Currency and the international sleeve: the position nobody chose

Question. The portfolio holds 35% international — VEA, DFIV, IDMO, IEMG, AVES — every one unhedged. That is a long position in a basket of foreign currencies, funded in dollars, sized at roughly a third of the equity book. Is it compensated, what does it cost in risk, and should any of it be hedged?

Decision it informs. Whether to hedge the currency exposure of the international sleeve — all, half, developed only, or none — and in which account. It does not change the size of the international allocation; that is the equity share and the recommendation.

Out of scope. Currency as a return sleeve in its own right (carry, value, momentum as factors). Hedging foreign bonds, where the literature is unanimous and this portfolio holds none. Tactical or dynamic hedge ratios.

as of 2026-08-22. Measured figures regenerate from studies/currency_hedging.py and its cache companion _currency_hedging_tables.py, run with uv run python -m portfolio_edge.studies._currency_hedging_tables; the arithmetic is pinned in research/tests/unit/test_studies_currency_hedging.py. Everything measured here is exploratory: no specification was frozen before the numbers were seen and no experiment is registered, so nothing below may support a promoted claim (decision 0010).

---

Conclusion

1. The mean is unresolvable and the variance reduction is large and certain. Make the case on the second, or do not make it. Across six developed-market panels — 150 years of annual data and three overlapping monthly windows — the point estimate lands between −0.30 and +0.93 pp/yr against a minimum detectable effect of 1.8 to 4.2 pp/yr. Not one window can sign it. Over the same windows a full hedge removes 17% to 21% of the sleeve’s volatility, and that estimate is precise. Any argument for or against hedging that rests on a forecast of the currency’s return is unsupported by this evidence, in either direction.

2. For this investor the currency leg adds risk rather than diversifying it, and it is worst exactly when it matters. The correlation between the currency basket and the sleeve’s own local-currency equity return is +0.26, not negative. In the worst decile of months for the international sleeve’s own equity, the currency leg lost a further −2.11% per month and was positive in only 14% of them. The variance-minimising hedge ratio is 1.47 — above a full hedge. This is not a hedge that happens to be unnecessary; it is a position that compounds the loss it is supposed to soften.

3. The dollar smile is real, recent, and not a law. In the four stress windows before 2000 that this repository’s currency data reach, the dollar weakened every time — 1973–74, the late-1970s inflation, the 1987 crash and 1998 LTCM. In the six since, it strengthened in five and weakened once, in February–April 2025, when US equity fell 8.5% and the currency basket gained 7.5%. So the pattern the hedging case rests on is a post-1997 one with a fresh exception, not a regularity of markets. The month-by-month tail statistic is the stronger design and it is clearly adverse; the episode list is the honest reminder of how few independent crises there have been.

4. Hedging currently pays about +1.5 pp/yr in carry, and that is the least durable part of the case. US three-month interbank is 3.77% against a VEA-weighted foreign 2.25%. But the Fed has stopped cutting while the ECB, BoJ and RBA have started hiking, the differential has compressed by roughly 89 bp in twelve months, and 82% of what remains comes from JPY, EUR and CHF alone. Underwrite the hedge on variance, not on this.

5. Emerging markets are the opposite trade and the answer is no. EM currency is more strongly pro-cyclical (correlation with its own local equity +0.54, variance-minimising ratio 2.67, positive in only 5% of the worst EM equity months), but hedging it costs the interest differential, which for ZAR and MXN is −3.0 to −3.3 pp/yr against the dollar today. The right response to EM currency risk is to size the sleeve, not to hedge it.

6. Recommendation: hedge about half the developed sleeve, in tax-advantaged accounts only, and nothing in emerging markets. Confidence moderate on “hedge some”, low on the exact ratio. Two implementation facts do most of the work: a hedged fund’s forward P&L cannot leave through in-kind redemption, so it belongs nowhere near the taxable account; and no product hedges the universe this investor actually owns — every large hedged fund tracks MSCI EAFE, which excludes Canada (10.9% of VEA) and Korea (8.0%), the two currencies whose behaviour most argues for hedging. §5 and §6 have the details.

---

1. The mechanism, and what it does and does not imply

Write r_L for a foreign asset’s return in its own currency and s for the appreciation of that currency against the dollar. A US investor holding it unhedged earns (1+r_L)(1+s) − 1. Selling the beginning notional forward at the covered-interest-parity rate F = S(1+i_USD)/(1+i_foreign) leaves

hedged = unhedged − [(1+i_f)(1+s) − (1+i_d)] / (1+i_f)

and the bracket is the excess return on foreign cash funded in dollars — the spot move plus the interest earned abroad, less the interest given up at home. So:

Unhedged minus hedged is the currency excess return, and nothing else.

“Should I hedge?” is therefore exactly “do I want to be long the developed-currency carry trade?”, and that decomposes into three questions with wildly different resolution: a mean (is it compensated?), a variance (what does it cost in risk?), and a crisis dependence (what does it do when it matters?). The derivation is pinned against an independently written forward payoff in test_the_forward_hedge_matches_an_independently_derived_payoff.

The theory is genuinely two-sided. A currency has no cash-flow claim, so the prior that it earns nothing is reasonable and old — Perold and Schulman’s “free lunch” argument (Financial Analysts Journal 44(3), 1988; the text is behind a paywall and is reported here only through Schmittmann, IMF WP/10/151 and Campbell et al.). Its formal condition is two things: a nil risk premium and zero correlation with the asset. Against it: hedging costs the interest differential, foreign currency has sometimes diversified dollar-asset drawdowns, and the hedge has operational and tax costs that the theory ignores entirely.

Which argument dominates is an empirical question about the second condition, not the first, and §2 answers it for this investor.

A caveat that runs through every hedged number below. CIP is assumed, not measured. Since 2008 a cross-currency basis has separated the traded forward from the CIP forward — and its sign favours a dollar-based investor, which is the thing most often stated backwards. The basis sits on the non-dollar leg and is negative because dollars are scarce, so whoever supplies dollars to the swap market is paid for it, and a US investor hedging foreign assets is supplying dollars. The BIS puts the turnover-weighted three-month basis at about −9 bp in October 2025, down from about −59 bp in October 2022, widening each autumn as three-month contracts begin to span the year-end (BIS Quarterly Review, December 2025). Every hedged return here is understated by roughly that much.

---

2. What the repository’s own data say

Five panels, because no single free source answers all three sub-questions. The long one is Jordà–Schularick–Taylor: local-currency equity total returns, bill rates and the xrusd exchange rate for 15 non-US countries, 1871–2020, annual. The monthly ones pair Ken French’s developed-ex-US market return — which he publishes in dollars only, so the local leg here is a residual — with a currency basket built from H.10 month-end noon rates and the OECD three-month interbank differential. See the evidence base for source fitness, including why a month-end sample and not a monthly average.

2.1 The decomposition

Read the resolved column first. The mean is unresolved in every window, including the one with 150 annual observations. Effective sample size — the number of independent draws the serial correlation leaves — runs 0.84 to 0.91 of the raw count, so the shortfall is not a statistical artefact that a longer window would fix soon. To resolve a 1 pp/yr currency premium at this volatility would take several centuries of floating rates, and there have been fifty years of them.

The volatility column is a different object. A variance ratio over the same months is estimated an order of magnitude more precisely than a mean, and the full hedge removes about a fifth of the sleeve’s volatility in every modern window.

Two internal contrasts carry information. First, the JST windows: the deep history’s currency losses are dominated by wars, occupations and currency reforms, not by a floating market — the 30 largest annual moves in that panel are almost entirely 1914–1950, and Germany 1914–1950 is excluded outright as redenomination arithmetic. Pre-1971 “currency volatility” is a devaluation-jump process under a peg, which is why the 1974–2020 row is the only JST row that describes the world this investor lives in. Second, the VEA and EAFE baskets differ only in that VEA’s includes Canada and Korea; adding them raises the volatility cut from 17.2% to 20.9% and the variance-minimising ratio from 1.26 to 1.47. Hold that thought for §5.

2.2 Volatility, and what the currency is correlated with

The third column decides the risk case. A currency leg negatively correlated with the local equity leg would be a genuine diversifier and would belong in the portfolio on risk grounds alone. It is positive for every sleeve this investor owns.

This reproduces independently. MSCI publishes both legs of the same index. As of 2026-07-31, MSCI EAFE 100% Hedged to USD had a ten-year annualised standard deviation of 11.82% against unhedged EAFE’s 14.99% — a 21% reduction, essentially identical to the 20.9% measured above on a differently constructed basket — and the hedged index sat within 0.03 pp of the pure local-currency index (11.85%), which is what a working hedge looks like (hedged factsheet, read 2026-08-22). Campbell, Serfaty-de Medeiros and Viceira (Journal of Finance 65(1), 2010, Table VII) put a US investor’s global equity volatility at 15.05% unhedged against 13.86% hedged over 1975–2005; AQR’s Risk Without Reward (2015) puts 16.7% against 14.5% over 1975–2015.

2.3 Crisis-conditional behaviour, which is the decision-relevant measurement

Average correlation is incomplete evidence about crisis dependence (charter), so both the tail statistic and the named episodes are reported.

In 86% of the worst months for the international sleeve’s own equity, the currency moved against the holder too. That is the answer to the question the charter asks: the currency leg is not neutral in the tail, it is additive to the loss. Note that the in-tail correlation is not comparable with the full-sample figure — conditioning on the base’s magnitude truncates its variance — and should be read against the same design’s other rows only.

The named episodes are the honest complication.

* partial coverage. – means no currency series in this repository reaches the window; none of them sees 1929, 1937 or 1973 in full, and that limit is itself the most important thing the top rows say. The 2025 and 2026 windows are the peak-to-trough months of French’s US market return since 2024-06, read off the panel rather than chosen.

Split the rows at 2000 and the pattern is unmistakable — and so is its age. In all four pre-2000 windows the panel reaches (1973–74, the late-1970s, 1987, 1998) the foreign-currency leg was positive: the dollar weakened. In five of the six windows since (2000–02, 2008–09, 2020, 2022, 2026) it was negative: the dollar strengthened. The GFC and 2022 rows are large — a 14 pp currency loss stacked on a 50 pp and a 25 pp equity loss — and they are the two episodes that matter most to a long-horizon holder. But the dollar’s crisis behaviour as this data records it is a quarter-century-old regularity with one fresh counter-example, not a law of markets, and a decision that leans hard on it is leaning on roughly five independent observations.

2025 is a published break, and its cause is disputed. The BIS regresses daily S&P 500 returns on the DXY and reports a fitted slope of −1.449 (R² 0.26) for 2020–24 flipping to +0.792 (R² 0.09) in 2025 (BIS Quarterly Review, December 2025, Graph 4.A). Its own reading of April 2025 is not a loss of confidence in dollar assets but ex-post hedging flow by non-US investors whose hedge ratios had fallen to historic lows — Japanese life insurers from about 60% in 2021 to 40% in 2024 (BIS Bulletin 105, June 2025). If that reading is right, 2025 is a positioning unwind inside an unchanged relationship. One year is not a sample either way, and the March 2026 review has the dollar rallying again on a genuine risk-off shock.

2.4 The hedge-ratio frontier

VEA basket, spot and carry, 2002-04 to 2026-01. The curve is close to linear: the first half of the hedge buys about 60% of the volatility reduction and about half of the drawdown and worst-month reduction. Campbell et al. found the same shape more sharply — a half hedge captured 1.14 of the 1.19 pp their full hedge bought — and it is why the choice between 50% and 100% matters much less than the choice between 0% and 50%. MSCI’s published maximum drawdowns say the same thing in realised terms: −54.60% hedged against −60.41% unhedged, both troughing on 2009-03-09.

2.5 The carry today

Weighted, +1.52 pp/yr before the basis pickup and before any fee. The same calculation on policy rates read on 2026-08-22 (Fed 3.50–3.75%, ECB deposit 2.25%, BoJ 1.00%, BoE 3.75%, SNB 0.00%, RBA 4.35%, BoC 2.25%, Riksbank 1.75%) gives +1.40 pp/yr, and shows the same twelve- month compression of roughly −89 bp: the Fed cut three times in late 2025 and stopped, while the ECB, BoJ and RBA have all raised since. Hedging GBP and AUD now costs money.

This is the least durable leg of the case. It is a positive number today, it is shrinking, and 82% of it comes from three currencies.

The decade table is the reason not to lean on any recent window.

Cumulatively the developed basket lost −25.83% over 2010–2024 (the Fed’s own AFE index says −27.09%), then gained +7.63% in 2025 and lost −2.77% in 2026 H1. A decade of drag followed by one large year is what an unforecastable near-random-walk looks like; it is not evidence of a trend in either direction.

---

3. Emerging markets are a different question with a different answer

Three findings. The first argues for hedging EM harder than developed; the second and third say the trade is not available at a price worth paying.

The risk case for hedging EM is stronger than for developed. Correlation with its own local equity is +0.54 against the developed sleeve’s +0.26; the variance-minimising ratio is 2.67; and in the worst decile of EM equity months the currency was positive in 5% of them. Conover, Garcia-Feijoo, Silverstein and Szakmary (Review of Financial Economics 43, 2025, open access) find the local-equity/currency correlation significantly positive in 12 of 13 emerging markets and the ex-post optimal hedge ratio equal to 1 in 11 of 13, against significantly negative correlations in six developed markets. The mechanism is not mysterious: EM currencies are commodity-linked, and dollar-denominated corporate debt gets harder to repay exactly as the local currency falls.

But the spot depreciation is roughly the interest differential, so hedging removes both. The spot-only row shows a −1.12 pp/yr currency drag; the carry-inclusive row on the four currencies with a comparable interbank series shows +1.21 pp/yr — the sign reverses once the interest earned abroad is counted. That is the carry trade, and it means the naïve reading of the spot table (“EM currency loses 1.1 pp/yr, hedge it”) is wrong.

And the cost of hedging is prohibitive and worst when it is needed. Selling a high-yielding currency forward locks in the differential as a loss: ZAR and MXN cost 3.0–3.3 pp/yr today. The two rows with a blank rate column are the expensive ones — Brazil and India have no comparable interbank series on this feed, and both currencies depreciated 3–4 pp/yr against the dollar over three decades, which is what a large positive differential looks like from the other side. Viceira and Shen (NBER WP 35498, July 2026) make the trade-off explicit: risk minimisation wants EM over-hedged at ratios of 2.7–2.9, but their factor-implied expected currency excess returns are +6.1% for BRL and +4.7% for MXN, which “offset, and in some cases reverse, portfolio-risk minimizing demands.” On top of that, much EM hedging runs through non-deliverable forwards, whose pricing separates from the onshore market precisely in stress (McCauley, Shu and Ma, BIS Quarterly Review, March 2014).

Verdict for EM: do not hedge. The measurement says EM currency is a crisis amplifier; the cost structure says the hedge is not worth buying; and the tools are worse. The correct control for EM currency risk is the size of the EM sleeve, which is a different decision on a different page. It also interacts with the foreign-tax-credit finding in structural and tax edges — an EM sleeve in a taxable account is already carrying an account-placement argument, and adding a hedge does not help it.

---

4. What the literature says, and where it genuinely disagrees

Converged, and worth stating plainly.

• Developed-currency excess returns are close to zero and statistically indistinguishable from zero. Nobody disputes this. The whole argument is about correlation.
• The dollar was negatively correlated with global equity through 2023. Campbell et al. (2010) over 1975–2005 and Viceira and Shen (NBER WP 35498, 2026) over 1975–2023 agree, and the latter states the implication for this investor without hedging its words: “the U.S. dollar is still negatively correlated with global equity markets… This means that US investors want to hedge and even overhedge their currency exposures.” Their implied optimal ratios using observed forwards are CAD 2.03, AUD 1.44, EUR 1.03, GBP 1.00, JPY 0.50.
• The correlation is unstable and regime-dependent — Schmittmann (2010) measures it swinging ±40 points decade to decade. Anyone quoting one number is quoting one sample.
• Foreign bonds should be hedged, unanimously. Not relevant here; noted so it is not mistaken for support of the equity case.

Genuinely open.

• Vanguard and AQR give opposite advice to the same investor. AQR (2015) reports hedged volatility of 14.5% against unhedged 16.7% over 1975–2015, Sharpe 0.37 against 0.33, a worst one-year drawdown of −60.4% against −78.7%, and a volatility-minimising currency allocation of about −10% — hedge fully, then a little more. Vanguard (May 2026) recommends leaving equities largely unhedged, at portfolio hedge ratios of 10–20% for an 80%-equity investor. They are measuring different things: AQR compares risk-adjusted outcomes, Vanguard compares a marginal volatility saving against a fixed cost of 0.25–0.50% a year. And Vanguard carves out this exact investor: its own Figure 4 reports that from a US perspective the ten-year rolling currency/equity correlation was positive — risk-adding — in 71% of 1972–2025, the highest of the seven domiciles it shows, and its text concedes that “the US dollar’s ‘safe haven’ correlation channel… may justify higher overall hedging for USD-based investors.”
• Whether hedging helps at long horizons. Froot (NBER WP 4355, 1993) is the standard citation for “no” — the minimum-variance ratio falls to about 35% at five to ten years. Two things are usually dropped: it uses one currency pair, GBP/USD, from a British investor’s perspective; and its own text concedes that “the data provide little power against the hypothesis that β = 1 at very long investment horizons.” Schmittmann (2010) reaches the opposite conclusion on four base currencies and says so explicitly. Neither settles it, and the burden of proof sits with the side whose data cannot reject a full hedge.
• Whether 2025 broke the dollar’s safe-haven property. §2.3. Published, quantified, and interpreted by its own publisher as a flow event rather than a regime change.

On the 50% rule, a correction worth carrying. The usual attribution of “minimum regret = 50%” to Gardner and Wuilloud (Journal of Portfolio Management 21(3), 1995) could not be verified — the citation is real, the paper is paywalled, and nothing found actually says they recommend 50%. The properly sourced version is Michenaud and Solnik (Journal of International Money and Finance 27(5), 2008), who derive it formally: the regret-minimising hedge ratio is always between 50% and 100%, and reaches 50% only in the limit of infinite regret aversion. The 50% choice is “always wrong, but the maximum regret is cut in half” — and, decisively for this page, it requires no forecast of the currency mean, which is the one quantity none of the evidence above can supply. Black’s universal hedging ratio (77% on his baseline inputs) is the other principled interior answer, and its own author’s worked examples span 30% to 73% across two four-year windows, which is the standard critique.

---

5. Implementation, honestly

5.1 The products, and the one that does not exist

Read 2026-08-22 from issuer pages and SEC filings; HEFA’s fee table and net assets were confirmed independently against ishares.com, and VEA’s 0.03% against Vanguard’s own profile endpoint (stated as of 2026-04-28).

Prefer DBEF to HEFA on structure, not on price. They cost the same and hedge the same index, but HEFA’s fee is a waiver with an expiry and its fund-of-funds form carries a tax wrinkle its own prospectus discloses: “the Fund’s realized losses on sales of shares of the Underlying Fund may be indefinitely or permanently deferred as ‘wash sales’”, and the underlying fund’s own loss carryforwards cannot offset the hedged fund’s gains. DBEF holds the securities directly, charges a flat unitary fee with no waiver to expire, and has no acquired- fund layer at all.

Fees are not the obstacle. Against the VEA counterfactual a hedge costs about 32 bp, which is real but small against a 3 pp/yr volatility reduction. HEFA’s 0.35% is a contractual waiver running to 2030-11-29 and structured as “EFA’s fee plus 0.03%”; DBEF has no waiver at all, which makes it the more robust choice on this axis. Realised tracking against the hedged index is tight — HEFA within 3–28 bp over 1, 3, 5 and 10 years — so the hedge itself is not where value leaks.

The obstacle is that no product hedges what this investor owns. Every broad hedged fund tracks MSCI EAFE, which excludes Canada (10.9% of VEA) and Korea (8.0%) — 19% of the sleeve, as of Vanguard’s 2026-07-31 allocation. HFXI’s index is FTSE Developed ex North America, which also excludes Canada. And Canada and Korea are precisely the currencies whose behaviour most argues for hedging: §2.1 measured the variance-minimising ratio rising from 1.26 to 1.47 and the local-equity correlation from +0.14 to +0.26 when they are added, and Conover et al. measure CAD’s equity correlation at +0.18 where France’s is −0.18. The available hedge removes the currency risk that needs removing least.

This is not fatal — 81% of the exposure is hedgeable — but it means switching VEA for DBEF is not a currency-only change. It is also a change of equity index, and the charter is explicit that two decisions bundled into one trade should be separated. A partial position beside VEA keeps them separate; a wholesale swap does not.

Also relevant to the 50% question: Xtrackers has an MSCI EAFE 50% Hedged Equity ETF in registration on Form 485APOS, filed mid-2026. Its ticker and its entire fee table are blank in the filing and it has not launched. Worth checking at the next review; do not plan around it.

5.2 The mechanics leave a residual, and the issuers say so

The MSCI hedged indices sell each foreign currency forward at the one-month forward rate and reset the hedge monthly, without intra-month adjustment for equity price moves. So the fund is over-hedged when equities fall inside a month and under-hedged when they rise. This is the same r_L × s cross term the arithmetic in §1 keeps rather than dropping, and it is why a “100% hedged” fund is not fully hedged.

5.3 Tax, which is the part that actually decides the account

The mechanism is §852(b)(6). An ETF flushes appreciated stock out through in-kind redemption with no gain recognised. A cash-settled currency forward cannot be delivered in kind. The equity leg of a hedged ETF is therefore exactly as tax-efficient as its unhedged twin, while the currency leg is forced through the income statement and out to shareholders in cash. HEFA’s FY2025 statement of operations shows the two legs side by side: a realised loss on forward FX of $(581.9)m against in-kind redemptions of +$705.4m.

The consequence in a dollar-strength year is not subtle.

EFA and IEFA have distributed zero capital gains in every year 2016–2026. HEFA has done so in 2014, 2015, 2016, 2018, 2019 and 2022. The realised after-tax drag, from iShares’ own standardised after-tax-pre-liquidation figures, is 204 bp over 5 years and 144 bp over 10 for HEFA against 78 and 69 for EFA — roughly 2.5 to 3 times.

On character: a foreign currency forward is a §1256 contract, so it is marked to market annually whatever else happens. Its character is ordinary under §988(a)(1)(A) unless the fund elects out under §988(a)(1)(B). HEFA’s 2022 distribution split of 60.08 / 39.92 is the §1256(a)(3) signature and implies the election was made, but no fund document states it; DBEF’s treatment is unverified. Even under 60/40 the short-term 40% is distributed as a non-qualified ordinary dividend.

On the foreign tax credit, the worry is half right. §853(a)(1) requires more than 50% of assets in stock of foreign corporations, and a fund-of-funds holding a US-listed ETF fails that on its face — but §852(g)(1)(B), added by the RIC Modernization Act of 2010, waives the test for a qualified fund of funds. HEFA’s own 1099s confirm the credit reaches shareholders at the same rate as EFA’s (5.28% of ordinary dividends in 2025 against EFA’s 5.14%, on QDI shares of 76.00% and 76.15%). The election is annual rather than guaranteed — neither HEFA nor EFA passed any foreign tax through in 2021 or 2022 — but that failure hit both funds identically and is not a hedging problem. The real problem is second-order and specific to hedging: §988(a)(3)(A) sources ordinary §988 income by the taxpayer’s residence, so hedge gains are US-source income that dilutes the foreign-source fraction in the §904 limitation. HEFA’s foreign-source income was 22.61% in 2022 against EFA’s 100%. The credit survives; the ability to use it shrinks in exactly the years the hedge pays.

No Cayman subsidiary is involved and none is needed — §851(b)(2)(A) already treats currency forward gains as qualifying income when they are derived with respect to the business of investing in the underlying securities. This is not analogous to the commodity-fund structure structural and tax edges covers. The analogy that does hold is the ordinary-income one: like a §1256 commodity structure, a hedged equity fund converts return into currently recognised income the holder cannot defer.

Conclusion: a currency-hedged fund is disqualified from the taxable account. Not marginal — disqualified. The one exception in the data is a fund carrying a large disclosed loss carryforward (HEDJ holds about $2.1bn against $1.7bn of net assets, a shield worth roughly 124% of NAV), but that is depletable, fund-specific, and HEDJ is a Europe-only 0.58% product that does not fit this sleeve.

---

6. The decision

Hedge about half the developed portion of the international sleeve, hold the hedged position only in the Roth or traditional accounts, and hedge nothing in emerging markets.

Concretely: leave VEA, DFIV, IDMO, IEMG and AVES exactly as they are in the taxable account, and in the sheltered accounts hold DBEF beside the developed exposure, sized so that roughly half the developed sleeve’s currency notional is hedged. The developed sleeve is 25% of notional equity exposure (DFIV 10, VEA 10, IDMO 5) against 10% emerging (IEMG 5, AVES 5), so a half hedge is a position of roughly 12% of the portfolio — which fits comfortably in two accounts holding about two-thirds of it between them, without displacing the higher placement priorities in structural and tax edges.

Confidence: moderate that some hedge is right; low on the ratio.

Why moderate rather than high: the variance reduction is large, precisely estimated, reproduced on six internal panels and on MSCI’s own published index pair, and corroborated by three independent literature strands including one 2026 paper by an author of the canonical one. The crisis evidence points the same way and is the measurement the charter asks for. The carry currently pays.

Against that, the portfolio-level effect is modest, and this page has not measured it. A half hedge takes the sleeve from 16.15% to 14.12% annualised volatility, about 2.0 pp. At a 35% weight that is an upper bound of roughly 0.7 pp on portfolio volatility, and the true figure is lower because the sleeve is imperfectly correlated with the rest of the book — Vanguard’s independent estimate for a US investor is 0.3 to 0.7 pp. Whether that is worth a certain 32 bp and a permanent account constraint is a judgement, not a measurement, and it is the weakest link in the chain. The mean is unresolvable. And the product hedges the wrong universe.

Why 50% and not 100%: because the frontier is nearly linear and the first half buys about 60% of the benefit; because the regret-minimising ratio requires no forecast of the quantity this evidence cannot resolve; and because 2025 demonstrated in one year that the regret is real — unhedged EAFE beat hedged EAFE by 8.1 percentage points (31.22% against 23.10%). A 50% hedge is not a compromise between two estimates. It is the correct answer to a question whose central input is genuinely unknown.

Why not 0%: because the strongest argument for leaving it alone — that foreign currency diversifies dollar-asset risk — is the one thing measured here that is clearly false for this portfolio. The correlation with the sleeve’s own local equity is +0.26, the variance-minimising ratio is above one, and in 86% of the sleeve’s worst months the currency made it worse.

What would change this.

• The carry turning negative. It is +1.4 to +1.5 pp/yr today and has compressed 89 bp in twelve months, with the BoJ, ECB and RBA all hiking. If US-minus-foreign goes to zero, the variance argument has to carry a certain fee on its own, and the case for 50% weakens toward 25%. Recheck at the next annual review.
• The currency/local-equity correlation turning durably negative — AQR puts the threshold at about −0.5, Vanguard’s grid at about −0.3. It is +0.26 now. A sustained move below −0.3 would flip the recommendation to 0%.
• A second 2025-style episode. One dollar decline inside an equity drawdown is a data point the BIS itself reads as a positioning unwind. A second one, especially with a different proximate cause, would be evidence of a regime change and should reopen this page.
• A hedged product on a Canada-and-Korea-inclusive index at a comparable fee, which would remove the largest implementation objection. Watch the Xtrackers 50%-hedged EAFE filing and any FTSE Developed ex-US hedged launch.
• A frozen evaluation. Everything here is exploratory. The whole-portfolio effect at the actual 35% weight has never been scored against the charter’s outcome set — after-tax relative terminal wealth, drawdown, benchmark-relative risk, and holdability — and that is the natural next test.

---

What could not be established

• The sign of the currency premium. Unresolved on all eight panels, developed and emerging, with detection floors of 1.8 to 4.2 pp/yr against point estimates inside ±1.3. This is not a small effect; it is an unmeasured one, and no free data this repository can reach will change that soon.
• Whether 2025 was a break. Published, quantified, and disputed by its own publisher.
• A local-currency developed-ex-US equity series. None is free. The local leg here is a residual after dividing out a constructed basket and inherits every error in the weights; it is not a measurement of a local-currency index and is not quoted as one.
• Point-in-time currency weights. The baskets apply constant 2026 fund weights to windows starting in 1995 and 1999. Vanguard’s own endpoint shows Korea going 4.6% to 8.0% in twelve months, so the constancy assumption is known to be wrong; the Fed’s trade-weighted indices are reported beside every basket result as a check that the answer survives a different scheme.
• EM carry for BRL, INR and TWD. No comparable interbank series exists on this feed, so the carry-inclusive EM panel covers four currencies of seven and is not the fund’s basket.
• The realised cross-currency basis inside a fund’s own forward roll, and whether DBEF has made the §988 election out.
• The after-tax comparison for this investor. It depends on his marginal rates, contribution path and account capacity, and the charter requires that dependency be shown rather than silently resolved. The direction is unambiguous and is enough for the account recommendation; the magnitude is not established.