 988(a)(1)(B) currency election. That is a disclosure difference, not an exposure difference — the rule is statutory and reaches all three identically.
• A Cayman subsidiary converts the commodity sleeve into ordinary income. Each fund’s subsidiary is a controlled foreign corporation, so the fund must include its subpart F income in gross income annually whether or not it is distributed; “it is expected that all of the Subsidiary’s income will be ‘subpart F income’”; and that income “is generally treated as ordinary income, regardless of the character of the Subsidiary’s underlying income”. The loss side is worse: a subsidiary net loss “is not generally available to offset the income earned by a Fund and such loss cannot be carried forward” (J.P. Morgan ETF Trust SAI Part II, effective 2026-07-01, read 2026-08-22). RSST’s prospectus states the same expectation of annual distribution.
• A cash-creation ETF forfeits the in-kind shield. JPFP “expects to generally effect its creations and redemptions entirely or partially in cash”, so it “will be required to sell portfolio securities and subsequently recognize a gain” — on its equity leg as well as its overlay.

The placement that follows is the pre-tax account, not the Roth and not the taxable one. A wrapper’s tax cost is the conversion of return into currently recognised ordinary and short-term income. A traditional account already converts every dollar it holds into ordinary income on withdrawal, so that conversion costs nothing there. A Roth would spend the most valuable shelter in the portfolio on the sleeve with the least established expected return. A taxable account pays the cost in full, including phantom income in any year when the fund’s unrealised futures gain exceeds the cash it distributes.

Two limits on that conclusion. No SEC-standardised after-tax table exists for MATE or JPFP, because neither has completed a calendar year, so the only measured distribution drag on a trend wrapper here is RSST’s, over a window that is entirely a rising equity market — the wrong window for the year the sleeve exists for. And a 30%-of-portfolio wrapper consumes roughly nine tenths of a one-third pre-tax account, which concentrates that account in a single product before any return argument is made.

What the wrapper actually delivers, as opposed to what it holds. Everything above reads positions off a filing. The complementary question — how much of the trend exposure reaches the return — is answered from the same filings’ Item B.5 total returns and lives in loading comparability and wrapper exposure: RSST’s trend loading is +0.681 [+0.406, +0.955] over 31 months to 2026-04, against an equity beta of +0.979, with RSSB as a negative control at −0.101. A filed notional and a measured loading are different quantities and this page owns only the first.

Why levering equity is not the answer

Historical growth optimization of equity alone can select exposure no investor can survive. In the tested long US sample, the unconstrained optimum was around 2.2× with an almost total drawdown and decades under water. The useful conclusion is not that the estimate is the right leverage; it is that feasible exposure is set by drawdown, liquidity, withdrawal, and holdability constraints before an estimated growth optimum binds.

Accessible return sources

Use “engine” to mean a distinct economic mechanism and failure mode, not a ticker or a low full-sample correlation.

“One engine” was too strong a phrase. The evidence supports a narrower statement: among the financed retail vehicles examined, trend currently has the clearest multi-layer case. The shelf can change and does not define market ontology.

Counting engines is now a measurement rather than a judgement. Stacking and effective breadth computes 1'R^-1 1 on the excess returns themselves: five French factors across three regions plus trend are worth 10.23 of 16, one factor across three regions is worth 1.35 to 1.55 of 3, and five factors inside one region are worth 5.52 of 5. Geography is nearly free breadth and style is real breadth. That page also owns the probability arithmetic — what a stack of sleeves each n% likely to win is jointly likely to do — which is a different question from the growth arithmetic here and does not aggregate with it.

What sets an overlay weight

No single historical optimum should decide. Size an overlay across a surface of:

• investor drawdown and tracking-error limits;
• account capacity, taxes, and withdrawal needs;
• financing cost and stress behavior;
• expected net sleeve return;
• average and crisis-conditional dependence;
• product closure, method change, and operational failure;
• regret if the expected return is overestimated.

The current missing contribution and withdrawal path can change feasible sizing more than a small refinement to expected return. Until investor constraints and the joint portfolio test exist, exact overlay weights are scenarios.

One measured boundary is available and is not a scenario. For a financed trend overlay on a 65/25/10 global equity blend over 1990-07…2025-12, portfolio variance is minimised at 21.6% of notional, from w* = −rho × sigma_equity / sigma_trend and three measured numbers — no premium, no benchmark and no forecast. Its sensitivity is the correlation’s: the 95% interval on rho of [−0.277, −0.088] puts w* between 10.3% and 32.8%, so a 30% notional sits inside what the instrument supports but past the centre of it (stacking and effective breadth). The credit is much smaller than the pro-rata identity in marginal sleeve value suggests — about +0.03 pp/yr of growth rather than +0.22 at a 10% weight — because a substitution sells the base and removes its variance while an overlay keeps the variance and the return. That difference is the funding-rule gap seen from the variance side.

Global versus US is a separate decision

Local-currency long-history and modern USD evidence can point in different directions. That disagreement is not resolved by selecting the sample with the preferred answer. Currency, valuation, tax credit, survivor composition, and investable vehicle all matter. Treat the US/global split as a robustness and regret decision; do not bury it inside a diversifier weight.

Evidence and provenance

The funding identities and portfolio primitives are tested in research/. Experiment 011’s financed overlay result is recorded in research/artifacts/1e504717fa664a0dabf1cceca5c0a7d6/summary.md. Product facts and primary filing links — including each wrapper’s legs, delta, all-in cost, filing date and read date — are canonical in research/data-manifests/ and the typed client shelf under src/content/; they should be rechecked before a transaction.

The earlier long tables, one-investor tax examples, and product-by-product narratives are not repeated here. They remain in manifests, tests, client content, and Git history; the current synthesis owns interpretation rather than duplicating those stores.

Next informative tests

1. Run the portfolio-level tournament in the research agenda under explicit pro-rata and financed comparators.
2. Estimate crisis-conditional dependence and predeclare a monitoring boundary.
3. Build a point-in-time live-fund panel including closures and method changes.
4. Independently implement diversified trend from contract-level data.
5. Read JPFP’s first Form N-PORT and compute its delta. It is the only candidate whose structure is still unknown and the one whose 59 bp would reorder the cost ranking; the filing date and what the answer would change are on its shelf entry’s review trigger.
6. Revisit other mechanisms when a cheaper vehicle, longer live record, or better instrument changes the access or evidence constraint.

No candidate is promoted by this page. The conclusion is designed to be easy to update when the investor, evidence, or retail shelf changes.