How to Choose an ETF: The 2026 Framework (Index, Costs, Liquidity, Taxes)

Rankings promise a winner. The data only knows measurable criteria: what the fund tracks, how it tracks it, what it charges all-in, whether it trades at a fair price, and how the wrapper is taxed. This page turns those five criteria into a working grid, then reads the grid through the macro regime.

1. What choosing an ETF means once the wrapper is understood

This page assumes the mechanics are already clear. For what an ETF is and how it works, the creation and redemption process, and the role of authorized participants, the primer covers the ground; nothing here re-explains it. The decision problem starts after that point, when several hundred funds offer overlapping exposures and the differences hide in documents most buyers never open.

The framing matters. Marketing presents ETF selection as a search: scan a ranking, spot a winner, click. The arithmetic points the other way. On a given index, the funds are near-substitutes, and the persistent differences between them are costs, structure and size, all of which are published. Screening on those criteria removes most of the field in minutes. You do not find the right ETF; you eliminate the wrong ones.

Elimination needs measured inputs, not adjectives. The 2026 data on ETF costs and size compiles the observed figures category by category; this page supplies the grid that makes those figures usable. Five criteria, in decreasing order of consequence: the index, the replication method, the total cost, the assets under management, and the tax structure of the wrapper. Each is observable from public documents. Each has a known trap. And each carries a different weight depending on the investor: the tax criterion dominates in a taxable account and nearly vanishes in an IRA, while the liquidity criterion barely registers for a monthly contribution plan and becomes central for anyone who might exit in a hurry.

2. The five-criterion grid

The grid below is the whole method in one table. Everything after it is commentary, criterion by criterion.

One practical note before the detail: the grid is only as good as the account it feeds into. A fund screened to perfection still trades through an intermediary, and evaluating the broker that will hold them is a separate exercise with its own criteria, covered in a dedicated page. The two decisions interact: commission-free trading changed which fee layers matter, and cash sweep rates can cost more than any expense ratio on the uninvested portion.

3. Criterion by criterion

3.1 The index before the ticker

Two ETFs on different indices are different investments, whatever their names suggest. The MSCI World index carries 1,283 stocks across 23 developed markets, yet the United States alone weighs 72.45% of it (MSCI, June 30, 2026) and information technology 30.66% (MSCI, May 29, 2026). A saver who wanted balanced global exposure has, in fact, acquired a concentrated position in one country and one sector. The hidden concentration inside global equity ETFs documents how that weight built up and what it implies.

The concentration is not an accident; it is the weighting rule at work. Capitalization weighting sizes each position by market value, which means the index rebalances itself toward whatever has already risen. The property cuts both ways: it keeps turnover and costs low, and it embeds a momentum tilt that compounds when leadership persists and reverses painfully when it rotates. An investor comparing a capitalization-weighted fund to an equal-weighted one on the same universe is comparing two different bets, not two prices for the same thing.

The comparison across index providers repeats the lesson at another scale. MSCI ACWI adds emerging markets at a 12.35% weight and still assigns 63.59% to the United States (MSCI, May 29, 2026), while FTSE All-World covers 4,254 constituents against ACWI’s 2,513 (FTSE Russell and MSCI data, 2026). Same words on the label, different portfolios inside. Weighting rules diverge further once a fund leaves plain capitalization weighting: the factor risk behind smart beta ETFs shows how a screen sold as diversification can concentrate exposure instead.

Reading the index methodology sheet takes ten minutes. It is the highest-yield ten minutes of the whole selection process.

3.2 Replication and structure

How the fund delivers the index is the second criterion. Full physical replication holds every constituent; sampling holds a representative subset; synthetic replication holds a substitute basket and swaps its return with a counterparty. The trade-offs between fidelity, cost and counterparty exposure are documented in physical versus synthetic replication, and they are not cosmetic: sampling can drift from the index in stressed markets, and a swap introduces a bank on the other side of the return. Collateral rules cap the exposure, but capped is not zero, and the cap is tested precisely in the conditions where it matters.

Structure also separates ETFs from their older cousins. The mechanical difference between ETFs and mutual funds runs through creation and redemption: an ETF exchanges baskets of securities in kind with authorized participants, while a mutual fund meets flows in cash, liquidating holdings when redemptions arrive. That single mechanical fact drives the tax behavior discussed below, and it explains why an ETF’s price can deviate from its net asset value intraday while a mutual fund transacts once a day at NAV. It also explains who pays for other people’s exits: in a mutual fund, remaining holders absorb the trading costs of departing ones; in an ETF, the departing holder pays the spread.

3.3 Total cost: expense ratio, tracking difference, spread

The expense ratio is the visible layer. On S&P 500 exposure, SPLG charges 0.02%, VOO and IVV 0.03%, and SPY 0.0945% a year for the same 500 stocks (issuer figures, May 2026): on $10,000, that is $2, $3 and $9.45 respectively. Small numbers, but they compound for decades, and they are the one cost known in advance. The compounding is worth stating in full: a 0.30 point annual fee gap on a portfolio growing at 6% gross leaves roughly 8% less capital after 30 years, an entire year of returns surrendered to a line item most holders never re-read after purchase.

The second layer is the tracking difference: the gap between the fund’s realized return and the index’s, measured over one or several years in the annual report. It bundles the fee, transaction costs, withholding tax treatment and securities-lending revenue. A fund can charge more and track better; a fund can even finish ahead of its index when lending revenue exceeds costs. Tracking difference is a result; the expense ratio is a promise. The distinction with tracking error matters too: tracking error measures the volatility of the gap, tracking difference its cumulative size. A long-horizon holder cares about the second; an arbitrageur cares about the first.

The third layer is the bid-ask spread, paid at every transaction. On the largest US-listed funds it rounds to a basis point or less; on small or exotic listings it can reach several dozen. The spread is invisible in any fee document and fully visible in the order book, which is where it gets checked, ideally at mid-session rather than at the open or close, when quotes are widest. Its weight depends entirely on turnover: paid twice over a twenty-year holding, it is noise; paid monthly by a strategy that rotates, it becomes the dominant cost. The simulator below makes that asymmetry explicit.

Order mechanics finish the cost picture. A market order accepts whatever the book shows at that instant, spread included; a limit order caps the price paid and turns the spread from a tax into a negotiation. The difference is invisible on a calm Tuesday and worth whole percentage points in a fast market, when quoted spreads on the same fund can widen sharply within minutes. Execution is the one cost layer the holder controls entirely at the moment of trading, which makes it the cheapest one to fix.

3.4 AUM and liquidity

Assets under management condition three things: the spread, the probability the fund survives, and its behavior in stress. Large funds attract market makers and arbitrage flow, which tightens quotes; sub-scale funds face closure, a taxable event imposed on holders at a date they did not choose. The economics of closure are plain arithmetic: a 0.05% fee on $30 million of assets generates $15,000 a year, which does not cover an index license, administration and listing fees, so the issuer eventually pulls the product. Fund closures are routine industry events, and each one converts a paper position into a realized gain or loss on the fund’s schedule rather than the holder’s.

The stress dimension is the least visible and the most consequential. In March 2020, quoted prices on some fixed income ETFs moved away from stale net asset values as the fund became the price-discovery venue for its underlying market. ETF liquidity under market stress examines those episodes and what the wrapper can and cannot promise. The nuance worth keeping: an ETF is at most as liquid as what it holds, plus a layer of secondary-market netting on top. AUM is a reasonable proxy in calm markets and an incomplete one in fast markets.

3.5 Tax structure: the criterion most portfolios underweight

The wrapper’s tax mechanics are the last criterion and, for taxable accounts, often the decisive one. The in-kind creation and redemption process lets an ETF hand appreciated securities to authorized participants instead of selling them, which is why broad equity ETFs rarely distribute capital gains while mutual funds routinely do. A holder who never sells can still receive a taxable distribution from a mutual fund because someone else sold; the ETF structure largely severs that link. Distributions also differ in kind: qualified dividends, non-qualified income and return of capital carry different rates, and the mix is published in the fund’s tax documents. A companion piece: physical versus synthetic ETF replication compared.

Two further mechanics deserve a check. The wash sale rule disallows a loss when a substantially identical security is repurchased within 30 days, which constrains tax-loss harvesting between near-identical index funds and rewards pairs that track similar but distinct indices. And the listing venue matters for cross-border holders through treaty withholding rates on dividends, a structural difference that no fee cut can offset. None of this changes the index; all of it changes the after-tax return, and the after-tax return is the only one spent. The account question extends one level up, to the choice between tax-advantaged and taxable wrappers, which precedes any fund screen.

4. The total cost of ownership, made visible

The simulator below compares two generic fee profiles under one return assumption that you set: profile A with a low expense ratio and a wide spread, profile B with a higher expense ratio and a tight spread. It shows how the gap in terminal value builds year by year, and where one cost structure overtakes the other. No fund is named and neither profile is designated preferable; the point is the arithmetic, and the arithmetic depends on the holding period you choose. The two profiles bracket the ranges observed on the market: expense ratios on MSCI World UCITS funds span 0.05% to 0.50% a year across 31 funds (justETF, July 5, 2026), and spreads run from about a basis point on flagship listings to several dozen on thin ones.

5. The grid, read through the macro regime

The five criteria are stable; their relative weight is not. In a regime of low expected returns, cost differences absorb a larger share of the outcome: a 0.20 point fee gap is background noise against a 12% year and a meaningful slice of a 3% year. In a stress regime, liquidity moves to the front: the spreads paid in March 2020 dwarfed a decade of expense ratio differences for anyone forced to trade that week. The regime shift of 2022 made the same point through the index criterion: MSCI World returned about minus 18.1% that year and plus 23.8% the next (iShares URTH factsheet, March 2026), a 42-point swing that no fee optimization touches. The grid does not change; the sorting order does.

Placing the current environment is a measurement problem, not an opinion. At the time of writing, the current macro regime reads as a transition state on the Eco3min classifier: mixed cyclical signals, no clear growth-inflation direction (June 2026 reading). The classifier is rules-based and its thresholds are published; how Eco3min classifies macro regimes details the three-axis grid behind the verdict, and the reading updates daily as the underlying indicators cross their thresholds.

History supplies the base rates. Asset performance across macro regimes compiles how the main exposures behaved in each configuration since the 1970s, and reading investments through the macro cycle extends the exercise from asset classes to the vehicles that carry them. Neither turns the grid into a timing device; both tell a holder which criterion is likely to be tested next. A transition state, by definition, does not say. That is an argument for weighting the criteria that survive every regime, cost and structure, over the ones that shine in one.

6. The bond special case

Everything above applies to bond ETFs with one addition: the index has a duration, and duration is a direct exposure to the rate cycle. The first-order arithmetic is unforgiving: a fund with a seven-year duration loses roughly 7% of its value for each percentage point rise in its yield, a move that rate regimes deliver routinely and that swamps any fee difference between competing funds. In 2022, that arithmetic turned aggregate bond indices into double-digit losers, an outcome written into the index, not the wrapper.

The consequence for the grid: on the fixed income side, criterion one carries even more of the decision than in equities, because choosing a bond index means choosing a duration, a credit mix and a currency exposure in a single line. The mechanics, the 2022 episode, and the way duration interacts with the cycle are treated in bond ETFs read through the rate regime; the fee and liquidity criteria then apply unchanged, with the caveat that bond ETF spreads widen faster than equity ETF spreads when the underlying market seizes.

7. The traps that survive the grid

Three failure modes recur even among buyers who screen carefully. The first is folklore: performance chasing across near-identical funds, spread-blind orders at the open, the belief that a larger fund always returns more, the conviction that a distributing share class pays something an accumulating one does not. The most common ETF misconceptions catalogs them with the data that refutes each one.

The second is instrument confusion. Leveraged and inverse products carry ETF branding and daily-reset mechanics that make them behave nothing like their index over weeks or months; the long-term risk of leveraged ETFs quantifies the decay. They fail the grid at criterion one, since the index they track is a daily multiple, not the market. Their presence at the top of volume rankings says something about who trades them, and nothing about whether they belong in a portfolio built to be held. A portfolio built to be held is also what the standard robo-adviser mandate assembles, from a deliberately narrow menu of broad index funds.

The third is scope creep: treating the fund decision as the portfolio decision. A perfectly screened ETF in the wrong proportion, or the wrong account, solves the small problem and leaves the large one. The grid ranks vehicles; it does not size positions, set horizons, or decide what share of a portfolio equities deserve in the first place. Those questions sit upstream, and no expense ratio answers them.

8. FAQ

What criteria distinguish two ETFs tracking the same index?

Four observable ones: the expense ratio, the realized tracking difference published in annual reports, the bid-ask spread on the relevant exchange, and assets under management. Replication method and distribution policy separate funds further. On identical indices, these criteria, not past returns, explain persistent performance gaps between funds.

How do expense ratio and tracking difference differ?

The expense ratio is the announced annual fee. The tracking difference is the measured gap between fund return and index return over a period: it includes the fee plus transaction costs, withholding taxes and securities-lending revenue. Funds are compared on realized tracking difference; the expense ratio only sets an expectation of it.

How does the bid-ask spread affect long-term returns?

The spread is paid at each transaction, on entry and on exit, so its weight depends on turnover. A holder who trades twice in twenty years amortizes it to nearly nothing; frequent rebalancing multiplies it. On the largest funds it rounds to a basis point; on thin listings it can exceed the annual fee.

How are US-listed ETFs taxed compared with mutual funds?

The in-kind creation and redemption mechanism lets ETFs transfer appreciated securities out of the fund without selling, so broad equity ETFs rarely distribute capital gains. Mutual funds meet redemptions in cash and routinely distribute realized gains to all remaining holders. Dividend taxation is identical; the structural difference sits in embedded gains.

What does AUM indicate about an ETF’s viability?

Assets under management proxy three risks: closure risk, since sub-scale funds get liquidated and force a taxable exit; spread width, since scale attracts market makers; and stress behavior, documented in past liquidity episodes. AUM says little about future returns, which the index, not the size, determines.

9. Where this grid sits in the larger decision

Fund selection is the last mile of a longer chain. Upstream sits the structural shift that made this page possible at all: index funds now anchor equity markets, and the market structure of passive investing examines what that concentration of flows does to pricing. Further upstream still sits the allocation itself, the split across asset classes that dominates any fund-level optimization; the broader architecture of asset allocation covers that layer. The grid on this page assumes those questions have answers. When they do, choosing the vehicle stops being a matter of taste: five public criteria, one afternoon of reading, and a field that eliminates itself.

Last updated — 30 July 2026