Alternative sleeves: which engines are distinct, and which are an expensive cash

Question. The candidate portfolio holds equity beta, equity factor tilts and one managed-futures overlay — two or three distinct return engines. The investor is explicit that they want more: “creative funds, assets that don’t correlate, assets that perform better in black swan events without excessive long-term drawdowns, crypto, anything.” Which further mechanisms are economically distinct, accessible to a US retail investor at a cost that leaves something over, and worth a weight?

Current answer. Two genuinely new engines, and one free improvement that is worth more than either. Duration-hedged credit is the largest change: this repository rejected credit on a +0.835 correlation to Treasuries measured with twenty years of duration attached, and once the duration is hedged out the correlation is +0.016 over 1,068 months, at the same return as long Treasuries for half the volatility and a third of the drawdown. Catastrophe risk is the only mechanism screened whose payer is not a financial market and whose trigger is a hurricane; its access problem has been solved and a pricing problem has replaced it. And a larger cash and short-Treasury allocation is the cheapest tail hedge on this panel — it beats almost everything sold as one. Everything else is already owned, an expensive form of cash, or an equity beta with a different name.

Two answers the investor asked for directly. Crypto: at most 1–2%, and only as a declared speculation, never as a diversifier — in the worst decile of equity months since 2015 bitcoin’s mean return was −7.51% and it was positive in 1 of 13 of them. Explicit tail hedges: no. The bleed is measured at roughly 12 percentage points a year against the index, and the convexity it buys is not statistically resolvable in any asset on this panel.

Decision it informs. What to add to the construction, at what weight, in which account, and what evidence would move each weight.

Out of scope. The equity share (setting the equity share), the trend overlay’s own sizing (trend, live funds), and the current construction as a whole (recommendation).

as of 2026-08-22 for product facts; each carries its own source and read date. The measured tables regenerate from studies/_stress_dependence_tables.py and the arithmetic is pinned in research/tests/unit/test_studies_stress_dependence.py. Everything measured here is exploratory: no specification was frozen before the numbers were seen, the stress windows were chosen by eye from the standing episode list, and four of the legs are AQR vendor series their author reconstructs on every release. The stress windows and the tail quantile are hypothesis-bearing analytical choices and owe a ledger entry; the module and its tests are committed so the choice is inspectable.

---

1. What the data says before any product is named

1.1 The panel

Eight candidate engines against US equity, each on its own longest window, every leg an excess return over cash so the rows are comparable. The base is Ken French Mkt-RF and cash is the same file’s RF; over 1926-07…2026-06 that base returned +6.86%/yr excess at 18.38% volatility, Sharpe 0.45, with a maximum drawdown of −84.6% relative to cash.

Sources and their traps: long Treasury and unhedged corporate are Goyal–Welch ltr and corpr less Rfree; duration-hedged credit is AQR CORP_XS, defined as the corporate total return less a duration-matched government return and therefore the only true credit-spread series held — it may never be summed with the Treasury leg and it ends in 2014-12; commodities are AQR’s equal-weight long-run excess series; trend is AQR TSMOM, a vendor series that states no fee, transaction cost, slippage or financing basis anywhere; gold is the World Bank Pink Sheet less cash and an assumed 25 bp carry, from 1975-01 because private US bullion ownership was illegal before 1974-12-31; bitcoin is FRED CBBTCUSD, one venue’s daily print rather than the CME CF rate an ETP prices against. Provenance and licence for each is in the evidence base.

Two rows deserve a second look. Duration-hedged credit has the best Sharpe and by far the shallowest drawdown of any long-only engine here — 0.53 at 4.12% volatility, against equity’s 0.42 over identical months. And trend’s headline Sharpe of 0.98 is gross of everything; this repository’s own live-fund panel measures what the funds actually paid (live managed futures), and the last 78 months of the same vendor series returned +1.95%/yr, not +12%.

1.2 There are two kinds of shock, and no single asset covers both

Cumulative excess return inside each of the standing named episodes. — means the panel does not reach the window; * marks partial coverage, so the number beside it is not the episode.

Read down the columns and the shocks separate cleanly:

• Growth and deflation shocks — 1929-32, 1987, 1998, 2000-02, 2008-09, 2020 Q1. Treasuries pay in every one. Commodities lose in four of six.
• Inflation and rate shocks — 1973-74, the late 1970s, 2022. Treasuries lose in all three, by up to 40%. Commodities and gold pay.
• Trend is the only engine positive in both kinds, and it has no data before 1985, so it has never been observed through an inflation shock larger than 2022.
• Credit’s hedged spread is the mildest loser in the growth shocks and nearly immune to the inflation ones — −3.2% through five years of the late 1970s against the Treasury leg’s −40.5%. That is the whole point of hedging the duration out.
• BAB is a trap. Its full-sample correlation to equity is −0.139, which reads as a diversifier. It lost 30.6% through 1929-32 and 31.6% through the GFC, because its mechanism is selling leverage to people who cannot borrow, and a crisis is exactly when leverage is withdrawn. This matters here because the investor already holds equity factor tilts that lean the same way.

1.3 In the lower tail, almost every candidate is a worse cash

The black-swan question stated as an estimand. Split months by the base’s own return, take the worst decile, and ask what the engine did. offset at 10% is what swapping a tenth of the equity base into the engine adds back in the average worst-decile month; same for cash is the identical swap into T-bills, on the identical months.

Four readings, and the third is the one that should change a portfolio:

1. Trend is the only engine that materially beats cash in the lower tail — +1.08% against +0.82%, 26 bp a month of genuine protection, delivered in 69% of those months rather than by one enormous outlier.
2. Long Treasuries buy two basis points a month over T-bills. +0.94% against +0.92%. The entire crisis-hedging case for twenty-year duration, on a century of data, is 2 bp per worst-decile month — bought with 8.42% of annual volatility, a −59.1% drawdown, and a −40.5% loss through the late 1970s that cash did not have.
3. Gold buys twelve basis points a month over T-bills, at 16.24% volatility and a −91.2% peak-to-trough. This is the same conclusion as marginal sleeve value reached by a different route, and it is why gold keeps failing: it is not that gold does not diversify, it is that cash diversifies nearly as well for nothing.
4. Commodities and bitcoin are worse than cash in the lower tail, and bitcoin is worse by 75 bp a month. Two assets frequently sold as diversifiers are, on this measurement, negative-value tail hedges.

Conditioning on the base’s own magnitude truncates its variance, so ρ low and ρ high are biased toward zero and are comparable with each other, never with ρ full.

1.4 Nothing on this panel is convex, and the only measured convexity has the wrong sign

Fit engine = α + β·equity + κ·min(equity, 0) with Newey-West standard errors. κ is the convexity: negative κ means the engine’s slope against equity falls when equity falls, which is what “performs better in a crash than a linear exposure would” means as an estimand. α is the per-month price of the shape.

• Not one engine has statistically resolvable convexity. The largest |t| on κ among the seven non-BAB rows is 1.37. What looks like crisis protection in §1.3 is a low beta plus a positive mean, not a payoff that accelerates.
• Trend’s crisis case is an alpha, not a shape. α = +0.91%/month at t = 3.09 on a vendor series with no costs in it; κ is indistinguishable from zero. That is consistent with the evidence base, where the crisis-conditional trend benefit has ≈4.4 effective observations and cannot be resolved at all.
• BAB is measurably concave at t = 3.49: β = −0.264 when equity rises and +0.118 when it falls. This is the sharpest available demonstration of the charter’s rule that low average correlation is only an admission signal.
• Bitcoin is a levered equity beta. 1.53 up, 1.62 down, no convexity, and an α of +4.46%/month whose t is 1.44 — economically enormous and statistically nothing.

1.5 What a sleeve is worth, at the weight anyone would actually hold

Realised marginal growth of a pro-rata-funded sleeve (sell the equity base), which is Experiment 010’s rule and the least favourable one for a diversifier. No trading cost is charged; the gold leg carries its 25 bp.

Units are pp/yr of realised geometric growth. Read this against the design’s own floor: the evidence base measures the marginal-sleeve instrument’s MDE₈₀ at ≈0.58 pp/yr, so every row except bitcoin’s lies inside the noise, and bitcoin’s row is 137 months containing the largest bull market in the asset’s history against a matched-volatility floor of 15.58 pp/yr. The one thing in that table that is not noise is the last column: a bitcoin sleeve is the only candidate here that made the portfolio’s drawdown deeper at every weight tested.

---

2. Candidate map

Every mechanism screened, with who pays and why they keep paying, what the payoff looks like, and the verdict. Access is separated from evidence throughout: a missing retail vehicle is an implementation finding, not proof that the return source is absent.

---

3. Crypto: the investor asked, so here is the arithmetic

Verdict. At most 1–2% of the portfolio, funded from the speculation budget rather than from the defensive sleeve, held in a taxable account, and labelled a speculation rather than a diversifier. Zero is also defensible. Anything above about 5% is a leveraged equity position that the investor could obtain more cheaply and with a shallower drawdown by holding more equity.

The mechanism, stated honestly. There is no cash-flow claim. A bond pays a coupon because a borrower is contractually obliged; an equity pays because a firm earns; a cat bond pays because an insurer needs capacity. Bitcoin’s expected return is entirely a claim that the future marginal buyer will pay more than today’s. That is not a risk premium and it must not be written as one. What can be defended is narrower and worth stating: a fixed supply schedule, a settlement network with no counterparty, and a payoff that is not a claim on any government’s solvency. Those are properties, not premia.

What is measured, on 137 months of FRED CBBTCUSD:

• Equity beta 1.526 up and 1.616 down, κ indistinguishable from zero. A 2% bitcoin sleeve is, to first order, 3% more equity plus a large idiosyncratic risk.
• ρ to equity +0.342 over 137 months, and the 81-month sub-window reads +0.531, which is outside the 0.5 boundary at which the repository’s own admission arithmetic stops being usable (evidence base). Correlation has risen as the asset has been financialised — the direction that removes the case.
• In the worst decile of equity months, mean −7.51%, positive in 1 of 13. It has never been observed doing anything else in an equity crisis. −10.5% through 2020 Q1 and −58.3% through 2022, when equity fell 25.3%.
• Maximum drawdown −75.9% on monthly data over a window in which equity’s was −25.3%.
• Its measured α of +4.46%/month carries t = 1.44. Against equity at matched volatility the measured gap is +0.10 pp/yr against a floor of 15.58 pp/yr: 137 months of the best returns in the asset’s history cannot distinguish it from the S&P 500.
• And the most recent evidence is the same evidence again. In the first half of 2026 bitcoin fell −33.2% while US equity returned +9.9% — measured on the same panel, and independently corroborated by the CME CF BRRNY rate falling from $87,549 to $58,605 over the same six months (CF Benchmarks, read 2026-08-22). It has since recovered to $77,338. A 33% fall against a rising equity market in the sample’s final six months is not a diversifier failing to help; it is an asset behaving as the beta measurement says it does, with the idiosyncratic risk on top.

Why not zero, then. Because the loss from a 1–2% position that goes to zero is 1–2%, the position is not correlated with the investor’s human capital, and the investor has explicitly asked for it. A holding an investor wants and understands is easier to keep than a holding they resent, and holdability is in the objective. That is a behavioural-and-preference argument, not an evidence argument, and it should be written down as one.

Vehicle and account. Spot ETPs are 1933-Act grantor trusts, not 1940-Act funds: no K-1, gain and loss flow through as if the holder owned the coin, brokers report on 1099-B, and each sale of coin to pay the sponsor fee is a taxable disposition for the holder. The 28% collectibles rate that applies to a physically-backed gold trust is not asserted here — IBIT’s own prospectus tax section contains no collectibles discussion at all, and the IRS treats bitcoin as property that can be held as a capital asset. Fees and sizes from each trust’s Q2-2026 Form 10-Q, net assets as of 2026-06-30, read 2026-08-22:

Two rows have been re-read since, and both hold. HODL’s waiver expiry has now passed, and the same Form 10-Q (filed 2026-08-13, read 2026-08-24) states the rate on the far side of it: the Sponsor Fee is 0.20% of average daily net assets, the waiver of it on the first $2.5bn ran “from November 25, 2024 through July 31, 2026”, and “[a]fter July 31, 2026, the Sponsor Fee will be 0.20%”. EZBC’s Q2-2026 10-Q (filed 2026-08-14, read 2026-08-24) accrues its sponsor fee “at an annualized rate equal to 0.19%”. Data aggregators showed 0.25% and 0.29% for these two on 2026-08-22 and both readings are wrong; the fee of a 1933-Act grantor trust is in its own quarterly filing, and a fee under waiver is where an aggregator is most likely to be stale. Sizes move faster than fees: IBIT was $58.78bn at 2026-08-22 against the $43.4bn above, which is a later date and mostly mark-to-market rather than flow, and every other net-asset figure in the table is still as of 2026-06-30.

Only six of eleven US spot bitcoin ETPs price against the CME CF rate, and the methodologies genuinely differ: at 4:00 p.m. ET on 2026-06-30 the same bitcoin was marked at $58,605 (BRRNY, an hour-long volume-weighted median across seven venues), $58,717 (Lukka Prime), $58,732 (CoinDesk) and $58,745 (Grayscale’s single principal market) — 23.8 bp of dispersion at a single instant, from audited filings. That is small, it is the benchmark working rather than failing, and it is the reason this repository’s own FRED CBBTCUSD leg is labelled one venue’s print rather than the asset.

Hold it in taxable: it pays no income, so a tax-deferred account wastes shelter a bond would use, and a taxable holding preserves the loss-harvesting option that a 75% drawdown makes unusually valuable. Avoid the crypto covered-call and “income” funds entirely — their headline distribution rates run 27% to 73% against 30-day SEC yields of 0.3% to 3.8%, and at least one fund’s most recent 19a-1 notice estimates the distribution as 100% return of capital. That is your own money handed back with a tax form attached.

What would change this — and one condition has moved. The three triggers were a correlation to equity back below +0.2 on a window containing a recession; a cash-flow claim with a contractual payer; or a realised equity bear market in which the asset does not fall harder than equity.

The second has partially arrived, and not for bitcoin. Rev. Proc. 2025-31 created a safe harbour letting a grantor trust stake proof-of-stake assets without losing trust classification, and staking ETPs are now live and material: Grayscale’s ETHE and ETH stake about 82% and 83% of their ether and recognised $18.8M and $16.9M of staking income in H1 2026 against $0 in 2025, and BlackRock launched a separate staked trust (ETHB, 86.9% staked) rather than turning staking on inside ETHA. That is a genuine contractual payer — the protocol pays for validation — and it is the first thing in this family that is a yield rather than a price expectation. It is also not bitcoin, it carries slashing and validator risk, the sponsor takes a cut of the reward, and the income is ordinary. It is a reason to reopen the ether question with a real estimand, not a reason to raise a bitcoin weight.

---

4. Tail hedging: the bleed is measured, and the cheaper substitutes win

Verdict. Reject explicit tail hedges. The investor’s stated requirement — “assets that perform better in black swan events without excessive long-term drawdowns” — is best met on this evidence by (a) a larger short-Treasury and cash allocation, (b) the trend overlay already held, and (c) not holding the levered and concave things in §1.4. An option-based hedge converts a diffuse long-term drawdown into a certain annual bleed, which is a different risk, not less of it.

The mechanism, stated honestly. A long put is the short side of the variance risk premium. Index options have been persistently rich relative to subsequent realised volatility because someone is being paid to bear crash risk — and a tail-hedge fund’s investor is the one paying. The strategy is therefore negative expected return by construction, and the case for it can only ever be that a convex payoff at the right moment is worth more than the premium. That is a claim about the path, and it requires the holder to monetise the hedge at the bottom, which is the moment they are least likely to.

The bleed, from issuer-published standardised returns.

Read 2026-08-22 from the TAIL factsheet PDF, the Alpha Architect CAOS page and the SSGA SPY page. Two cautions that belong with the numbers: do not use the figures rendered on cambriafunds.com/tail — that page is JavaScript-hydrated and returns numbers irreconcilable with the issuer’s own PDF; and CAOS’s ten-year record is inherited from the Arin Large Cap Theta mutual fund, a differently-mandated, higher-turnover predecessor, with only about 3.4 years of it as the current ETF (497K).

The bleed is roughly 12 pp/yr against the index on the honest comparison (CAOS, identical window, both figures issuer-published) and worse on the pure hedge. To first order a 10% TAIL sleeve funded pro rata would have cost about 2.2 pp/yr of portfolio growth over its life — 0.10 × (−7.15 − 15.35) — which is roughly four times the entire measured marginal value of the trend overlay at the same weight in §1.5. That is arithmetic on two annualised returns over slightly different windows, not a backtest.

The calendar years say something the annualised figure cannot, and it is the decisive thing. From TAIL’s own prospectus bar chart and the issuer’s return feed, read 2026-08-23:

TAIL made 6.98% in 2020, the year of a 33.9% peak-to-trough fall in the S&P 500. And it lost 13.15% in 2022, a year the S&P 500 returned −18.11%. A put-buying programme collateralised with Treasuries loses on both legs when a bear market is a slow grind with rising rates rather than a volatility spike, and 2022 was the second kind. A crash hedge that loses money in a bear market is a bet on the shape of the decline rather than a hedge. On the same feed the fund’s since-inception return to 2026-07-31 is −7.30%/yr, −50.64% cumulative, its trailing five years −9.06%/yr, and its assets $147.6M at 2026-08-21. (TAIL’s 2025 calendar return is deliberately absent: no primary source publishes it yet — the issuer shows trailing periods only and the next prospectus bar chart has not been filed. Cambria’s own two sources also disagree by one day on the inception date, 2017-04-05 in the SEC 485BPOS against 2017-04-06 on the fund page.)

And the distribution has the other tail too. Simplify’s CYA, launched 2021-09-14, reported −99.16% since inception on the issuer’s own site at 2023-12-31, took a 1-for-20 reverse split effective after the close on 2024-02-09 (SEC 497, filed 2024-01-23) and was liquidated “on or about March 14, 2024” (SEC 497 filed 2024-02-20). Its last filed net assets, in the Form N-PORT for 2023-12-29, were about $2.16M; no filing covers the final quarter, so the frequently repeated “$1.7M at liquidation” is unconfirmed and is not used here. A total loss on a product sold as crash protection, over a period that contained a bear market. CYA is also the seventh ticker in Simplify’s run of alternative-strategy closures — SCY, SPQ, FIG, NXTV, EQLS and WUSA all last traded 2025-05-23, about fourteen and a half months after CYA — which is the survivorship point: a shelf screened today shows the products that lived.

One point in TAIL’s favour, because it changes what the number means without changing the verdict: TAIL is about 91% ten-year Treasuries and only about 5% options, so its −7.15%/yr is not all option bleed — a large part of it is the worst bond market in forty years, and §1.2 already shows that Treasuries lost 13.6% through 2022. The option leg’s own cost is nearer the adviser’s stated spend of roughly 1% of assets a month on puts. That makes TAIL a worse proposition rather than a better one for this investor: the Treasury part they can buy for 3 bp, and the part they are paying 59 bp to obtain is the part with negative expected return.

Long volatility is worse, and the reason is arithmetic. VIX futures were in steep contango across the whole visible curve on the last settlement before this page was written: spot VIX 15.13 against 17.50 for September and 19.15 for October (Cboe VIX futures, read 2026-08-22; the page carries no timestamp, so these are most likely the 2026-08-21 settlements). A constant-one-month index rolling M1→M2 pays roughly that spread every cycle. The consequence is in the issuer-published returns: VIXY −46.56%/yr over ten years and UVXY −71.37%/yr, NAV as of 2026-07-31 (ProShares). These are also commodity pools, not 1940-Act funds, and they issue K-1s; VXX is an unsecured Barclays note maturing 2048, callable by the issuer at its sole discretion, and subject to UK bail-in powers, with a tax treatment its own prospectus calls uncertain (424B2, read 2026-08-22). None of this belongs in a long-horizon portfolio.

Buffered and defined-outcome funds are the same trade at a worse price. This repository has already priced the cap-and-buffer package from 1,183 overlapping twelve-month price returns and found it worth −2.4 to −4.1 pp/yr, with the funds’ realised −4.1 landing inside that range from disjoint data (evidence base). Current shelf facts are consistent: the July buffer series carries a 0.79% fee for a 9% buffer against an 18.14% cap, and the 100%-protection two-year funds cap at 13.61% and 18.32% (Innovator factsheets, read 2026-08-22). Two structural notes worth keeping: Innovator’s benchmark on its own factsheets is the S&P 500 price-return index, which omits dividends and flatters the comparison; and because the FLEX options are written on SPY and VOO rather than on a broad-based index, they are generally not §1256 contracts and get no 60/40 treatment — unlike TAIL’s and XTR’s listed SPX options. Innovator has been an indirect wholly-owned Goldman Sachs subsidiary since 2026-04-01 (497), which is a counterparty-and-continuity fact rather than a pricing one.

What the investor should hold instead, and why it is not a compromise. §1.3 measures it: in the worst decile of equity months, swapping 10% of equity into T-bills adds back +0.92% in the average month, at zero fee, zero drawdown and zero path risk. Long Treasuries add +0.94% — two basis points more — and charge 8.42% of volatility and a −59% drawdown for the privilege. Trend adds +1.08%. No option structure on the current shelf offers a lower-tail offset larger than those net of its bleed, and the two that come closest do it by holding Treasuries and spending 1% of assets a month on options.

One structural defensive worth naming so it is not re-screened. BTAL (AGF U.S. Market Neutral Anti-Beta) is the non-option version of the same idea, and it is BAB’s short side. Its gross expense ratio is 1.65% and its net is 1.40% — the 0.45% “adjusted” figure excludes dividend and brokerage expense on short positions, which is where the cost actually is — and its since-inception NAV return is −3.63%/yr against the S&P 500’s +15.41% (AGF factsheet, as of 2026-07-31, read 2026-08-22). It also changed from index-tracking to active in 2022-02. §1.4 explains the shape: the anti-beta leg is BAB’s mirror, and BAB’s measured concavity is t = 3.49. Reject.

---

5. Catastrophe bonds: the vehicle problem is solved and a price problem has replaced it

Verdict. Admit the mechanism; hold at 0–3% or wait. This is the only candidate on the page whose loss trigger is meteorological rather than financial, and the access finding has genuinely changed — a 1940-Act ETF now exists with daily liquidity at 1.58% net. But the risk spread has compressed by roughly half since 2023, and the net-of-fee record of actual retail vehicles over nine years is about one percentage point a year over cash. The reopening condition is a number, and it is stated below.

The mechanism, and why it is the best one on this page. An insurer or reinsurer with a concentrated peak-peril exposure — Florida wind, California quake — cannot retain it under its own regulatory capital rules and cannot always cede it to the traditional reinsurance market at an acceptable price. It sells the risk to capital markets through a special-purpose vehicle: the investor’s principal sits in Treasury money-market collateral, the investor receives that collateral yield plus an insurance risk spread, and loses principal if a defined event occurs. The payer is a balance sheet with a statutory capital constraint, and the loss trigger is a hurricane. Nothing else screened here has a return whose driver is outside financial markets entirely.

What it has actually paid, and the two indices are not interchangeable.

Swiss Re figures as reported by Artemis and, for 2022, from Swiss Re’s ILS Market Insights March 2023; ILS Advisers from Artemis. Read 2026-08-22. Swiss Re’s pages return HTTP 403 and were not fetched directly.

Compound the second column and the case gets much quieter. The ILS Advisers fund index returned +3.31%/yr geometric over 2017–2025 — nine years that include the three best in the market’s history — against a US T-bill rate that averaged something close to 2.3% over the same span. That is roughly one percentage point a year over cash, net of fees, from the actual vehicles. The gap between the two columns is the cost-and-implementation gap that this entire page keeps rediscovering: Swiss Re ran 17.29% in 2024 against 12–15% for most managed funds.

The 2022 week is the shape of the risk. The Swiss Re index was −0.35% at the half-year and then fell −9.65% in the single week to 2022-09-30 on Hurricane Ian. SHRIX’s worst quarter is Q3 2022 at −10.29%. This is not a smooth carry.

Cascading loss is the failure mode that a correlation cannot see. Roughly 36% of the outstanding market is annual-aggregate structure ($22.78bn against $40.23bn occurrence), where each qualifying event erodes the retention beneath the attachment point, so a season of moderate events leaves a bond exposed to a later event it would have survived standalone. The 2024–25 sequence is the clean illustration: Helene and Milton eroded aggregate retentions in autumn 2024, and the January 2025 Palisades and Eaton wildfires landed inside the same annual risk period for Allstate’s Sanders Re programme — two tranches were marked down about 50% purely on the increased probability of attaching over the remaining risk period, with no payout having occurred. Peak-peril concentration compounds it: explicitly named-storm buckets are 24.2% of outstanding and every bucket containing any US wind totals 56.1%, so “diversified across perils” is a weaker claim than it sounds.

The price. This is the part that decides the weight.

Artemis market data, read 2026-08-22. Q2 2026 was the first quarter below 4% spread-above-EL in twenty. The arithmetic that follows is simple and should be done before buying: a 5.53% risk spread against a 2.50% market-average expected loss leaves about 3.0 percentage points of gross expected compensation, from which a retail vehicle takes 1.58% to 2.36%. What is left is roughly one point a year of expected excess over the collateral yield — which is exactly what the fund index has delivered — for an asset that can lose 10% in a week.

Access, which is the part that has changed.

Read 2026-08-22 from SEC filings and issuer pages. Four corrections to the usual account of this shelf, each of which changes a conclusion. SHRIX is not an interval fund — it is an open-end mutual fund with daily redemption, which removes the liquidity objection entirely for an investor who can meet the $250k Class M minimum. SRRIX is not a cat bond fund: at 2026-04-30 it held 19.4% event-linked bonds and 65.2% private quota-share paper, so comparing its 2.36% with an ETF’s fee is not like for like. CNRLX’s 0.99% is materially understated — its own prospectus says the Neuberger Berman segregated accounts have fees “not reflected in the fee table,” and they never appear as acquired-fund expense. And none of these funds imposes an accredited-investor or qualified-purchaser test; the gate is minimum size, not investor status.

On the ETF specifically, which is the vehicle that makes this reachable at all: it holds 144A cat bonds inside a 1940-Act wrapper because Rule 144A securities are not automatically illiquid — a board-approved liquidity risk management program classifies each holding against the 15%-of-net-assets illiquid cap, and at 2025-12-31 144A paper was 85.4% of net assets with 100% of holdings at Level 2 and no Level 3 at any point in 2025. That is a defensible answer to the obvious objection. Two cautions remain: it trailed its own stated benchmark by 430 bp over its first nine months (+5.87% against the Swiss Re index’s +10.17%), and its tailored shareholder report shows costs actually paid of 2.00% annualised, not the 1.58% cap.

Where it belongs and what would change the verdict. Tax-deferred, without exception: the return is almost entirely ordinary income and short-term gain. The reopening condition is the spread-to-expected-loss multiple. At 2.21× secondary and 2.40× new issue this is the least attractive entry the market has recorded in the period observed here. At 3.5× or above, with the retail fee unchanged, the arithmetic gives roughly two and a half points a year over collateral for genuinely non-financial risk, and the weight should rise. That is a monitoring boundary, it is publicly observable weekly, and it is the reason to keep this family open rather than screen it again from scratch.

---

6. Duration-hedged credit: the rejection was about the instrument, not the mechanism

Verdict. Admit. This is the largest single change this page makes. The earlier finding that “credit is not a second engine, its correlation to Treasuries is +0.835” is reproduced here at +0.826 — and it is a property of the unhedged corporate leg. The duration-hedged credit spread correlates +0.016 with long Treasuries over 1,068 months. It is a separate engine, and it was rejected because the instrument that measured it had twenty years of duration bolted to the front.

The mechanism. Insurers, pension funds, banks and rating-constrained mandates must hold investment-grade paper and cannot bear either default loss or mark-to-market volatility in it. They pay a spread above a duration-matched Treasury for that. The payer is a balance sheet with a regulatory or actuarial constraint, which is why the premium has survived a century of publication: the buyer is not choosing to bear the risk cheaply, they are required not to bear it at all.

What is measured. AQR’s CORP_XS, defined as the corporate bond total return less a duration-matched government return estimated by rolling empirical-duration regressions, 1926-01…2014-12. On the 1,062 months it shares with the equity and Treasury legs:

Identical return to long Treasuries, at half the volatility and a third of the drawdown. It is also nearly distinct from everything else on the panel: ρ −0.082 to trend, +0.064 to commodities, −0.050 to gold.

The honest counterweight, because this is a risk-premium and not a hedge. Its mean in the worst decile of equity months is −0.24%, its lower-tail offset at 10% weight is +0.91% against cash’s +0.94%, and it lost 13.1% through the GFC while Treasuries gained 13.4%. Credit’s left tail is the same corporate-default state that kills equity. Add it for return breadth, never for crisis protection.

Which is why the recommendation is a substitution, not an addition. Holding half the defensive sleeve in each is better than holding it all in long Treasuries on almost every axis measured:

The blend is flat through the GFC and loses 23% rather than 41% through the late 1970s, because the two legs fail in different states and correlate +0.016. That is what breadth is supposed to look like, and it is available inside the allocation the investor already intends to make.

The scale problem, stated plainly. At 4.12% volatility a 10% sleeve contributes about 21 bp/yr of gross excess return, which is inside this repository’s 0.58 pp/yr detection floor. This engine cannot be made to matter at a satellite weight. It matters as a replacement for defensive assets already held, or not at all. Levering it to Treasury volatility (2.03×) doubles the return to +4.20%/yr at the same Sharpe — and takes the drawdown to −39.2% and the GFC loss to −26.6%, which is the whole point of not doing that.

What would change this. The series ends in 2014-12, so it has never been measured through 2020 or 2022 — the two episodes in which a hedged-credit sleeve’s behaviour would be most informative, since March 2020 was a liquidity event in exactly this instrument. Acquiring a duration-hedged credit series that reaches 2026 is the single most decision- relevant acquisition this page identifies. It also has no net-of-cost version: CORP_XS is a vendor construction, gross of fee, spread and financing, and the retail wrapper’s real cost is what decides whether +2.13% survives.

6b. Retail access, and the conditions the sleeve would be bought into

The vehicle exists, it is cheap, and it does exactly what the series describes. The iShares rate-hedged funds are fund-of-funds: they hold the underlying credit ETF and overlay centrally-cleared interest-rate swaps at the 1, 2, 3, 5, 7, 10, 15, 20 and 30-year points, weighted to the underlying’s composition and rebalanced daily (iShares product brief, read 2026-08-22). ProShares does the same job with short Treasury futures and states an explicit target duration of zero (ProShares HYHG).

Fees and durations as of 2026-08-20/21 from the iShares and ProShares product pages, read 2026-08-22. The net figures depend on contractual waivers that expire 2027-02-28; the waiver sets total expenses equal to the underlying ETF’s acquired-fund fee plus 10 bp (LQDH, IGBH) or 5 bp (HYGH). If a waiver lapses the cost roughly doubles, and at a 2.13%/yr gross premium that is not a rounding error.

The category is shrinking, which is an implementation finding and should be recorded as one. iShares liquidated AGRH, the rate-hedged aggregate fund, effective 2026-08-17 (board approval 2026-06-12, trading halted 2026-08-13; prospectus supplement), and EMBH is no longer on the iShares screener. No new entrant was found in 2024–2026. A category that is losing funds while its mechanism is intact is a liquidity and continuity risk, not a verdict on the premium.

A second, cleaner instrument for the same idea: AAA CLO tranches. These reach near-zero duration natively — they float over SOFR — rather than by overlaying a derivative on a fixed-rate bond, so there is no swap carry, no daily rebalance, and no waiver to lapse.

Janus Henderson factsheets, data as of 2026-06-30; VanEck and Eldridge pages read 2026-08-22. The failure mode is different from corporate credit and must not be assumed away. A AAA CLO tranche is structured credit over leveraged loans; its risk is the correlation of loan defaults and the behaviour of the structure’s tests, not a single issuer’s balance sheet. JAAA launched in 2020-10, so no CLO ETF here has a March 2020 record, and March 2020 is precisely the liquidity event that would test it. Treat this as a candidate with an excellent instrument and a short history — the mirror image of the corporate series, which has a long history and a stale instrument.

And the conditions it would be bought into are the least favourable part of the case. As of 2026-08-20/21, read from FRED and Treasury.gov on 2026-08-22:

Two consequences, and they point in opposite directions.

Against the credit sleeve: 82 bp is a tight spread. The +2.13%/yr measured over 1,062 months was earned across the full range of spread regimes including 1932 and 2008. Buying the engine at 82 bp of gross spread, before expected default loss and before a 14–24 bp wrapper, is buying it near the bottom of its own distribution. This does not refute the engine; it says the entry point is poor and the position should be small, or built over time, or sized to the spread.

For the defensive allocation generally: a 2.40% ten-year real yield and a 3.00% thirty-year real yield are the strongest contractual terms available to this investor anywhere on this page. A TIPS held to maturity at 3.00% real is a contractual line — the certainty class the repository reserves for statutes and accounting identities — against gold’s measured +1.75%/yr with a −91% drawdown and no coupon at all. When the risk-free real yield is 3%, the hurdle every sleeve on this page has to clear rises with it, and the honest consequence is that more of the answer to “what should I add” is now “a better-constructed defensive allocation” than it was when real yields were negative.

6c. TIPS and nominal bonds are one engine, and that is not an argument against holding TIPS

Two facts that are frequently confused. TIPS and nominal Treasury funds are not two engines: they correlate +0.761 to +0.851 across eighteen bond and TIPS ETFs’ filed monthly returns and +0.798 on the modelled long series, against the 0.75 threshold capital efficiency uses, so counting both toward breadth is double counting. But their equity relationship genuinely differs in sign — TIPS +0.131 against nominals’ −0.076 on identical months, a gap of 3.5 standard errors — and nominal bonds’ correlation to equity is decisively era-dependent, spanning 0.802 across twelve 60-month blocks, positive in seven and negative in five. Full working and provenance in the evidence base.

The consequence is a liability argument rather than a breadth argument, and it is stronger at today’s real yields than it has been in twenty years: an investor whose future spending is in real terms is matched by a real bond, and the era-dependence above is exactly the risk that a nominal bond leaves on the table. Hold TIPS because they match the liability, size the defensive sleeve as one engine, and take the diversification from §6’s credit leg instead — which is where it actually is.

---

7. Trend beyond the one already held: the mechanism is unchanged, the price is not

Verdict. Do not add a second trend engine — it would be the same engine twice. But the delivery cost of the one already held has fallen by roughly an order of magnitude, and that is a live implementation finding worth acting on. The evidence on trend itself is owned by trend and live managed futures and is not restated here; §1.3 and §1.4 above add only that its crisis case is an alpha of +0.91%/month (t = 3.09) with no statistically resolvable convexity, and that it is the single engine on this panel that materially beats cash in the lower tail.

What has appeared on the shelf since the last audit (all read 2026-08-22 from issuer pages; verify fee, waiver expiry and AUM before transacting):

Sources: Simplify fund pages, Return Stacked prospectus 2026-04-27, iMGP, KraneShares.

Why the fee matters more than it looks. The trend sleeve’s measured marginal value at 10% weight is +0.580 pp/yr gross, which is exactly the design’s own detection floor. A 99 bp wrapper consumes 10 bp of that at a 10% weight; a 10 bp wrapper consumes 1 bp. That does not make the sleeve resolvable — it is still inside the floor — but it removes the one term in the arithmetic that was known to be working against it. The financed-overlay funds ask a more favourable portfolio question than a pro-rata sale of the core, which is the point capital efficiency makes about funding rules.

Three cautions that travel with the whole category. These wrappers hold futures through a Cayman controlled foreign corporation to keep the income RIC-qualifying, which is what lets them issue a 1099 rather than a K-1 — CTA, KMLM, SDMF and CTAP all state “K-1: No” on their own pages, and the Return Stacked funds are 1940-Act RICs (their 1099 status is an inference from that, not an issuer statement). No issuer in this set discloses a numeric financing cost; the only hard figure available is the Return Stacked funds’ interest-expense ratio of under 0.005% of average net assets for the year ended 2026-01-31, which reflects derivative-embedded rather than borrowed financing. And CTAP’s 10 bp is a waiver expiring 2026-12-04, not a fee.

Where the sleeve belongs. Tax-deferred. Managed-futures funds distribute short-term gains and interest income, and a financed stack distributes both legs.

---

8. Gold and commodities: right about the state, wrong about the state that hurts

Verdict on gold. Optional, at most 5%, and only as a replacement for cash or bonds — never funded by selling equity. A financed wrapper changes the funding question and does not change the expected return. The finance-free version of the argument is in §1.3: gold buys twelve basis points a month of lower-tail protection over T-bills, at 16.24% volatility and a −91.2% peak-to-trough.

Verdict on long-only commodities. Reject as a diversifier; consider only if the investor has a specific inflation liability. Their mean in the worst decile of equity months is −1.84%, positive in only 36% of them, with ρ low +0.326 — they fall with equity in equity crises and pay only in the inflation shocks (1973-74 +143.9%, 2022 +10.5%). That is a real and valuable property, but it is a hedge against a different state, and the investor should not buy it believing it is a crash hedge.

Does a financed wrapper change the gold answer? No, and it is worth being precise about why. GDE (WisdomTree Efficient Gold Plus Equity Strategy Fund) holds US large-cap equity plus US-listed gold futures, at 0.20% with $496.0M of net assets, inception 2022-03-17 (factsheet, data as of 2026-06-30, read 2026-08-22). Its own summary prospectus dated 2026-01-01 says “approximately equal exposure” to the two legs, rebalanced quarterly, with the gold leg run through a Cayman subsidiary capped at 25% of total assets at each fiscal quarter-end to preserve RIC source-of-income qualification — so the “90/90” marketing shorthand is not the prospectus language, and whether §1256 60/40 treatment reaches the shareholder is unverified: gains realised inside a wholly-owned controlled foreign corporation are generally not passed through with that character, and the prospectus is silent. What the wrapper changes is the funding rule, and the marginal-sleeve work has already measured that the funding rule flips gold’s sign: −0.404 pp/yr at 10% funded pro rata against +0.18 to +0.22 financed. What it does not change is that both estimates sit inside the design’s 0.63–1.04 pp/yr detection floor, or that gold’s excess return over half a century has been +1.75%/yr at a Sharpe of 0.18. A cheaper wrapper for an uncertain expected return is a cheaper wrapper, not a better expected return. GDE also converts a grantor-trust holding taxed at the 28% collectibles rate into a 1940-Act fund that is not — which is a genuine tax improvement and should be scored as one, in the tax work rather than here.

One thing the financed form does buy, and it is the honest case for it. Because GDE’s gold leg is notional, holding it does not require selling equity. §1.5 measures that a pro-rata gold sleeve costs −0.40 pp/yr at 10%; a notional one costs its financing rate and its fee instead. For an investor who wants gold’s inflation-state payoff and does not want to reduce equity to get it, that is the correct instrument. The reason this page still caps it at 5% is §1.3: gold’s crisis contribution over cash is 12 bp a month, and financing does not make that number larger.

Commodity vehicles carry a structural tax split worth naming once. The broad commodity products divide into 1940-Act funds that hold futures through a Cayman subsidiary and issue a 1099, and commodity pools that issue a K-1. That difference is larger than the fee difference between them for most investors, and it decides which account can hold the position at all. Confirm the current form on the issuer’s page before transacting; the category has changed structure repeatedly.

The long/short version is not a separate idea. A carry- and momentum-aware long/short commodity strategy is, mechanically, the commodity leg of a diversified trend programme — which the investor already owns inside the managed-futures overlay. Reject on overlap, not on the premium. The measured evidence agrees: the AQR long-only commodity series correlates −0.064 with TSMOM, which is exactly what you would expect if the trend programme is trading the same markets from both sides.

---

9. Screened and set aside, with the reason attached

Each of these arrived with a mechanism. Each is set aside for a stated reason, and the reason is what a future round should attack.

Ideas the repository had not previously considered. These are the ones worth arguing about.

Families screened in earlier rounds and left where they were. The reasons below are scoped conclusions about instruments and vehicles, not statements that the mechanism is absent.

• Volatility selling and put writing. The premium is real — it is the same premium §4 says not to pay. Rejected on overlap, not on the premium: it duplicates equity’s own left tail, and its measured live-only alpha ran −0.09 to −0.88%/yr at correlations of 0.86–0.95 to equity, with an up-beta of 0.45 against a down-beta of 0.86.
• Merger arbitrage. The mechanism is genuine — sellers of deal risk pay for certainty before a deal closes — and the vehicles are real, cheap enough, and liquid. They are also too small to move a portfolio. Ten-year annualised NAV returns to 2026-06-30, all issuer-published and all read 2026-08-22: MNA +2.92%/yr at a 0.77% fee (NYLI factsheet), MERIX +4.29% and MERFX +3.98% at a 1.26%/1.55% net fee (Virtus factsheet), and ARB +4.25% since 2020-05 at 0.76% (AltShares factsheet). Those are total returns over a decade whose average T-bill yield was itself a large part of them, so the excess is a low single digit at best, and the left tail is a break-risk loss that clusters with equity. At any weight a retail investor would hold, the contribution is well inside the 0.58 pp/yr detection floor: unresolved, and too small to be worth resolving. Two housekeeping facts: ARB and EVNT reorganise from AltShares Trust into identically-named series of The Arbitrage Funds on or about 2026-09-25 at identical fees (497), and First Trust’s MARB stopped being a merger-arb fund on 2026-06-24, becoming the Equity Market Neutral ETF (NTRL) at a 0.95% fee (First Trust) — a reminder that a strategy shelf is not a market ontology. A stacked version exists, RSBA (100% US Treasuries + 100% merger arbitrage, 1.01%, $52.3M, inception 2024-12-17), which is the only form in which this premium could plausibly matter, because it does not compete with equity for capital.
• Alternative-risk-premia and multi-strategy funds. The examined shelf earned 0.3–1.0%/yr gross post-2019 at 2–5% volatility against a retail wrapper costing about 1.5%. The cost stack, not the premia, is the finding. The survivor worth naming is QAI (NYLI Hedge Multi-Strategy Tracker, 1.10% gross / 0.88% net, $1.02bn, inception 2009-03-25), whose ten-year NAV return to 2026-06-30 is +3.93%/yr (factsheet, read 2026-08-22) — a decade of hedge-fund replication that did not beat its own cash leg by much. Its index has held digital-asset ETPs at up to ±2.5% since 2026-05-01 (497), which is a mandate change a holder should notice.
• REITs and dividend funds. Dominated on Sharpe at correlations of +0.82 and +0.84; REITs gave 112% of the downside for 80% of the upside. Not distinct engines by label.
• Closed-end fund discounts, securities lending, direct indexing. Retained as candidates with the same next questions as before: a point-in-time discount panel, better N-CEN/N-PORT lending measures by fund and year, and after-fee modelling under the investor’s actual lots. None is a market-return engine; the last two are implementation lines and belong with the structural work.
• Short-term reversal, accruals, net issuance, and other published anomalies. Their post-publication premia sit inside the public library’s own detection floor, and no adequate registered implementation exists on the audited shelf. An implementation finding on the second clause, an underpowered null on the first.

---

10. Consequence for the portfolio

The ranked shortlist

Weights are scenario sizings for the reference investor, not an optimiser output, and they are stated as ranges because the investor inputs that would narrow them — contribution and withdrawal path, embedded gains, account capacity, tolerable drawdown — are still missing. Nothing here is promoted — decision 0004 stands, nothing on this page is production-eligible, and no measurement here was frozen before its numbers were seen. What has changed is which candidates deserve a frozen specification next, and one of them changed because the instrument that rejected it was wrong.

Nothing above is a new return engine except items 1 and 4. Items 2, 3, 5 and 6 are a cheaper wrapper, an absence of exposure, a declared speculation and an optional inflation-state hedge. Counting them as engines would be the error the charter names.

What was rejected, and on what grounds

Grounds matter more than verdicts, because a rejection on cost reopens when the cost changes and a rejection on overlap never reopens at all.

The three findings a reader should leave with

1. A rejection is only as good as the series that produced it. Credit was rejected here on a +0.835 correlation to Treasuries measured on an index carrying twenty years of duration. Hedge the duration out and the correlation is +0.016 over 1,068 months. Same asset class, different instrument, opposite conclusion. Every other rejection on this page should be read with that possibility in mind.
2. Shocks come in two kinds and no asset covers both. Treasuries paid in every growth and deflation shock and lost up to 40% in the inflation ones; commodities and gold did the reverse; trend was positive in both but has no data before 1985. Breadth means holding across shock types, not across ticker counts.
3. In the lower tail, cash is the benchmark almost nothing beats. A 10% swap from equity into T-bills adds +0.92% in the average worst-decile month. Long Treasuries add +0.94%, gold +0.92%, BAB +0.96%, commodities +0.74% and bitcoin +0.05%. Only trend, at +1.08%, materially beats it — and its case is a positive mean, not a convex shape.

What would change this page

• A duration-hedged credit series reaching 2026. The one held ends 2014-12 and has never seen March 2020 or 2022. This is the highest-value acquisition identified here.
• A cat bond spread-to-expected-loss multiple at or above 3.5×. Publicly observable weekly. It is a monitoring boundary, not a forecast.
• A CLO ETF with a liquidity-crisis record, which none has, because the oldest launched in 2020-10.
• Bitcoin’s correlation to equity falling below +0.2 on a window containing a recession, or a realised equity bear market in which it does not fall harder than equity. 2020 and 2022 both went the other way.
• The investor’s own inputs — contribution and withdrawal path, embedded gains, account capacity, tolerable drawdown and tracking error, and whether they carry a mortgage above the after-tax Treasury yield. Every weight above is a range because those are missing, and several would narrow more from one of those answers than from another experiment.
• A waiver lapse. CTAP’s 10 bp expires 2026-12-04; the iShares rate-hedged waivers expire 2027-02-28; the cat bond ETF’s cap expires 2027-04-30. Three of the six recommendations above rest on a fee that is contractually temporary.