The claim this study tests is narrow, and worth stating precisely before any numbers appear.

In scope. Two withholding layers: US withholding suffered inside a fund, and US withholding on the distribution paid to the investor. Fund fees. A one-off switching cost. In §6 only, a capital-gains charge when a position is sold. Nothing else.

Out of scope, explicitly. The investor’s own domestic taxation of either fund, other than the capital-gains charge in §6. That includes tax on distributions: no country’s rate is applied to them anywhere, and §6 reports its results under two stated alternatives rather than a real rate: no tax on distributions beyond withholding, or an additional tax on distributions after withholding at the capital-gains rate, with no foreign-tax credit modelled. Foreign tax credits. Regimes that tax accumulating funds on an imputed basis, such as the German Vorabpauschale or the Dutch box 3 system. Estate tax. Currency effects. Any of these can reverse the sign of the result in either direction, and none of them is modelled here. A reader in Germany or the Netherlands should read the 15% treaty figures in §4 as statements about withholding and nothing more.

Gate one: the same published index

Where two funds track different indices, the gap between them contains exposure as well as cost, and a fee-and-withholding comparison silently attributes the one to the other. VT against a FTSE All-World UCITS is not a like-for-like switch; it is a different index. BorderFolio’s catalogue carries index-equivalence as a field on every pair for this reason.

Gate two: how much of the dividend stream is US-source

This gate is narrower than the first, and it is not a pass or fail. It is a question of degree, and the answer changes what the study is entitled to claim about each pair.

The fund-level rate used throughout — 15% (modelled) — is the US–Ireland treaty rate on US-source dividends. It describes what an Irish fund suffers on a dividend paid by a US company. It says nothing about a dividend paid by a Japanese or British one, which is subject to an entirely different rate, on both the Irish and the US side. The supplied model contains no per-source-country withholding data, so the non-US portion of any fund’s dividend stream cannot be computed from it.

That puts the three candidate pairs in three different positions.

Exact — IVV → CSPX. S&P Dow Jones requires every S&P 500 constituent to be a US-domiciled company subject to domestic-issuer reporting. Every dividend in the index is therefore US-source, and the 15% rate is a treaty rate rather than an estimate. Everything this study says about this pair holds exactly, within the stated scope.

Approximate, with a stated bound — QQQ → CNDX. The Nasdaq-100 admits foreign Nasdaq listings by design, so some portion of its dividend stream is not US-source. That portion is not measured here, and applying a flat 15% to the whole stream is therefore an approximation. §7 states how large the non-US portion would have to be, and how different its withholding treatment would have to be, before the conclusion for this pair changed. Under the assumptions specified there, the result for this pair remains negative unless the non-US share of gross dividends exceeds roughly 88.7%. That is a conditional statement, not a measurement: the actual share is not established here, and the result is an approximation which is labelled as one wherever it appears.

Not established — URTH → IWDA. These funds hold global developed-market equities: a majority US, but with substantial Japanese, British, Swiss and French holdings. Here the non-US portion is not a residual, it is a large share of the whole, and no bound of the kind used for the Nasdaq-100 pair would be meaningful. URTH → IWDA is excluded from every withholding-dependent figure in this study — the annual comparison in §3, the treaty-rate table in §4, and the payback grids in §6. It appears only in §2 and §5, where the point being made is about fund fees and involves no withholding assumption at all.

The two pairs carried forward:

Before the withholding analysis, one observation that requires no modelled input at all.

The two pairs above both carry a fee gap running against the switch. That is not a general property. On the MSCI World pair the gap runs the other way:

These are published expense ratios. No proxy, no assumption, no model — six numbers read off six fund pages, and the direction of the difference is whatever it happens to be for the pair in front of you.

This matters because the common framing runs the other way: US funds are cheap, Irish funds carry a fee premium you accept in exchange for a withholding saving. On the MSCI World pair that framing is simply wrong, and a reader applying it would reach for a trade-off that does not exist.

The URTH → IWDA fee gap is the only thing this study says about that pair. What it is worth in total, once the withholding layers are counted, is not established here — see gate two above.

For a position of value V paying annual distributions D, with the investor’s US treaty rate r, the fund-level proxy L1 = 15% (modelled), and ΔTER the Irish fee minus the US fee:

annual difference  =  D × (r − L1)  −  V × ΔTER

Substituting D = V·y, where y is the distribution yield, the position value cancels:

the switch is favourable  ⟺  y × (r − L1)  >  ΔTER

Stated as a multiplication rather than as a crossover yield, on purpose. The division form y* = ΔTER / (r − L1) is undefined at r = 15% and silently flips its own inequality below it — which is exactly the region where the interesting results live. Where r is above 15% the crossover yield is well defined and quoted below.

At the statutory 30% rate, on a position of $100,000 (modelled):

Under the stated assumptions, the two pairs come out with opposite signs.

QQQ → CNDX is negative at the statutory rate. A withholding-only view reports $63.58 saved and stops there. The fee difference is nearly twice as large in the other direction. The crossover yield is 0.80% against a fund yielding 0.4238% — not a marginal call, a gap of nearly two to one. A tool that reports the withholding saving without the fee difference reports this switch as a benefit when, on these two layers, it is not one.

IVV → CSPX is positive, with a four-fold margin between its yield and its crossover. That sounds comfortable until §7.

The position value cancels out of the condition, so the sign of each result above is independent of the $100,000 figure (modelled). The dollar amounts are not.

The statutory 30% applies where no treaty rate does. Most readers of this study are not there. The same two pairs at a 15% treaty rate — Germany, the Netherlands, South Africa — and at 10% — Japan, China, Bulgaria, Romania, Mexico:

At 15% the withholding term is zero for IVV → CSPX. Not small — zero, for every yield, because the investor’s rate and the fund-level proxy are the same rate applied to the same income, and every S&P 500 dividend is US-source by index rule. Whatever remains is the fee difference and nothing else.

For QQQ → CNDX the term is approximately zero at 15%, not exactly. It is exactly zero on the US-source portion of the fund’s distributions, for the same reason. On the remainder it is whatever the difference between Irish and US treaty rates happens to be for those source countries, which this model cannot compute. The −$120.00 figure above is the fee difference alone, and is a close approximation to the extent the stream is US-source — see §7 for how far that can be pushed.

Both pairs are negative at 15% on that basis.

Below 15% the term goes negative. The Irish wrapper’s fund-level rate now exceeds what the investor would pay directly, so the switch costs withholding rather than saving it.

This is the finding most likely to be new to a reader, and also the one most often hidden. BorderFolio’s engine floors the withholding saving at zero, which displays a penalty as “no benefit”. The signed figures are in the table above, and BorderFolio documents the floor as a limitation in its supplied technical workings.

The caution from §1 applies with force here. These are statements about withholding under the stated assumptions. They are not conclusions about German, Dutch or Japanese investors, whose domestic treatment of a US fund versus an Irish one is not modelled anywhere in this calculation and can be larger than everything measured above.

This strand is out of scope, and the reason matters more than the omission.

The switching cost used throughout is 0.5% of position value (modelled) — purely proportional. The supplied model contains no fixed commission, no per-order minimum, no FX spread and no bid-ask modelling.

That has a consequence in each direction, and the direction depends on position size:

• At small tickets the allowance is too low. A $500 switch at a broker charging $3 per order (both modelled) is two orders, not one: $6 in commission before any spread, or 1.2% — more than twice the proportional allowance. Payback figures for positions of this size understate the cost.
• At large tickets it may be too high. A flat 0.5% (modelled) on a six-figure position is a large absolute sum, and a percentage allowance does not shrink the way real per-order costs do. Payback figures here may overstate the cost.

The study does not model where the crossover between the two sits, so no general claim is made about the direction of the error. What can be said is that the payback figures in §6 are results under a proportional-cost assumption, and that the assumption is least defensible at the small end.

Closing this strand needs a fixed-plus-proportional cost split that the supplied model does not currently contain.

Everything above assumes the money is already in the Irish fund. For new money there is no sale, so no capital-gains charge and no sell-side friction, and §3 and §4 give the result under the stated assumptions without a payback calculation.

For an existing position the one-off cost is:

capital-gains charge  =  V × embedded gain × CGT rate   (both modelled)
friction              =  V × 0.5%                        (modelled)

At $100,000 (modelled) with a 40% embedded gain and a 15% rate (both modelled), that is $6,000 + $500 = $6,500.

There are two defensible ways to charge it, and they answer different questions.

The simple basis divides the one-off cost by the annual difference. It charges the whole capital-gains bill to the switch. That is the appropriate treatment for a holder who would never have sold. It is the original benchmark: it uses the §3 annual difference, which excludes any tax on distributions, and every simple figure below is calculated this way whatever the distribution-tax assumption in that column.

The acceleration basis compares both paths after tax at a horizon where both are sold, each compounding at its own net return. For a holder who would eventually sell anyway, the tax is moved forward rather than created, and the new position starts from a higher cost basis. That is the appropriate treatment for a holder who would sell.

The acceleration basis is often assumed to be the friendlier of the two. Where distributions are untaxed, both bases rest on the same tax assumptions, so the difference between them is a difference of method. Where an additional tax on distributions is modelled, only the acceleration basis includes it, so the difference between the two figures reflects the tax assumption as well as the method.

IVV → CSPX at a 30% treaty rate, distributions reinvested and untaxed (both modelled) — payback in years, simple benchmark / acceleration:

The same grid with an additional tax on distributions after withholding, at the capital-gains rate, with no foreign-tax credit modelled (modelled), reinvested in full (modelled). At the 30% treaty rate the combined burden on each distribution is 40.5% at a 15% rate and 47.5% at a 25% rate. This is a sensitivity scenario, not a domestic tax treatment that applies in any particular country. Payback in years, simple benchmark / acceleration:

The simple figures are identical in both grids because the simple benchmark divides by the §3 annual difference, which excludes any tax on distributions. At a 0% capital-gains rate the two grids are the same, because the tax on distributions is then zero in both. The acceleration figures in this grid are therefore not a like-for-like comparison with the simple benchmark: they include a tax the benchmark leaves out. For context, at a 40% gain and a 15% rate, adding the modelled distribution-tax saving to the simple annual difference ($119.00 + $742.00 × 15% = $230.30) gives $6,500 ÷ $230.30, approximately 28.2 years, against 54.6 on the benchmark and 32.8 on the acceleration basis.

QQQ → CNDX has a negative annual difference at every gain and rate, so no simple figure exists anywhere. On the acceleration basis with distributions untaxed, no horizon exists within 1,000 years in any cell. With distributions taxed at the capital-gains rate, horizons appear only at a 25% rate (modelled): 111.8, 406.2 and 716.2 years at 0%, 20% and 40% embedded gains, and none within 1,000 years at 60%. None of them falls inside an investing lifetime.

With distributions untaxed, the acceleration basis is longer than simple in every cell where the capital-gains rate is above zero, and the two figures rest on the same tax assumptions. With an additional tax on distributions at the capital-gains rate, the acceleration figure is below the simple benchmark in every cell where both the rate and the embedded gain are above zero. That comparison changes the method and the tax assumption together, so it does not show the acceleration basis to be the more lenient method: in the example above, a simple figure that counts the same tax saving is shorter still.

Three effects run against each other. The higher cost basis on the new position helps the switch. The tax paid today stops compounding, which hurts it. And the accumulating fund turns dividends into capital gain, taxed at the capital-gains rate on sale, while the distributing fund’s dividends are taxed at whatever rate applies to distributions.

Where that rate is zero, the third effect runs heavily against the switch. At a 0% embedded gain no tax is paid at the switch at all, so the second effect is absent, yet at a 25% rate (modelled) the acceleration figure is 36.0 years against 4.2 simple. That gap is the third effect alone. Where an additional tax on distributions is modelled at the capital-gains rate, the third effect shrinks to a deferral advantage for the accumulating fund, and the acceleration figures fall sharply. With distributions untaxed, at a 60% embedded gain and a 25% rate (both modelled), the two bases are roughly fifty years apart.

The two are published side by side with the rule stated plainly: simple is the benchmark for a holder who would never sell, and excludes any tax on distributions; acceleration is the figure for a holder who would. Where distributions are untaxed, the ordering between them is a comparison of methods. Where an additional tax on distributions is modelled, it is not, and this study does not model any country’s rates.

Holders who spend distributions

Both grids above assume every distribution is reinvested (modelled). A holder who spends distributions cannot be compared with an accumulating fund by adding up the cash they received. The holder who switches receives no cash, so that comparison would measure a difference in saving, not a difference between funds.

The comparison used here is income-matched (modelled). The holder who stays spends each distribution, net of withholding and of any tax on distributions. The holder who switches sells enough CSPX units each quarter to raise the same net cash, pays the capital-gains rate on the gain portion of each sale, and pays no friction on those sales (modelled). Both positions are then compared after tax at the horizon, as above.

IVV → CSPX, distributions spent, income-matched, 30% treaty rate — payback in years, simple benchmark / acceleration:

none: no horizon within 1,000 years. “Taxed” means an additional tax on distributions after withholding, at the capital-gains rate in that column, with no foreign-tax credit modelled; “untaxed” means no tax beyond withholding. Simple figures are the benchmark and exclude distribution tax in every column.

Spending lengthens the acceleration figure in every cell where both the capital-gains rate and the embedded gain are above zero, under either tax treatment. The cause is the income-raising sales: each one pays capital-gains tax on its gain portion.

With distributions untaxed, the distributions those sales replace pay no tax at all, and the switch has no horizon within 1,000 years in seven of the eight cells with a capital-gains rate above zero.

With an additional tax on distributions at the capital-gains rate, acceleration is shorter than the simple benchmark at a 20% gain and longer at a 60% gain. At a 40% gain it is shorter at a 15% rate and longer at 25%. As in the reinvested grid, the benchmark excludes that tax, so these orderings reflect the tax assumption as well as the method.

QQQ → CNDX has no horizon within 1,000 years in any cell on this treatment, under either tax assumption.

Cost-basis treatment

These acceleration figures differ from those in BorderFolio’s supplied technical workings dated 12 September 2026, in one modelling choice.

Every US fund in this study distributes; every Irish fund accumulates. An investor reinvesting distributions from a distributing fund in a taxable account acquires new shares, and those shares carry their own cost basis. The supplied implementation holds the stay path’s cost basis fixed for the whole horizon while compounding net distributions into the terminal value, which taxes those distributions a second time as capital gain at the eventual sale. The accumulating side is modelled with a fixed basis, which is correct there — so the asymmetry is one-sided and runs in favour of switching.

This study adds reinvested distributions to the stay path’s cost basis.

Both treatments in this comparison assume distributions are reinvested in full and untaxed (both modelled), matching the first grid above.

To rule out the calculation method as the cause of the difference, the supplied fixed-basis treatment was reimplemented under this study’s quarterly discretisation. At a 30% treaty rate, payback in years:

* VOO → VUAA is not used in this study. It appears here only because it is the pair the supplied grid runs on, so the two treatments can be compared on identical inputs. The reimplementation reproduces the supplied figures to within two to four percent, which leaves the cost-basis handling as the remaining difference.

Under the fixed basis, the acceleration figure is shorter than the simple figure in some cells — 24.0 against 29.4 above. Under the step-up basis, with distributions untaxed, it is longer in all of them. The conclusion in the supplied workings, that the acceleration correction does not run the way one would expect, holds on that basis with distributions untaxed. Once an additional tax on distributions is modelled, the acceleration figures fall below the simple benchmark, but the benchmark excludes that tax; see the note below the second grid above.

Every figure in this study reproduces from a calculation script.

A study that reports sensitivity for its favourite scenario and not the others has not reported sensitivity.

Fee precision. The crossover condition depends on ΔTER, a difference between two small numbers. At a 30% treaty rate a two-basis-point error in either fee moves the crossover yield by 0.13pp:

How much room each pair has, at the statutory 30%:

For IVV → CSPX the fee gap would have to widen from 4bp to roughly 16bp before the sign changed. For QQQ → CNDX the yield would have to roughly double, or the fee gap roughly halve. Against a fund yielding 0.4238%, a two-basis-point fee error is about a third of the distance between the yield and its crossover.

Yield definition. Both yields are trailing twelve-month distribution figures. A forward or SEC yield would give different numbers. Distribution yields on equity index funds move with the market as well as with payouts, so a figure accurate on 31 August 2026 is an estimate by the time this is read.

Capital-gains rate and the tax on distributions dominate §6. The spread between the 0% and 25% columns (modelled) is larger than any difference between pairs, and the tax on distributions (modelled) moves the acceleration figures substantially while leaving the simple benchmark, which excludes it, unchanged. For a holder with a large embedded gain, their own tax position matters more than which pair they are switching.

Growth rate. Every grid in §6 uses 5% a year (modelled). The same grids at 0% (modelled), IVV → CSPX, 30% treaty rate — payback in years, simple benchmark / acceleration. “Taxed” and “untaxed” carry the same meaning as in §6: an additional tax on distributions after withholding at the capital-gains rate in that column, with no foreign-tax credit modelled, or no tax beyond withholding.

Reinvested, growth 0% (modelled):

Spent, income-matched, growth 0% (modelled):

none: no horizon within 1,000 years.

Growth changes which basis is harsher in one combination only. Reinvested with distributions untaxed, reinvested with distributions taxed, and spent with distributions untaxed all keep the ordering reported in §6 at both growth rates.

Spent with distributions taxed does not. At 5% growth, acceleration is longer than the simple benchmark at a 60% gain and at a 40% gain with a 25% rate. At 0% growth, it is shorter than the simple benchmark in all six cells where both the rate and the embedded gain are above zero. In both taxed combinations the benchmark excludes the tax on distributions, so the ordering there reflects the tax assumption as well as the method. The magnitudes move a long way in every combination, so no single acceleration figure in §6 should be read without its growth assumption.

QQQ → CNDX at 0% growth (modelled): no simple figure. Acceleration horizons exist only with distributions taxed at a 25% rate (modelled): 156.5 and 990.5 years at 0% and 20% embedded gains when reinvested, and 188.0 years at a 0% gain when income-matched. There is none within 1,000 years in any other cell.

The fund-level proxy, and how far the Nasdaq-100 approximation can be pushed. For IVV → CSPX the 15% figure (modelled) is a treaty rate on US-source income, so there is nothing to test. For QQQ → CNDX it is an approximation, and the question is whether the approximation could change the answer.

Write s for the share of the fund’s gross dividends that is not US-source, and δ for the per-unit withholding differential between the two wrappers on that non-US share. On the US-source portion the differential is 0.15 at a 30% treaty rate. The withholding term becomes:

withholding term  =  D × [ (1 − s) × 0.15  +  s × δ ]

The pair turns positive when that term reaches $120.00, against the $63.58 the flat proxy produces — it would have to rise by a factor of 1.9. Two consequences:

• If δ ≤ 0.15, no value of s flips the result. The non-US portion would have to be treated more favourably by the Irish wrapper than the US-source portion is, per unit of dividend, before s mattered at all.
• At δ = 0.30 (modelled), s would have to exceed approximately 88.7% for the result to turn positive — the fund’s distributions would have to be almost entirely non-US-source.

This is a conditional result rather than a measurement. The actual non-US share is not established here and would need constituent-level dividend data the supplied model does not carry, and δ is not established either. What the calculation establishes is the pair of conditions under which the conclusion would change, not that those conditions fail to hold. Anyone with the constituent data can substitute it into the expression above and read off the answer directly.

The common dividend yield is an approximation, not an identity. Both funds in a pair are modelled as receiving the same underlying gross dividend yield, taken from the US ETF’s trailing distribution yield (modelled). The Irish funds accumulate and distribute nothing, so no observable figure exists for them, and the US ETF’s distribution yield is not the same quantity as its underlying gross yield in any case — distribution timing, cash drag, securities-lending income and fee drag all sit between the two. The gap is small relative to the differences the study measures, but it is an assumption rather than an equality, and every figure derived from y inherits it.

Stated in advance so it can be checked rather than argued about:

• The S&P 500 fee gap widening to 16bp or more, whether from a CSPX increase or an IVV cut. That reverses IVV → CSPX at the statutory rate. A narrowing gap does not falsify the result; it strengthens it.
• Any published QQQ distribution total above roughly $5.72 per share at the current price. That puts the Nasdaq-100 pair above its crossover and reverses the negative result.
• An issuer KID showing a TER different from the justETF figure for either Irish fund. The Irish fees here come from justETF, which is also the source in BorderFolio’s pipeline for Irish ISINs — consistent, but not independent.
• Constituent-level dividend-source data showing the Nasdaq-100’s non-US dividend share above approximately 88.7%, at δ = 0.30 (modelled). Below that share, or at a lower δ, the result stated here stands. Above it, the sign for this pair changes. See §7.
• Per-source-country withholding data for a global fund. This would not falsify anything above; it would extend the analysis to the pairs currently excluded by gate two, where no result is claimed.
• A measured gap between the US ETF’s trailing distribution yield and the pair’s underlying gross dividend yield large enough to move either pair across its crossover. For QQQ → CNDX that means the true underlying yield being roughly double the distribution yield used here.
• Any domestic tax treatment that distinguishes accumulating from distributing funds. This reverses results in either direction. Beyond the two distribution-tax cases reported in §6 and §7, it is out of scope.

One row per document, with the date it was read. There is no site-wide verification date on this study.

Full method: quantroutine.com/methodology

When does switching from a US ETF to an Irish UCITS equivalent pay?

Under this study’s assumptions, the switch is favourable when the distribution yield multiplied by the gap between the investor’s US treaty rate and the 15% fund-level rate (modelled) exceeds the fee gap between the two funds. At the US statutory 30% rate, IVV → CSPX comes out at +$119.00 a year and QQQ → CNDX at −$56.42 a year, on a $100,000 position (modelled). The sign of each result does not depend on position size. Only two withholding layers, fund fees and a one-off switching cost are modelled; the investor’s own domestic taxation of either fund is out of scope.

Is the Irish UCITS version of an ETF always more expensive than the US one?

No. The fee gap is a fact about each pair. CSPX costs 0.04pp more than IVV and CNDX 0.12pp more than QQQ, but on the MSCI World pair the gap runs the other way: URTH charges 0.24% and IWDA 0.20%, a gap of 0.04pp in favour of the switch. These are published expense ratios with no modelled input.

Does switching help investors at a 15% US treaty rate, such as in Germany or the Netherlands?

On withholding and fund fees alone, no. At a 15% treaty rate the withholding term is zero for IVV → CSPX, leaving only the fee difference: −$40.00 a year. For QQQ → CNDX it is approximately zero, leaving −$120.00 a year. Both figures are on a $100,000 position (modelled) with a 15% fund-level rate (modelled). Below 15%, the switch costs withholding rather than saving it. These are statements about withholding only: the domestic treatment of a US fund versus an Irish one is not modelled and can be larger than everything measured in the study.

How long does it take to recover the cost of switching an existing position?

For IVV → CSPX at a 30% treaty rate with a 0% capital-gains rate (modelled), payback is 4.2 years on the simple benchmark, which excludes any tax on distributions, and 4.5 years on the acceleration basis. With a 40% embedded gain and a 15% capital-gains rate (both modelled), it is 54.6 and 75.8 years. QQQ → CNDX has no simple figure, and no acceleration horizon within 1,000 years unless distributions carry an additional tax after withholding at a 25% capital-gains rate (modelled), where the shortest is 111.8 years. These figures use a 0.5% switching friction (modelled), 5% growth (modelled), and distributions reinvested in full and untaxed (modelled). Under an additional tax on distributions after withholding, at the capital-gains rate with no foreign-tax credit modelled (modelled), the simple benchmark is unchanged because it excludes that tax, and the acceleration figure at a 40% gain and a 15% rate falls to 32.8 years. That comparison changes the tax assumption as well as the method: counting the same distribution-tax saving in the simple calculation gives approximately 28.2 years. §6 gives the full grids, including for holders who spend distributions.

Why is URTH to IWDA left out of the withholding figures?

The 15% fund-level rate (modelled) is the US–Ireland treaty rate on US-source dividends. MSCI World funds hold substantial Japanese, British, Swiss and French holdings, and the supplied model contains no per-source-country withholding data, so no bound of the kind used for the Nasdaq-100 pair would be meaningful. The study reports only the fee gap for that pair and claims no withholding result.

What would reverse the negative result for QQQ to CNDX?

Any published QQQ distribution total above roughly $5.72 per share at the current price would put the pair above its crossover. The flat 15% fund-level rate (modelled) could only change the sign if the per-unit withholding differential on the non-US share of dividends exceeded 0.15; at a differential of 0.30 (modelled), the non-US share of gross dividends would have to exceed approximately 88.7%. That is a conditional result, not a measurement of the actual share.

Withholding calculations, model documentation and the reproducible input register were contributed by BorderFolio, a portfolio tracker for international investors with a cross-border dividend withholding layer — borderfolio.app