Best ETFs 2026: What the Data Can Tell You, and What No Ranking Will

Typing “best ETFs” into a search engine returns rankings. This page returns data: the expense ratios, fund sizes and tracking records actually published in 2026, category by category, followed by what those numbers cannot decide, and how the rankings that claim to decide it are financed.

1. What “best” promises, and what the data measures

A superlative is a claim about the future: the best fund is the one that will serve you better than every alternative. No published dataset contains that information. What issuers and databases publish is narrower and more useful: the fee charged, the assets gathered, the gap between fund and index over past periods, the spread quoted on an exchange. Every one of those is verifiable, dated and comparable. None of them is a verdict.

The distinction is not pedantry; it changes what a reader can do. A verdict has to be trusted. A fee table can be checked. “Best ETF” is a claim; lowest verified cost is a fact. This page works exclusively with the second category, and the full ETF selection framework supplies the method for turning such facts into a decision, criterion by criterion. What fails first when reading fund data is rarely the arithmetic; it is the interpretation, and common mistakes when reading ETF data catalogs the recurring ones.

One superlative survives scrutiny: the factual kind. The lowest published fee of a category on a given date, the largest fund by assets on a given date, the smallest realized tracking difference over a stated period. Those appear below, each with its source and its date. They describe the field; they do not rank it.

Why the superlative question persists is worth naming, because the answer frames everything below. Aggregation is costly: reading five prospectuses to learn that the funds differ by one basis point feels like wasted effort, and a ranking sells the shortcut. The dominant reading follows: somewhere in the pile sits a fund objectively superior to the rest, and the job is to find it. The data supports a different picture. In the core categories, the leading products have converged so completely on price and index that the remaining differences are matters of account type, platform and structure, questions about the holder, not the fund. The search for the best fund fails not because the data is hidden but because the question points at the wrong object.

2. The criteria that can actually be measured

Four numbers carry the comparison. The expense ratio is the announced annual charge, published in the prospectus and deducted daily from net asset value. The tracking difference is the realized gap between fund and index over a period, published in annual reports; it absorbs fees, transaction costs, withholding taxes and securities-lending income, which is why it can beat the fee or undershoot it. Assets under management proxy survival odds and spread quality; the arithmetic floor is unforgiving, since a 0.03% fee on a sub-100-million fund cannot pay for an index license and a listing. The bid-ask spread is the cost of each transaction, visible on the order book and nowhere else. Of the four, the tracking difference is the least quoted and the most informative: it is the only number that records what the fund actually delivered rather than what it announced, and comparing it to the fee reveals the fund’s internal economics, from sampling skill to securities-lending policy, in a single subtraction.

Two framing rules keep those numbers honest. First, all of them are nominal readings on a nominal product: what a saver ultimately keeps is set one level up, by inflation, and real versus nominal returns covers why a fee comparison means little without that adjustment. Second, size and volume describe normal conditions; the liquidity risk hidden in ETF wrappers documents how quoted spreads behaved when the underlying markets froze, which is the scenario in which the liquidity criterion earns its place in the table.

The arithmetic behind the fee number deserves one paragraph, because its scale surprises in both directions. On $10,000, the gap between a 0.02% and a 0.0945% fund is $7.45 a year, lunch money. Compounded over 30 years on a portfolio growing at 6% gross, the same gap removes about 2.2% of terminal capital, most of a year’s typical withdrawal. Both statements are true; which one matters depends on the horizon, which is why the simulator below puts the horizon under your control rather than asserting one. The spread follows the same logic at a different frequency: it costs nothing to a holder who never trades and compounds ruthlessly for one who rebalances monthly, and it varies within a single day, widest at the open and close, tightest mid-session when the underlying markets are fully priced.

Everything below is those four numbers, applied to the categories US investors actually hold.

3. The state of play by category

Three categories cover the bulk of indexed assets: large blend, total market, and the aggregate bond universe. Figures are issuer publications and fund databases, each dated; they move, and the date is part of the data.

3.1 Large blend: the S&P 500 trackers

The factual superlatives, as of the sources above: SPLG carries the lowest published fee of the category at 0.02%, VOO the largest asset base, having crossed the trillion-dollar mark first among all ETFs, and SPY the deepest trading volume, at roughly $62 billion a day. The fee spread across the table, 0.02% to 0.0945%, is a factor of nearly five for the same 500 stocks. SPY’s unit investment trust structure explains part of its premium: it cannot lend securities or reinvest dividends between distributions, constraints its younger rivals do not carry. Which of the three properties matters, the cheapest fee, the largest base or the deepest book, depends entirely on whether the holder trades once a decade or once an hour; the table cannot know that.

The convergence at the bottom of the table is economics, not generosity. At 0.03%, a fund holding $686 billion still generates roughly $200 million a year in fee revenue, enough to run an index-tracking operation many times over; scale turned the fee war into a war the largest issuers could win at near-zero prices. SPY sits apart for a historical reason: launched in 1993 as the first US-listed ETF, it was built as a unit investment trust because that was the structure available, and its fee reflects a franchise whose clientele, traders and institutions pricing the deepest order book in the world, is largely insensitive to 6 basis points. Two products, one index, two different customers: the table records the split without resolving it.

A note on the empty cells, here and below: asset figures appear only where a dated source was verified for this revision. A blank is a refusal to estimate, not an oversight; the issuer page carries the live number, and an undated figure would defeat the purpose of the table.

3.2 Total market

Fees converged to 0.03% across the category (issuer pages, early 2026), which retires price as a tiebreaker. What remains are the indices: three different providers, three definitions of “the market”, differing mainly in small-cap depth and inclusion rules. The funds are near-substitutes, and near is doing quiet work in that sentence: the pairs are similar enough that switching between them for a tax loss invites the wash sale question, and different enough that returns diverge by basis points in small-cap-led years. The index names deserve a reading, because “the market” turns out to be an editorial decision: each provider draws its own line on minimum size, liquidity and inclusion timing, so the three funds hold several thousand constituents each but not the same several thousand. In large-cap-led years the difference is invisible; when leadership broadens, the provider’s small-cap rules become the only active decision inside a passive product. The extension from these tables to global exposure changes the question from fees to geography, and regional allocation inside global ETFs shows what a world label actually contains.

3.3 The aggregate bond universe

Identical fees, near-identical universes, one structural difference: BND’s float-adjusted index strips out bonds parked on central bank balance sheets, which tilts it marginally relative to AGG’s standard weighting. Both exclude TIPS, high yield and municipals despite the word total in one name and core in the other. The category’s defining exposure is not in the table at all: duration, currently translating a one-point rise in yields into a mid-single-digit price decline for either fund. With intermediate durations in the six-year area, both funds stand to lose roughly 5 to 6% of price per percentage-point rise in yields, and to gain the same on the way down; 2022 executed the first branch, handing the aggregate universe a double-digit annual loss with no help from fees whatsoever. That sensitivity, not the third decimal of a fee, decided bond fund outcomes in 2022, and bond ETF behavior across rate regimes carries the full analysis.

3.4 The thematic shelf, briefly

Thematic and factor products dominate launch calendars and ranking clicks while holding a fraction of indexed assets. Their published numbers deserve the same treatment, with one addition: the index itself needs auditing, because a theme label constrains almost nothing. What smart beta labels actually hold runs that audit on the factor shelf; the recurring finding is concentration sold as sophistication. Fee dispersion is wider there than anywhere in the tables above, and the burden of proof runs the other way: the product has to justify the premium over the 0.03% core, in a return stream that survives its own costs.

The launch calendar itself is a signal worth reading. Thematic products are created where demand already exists, which means after the theme has performed: the fund arrives once the story is priced, and its early holders systematically enter at rich valuations. That sequencing, launch follows performance, is a structural property of the product cycle, not a flaw of any single issuer, and it explains a pattern documented across the category: the gap between the returns thematic funds print and the returns their investors capture. A ranking of thematic funds by trailing return is, almost by construction, a ranking of stories the market has already told.

4. The fee gap, compounded

Everything in the tables reduces to one live question: how much does a given fee gap cost over a given holding period, under return conditions nobody can promise. The simulator below takes two annual fee levels and a horizon, and draws the terminal wealth gap between them across a full range of return assumptions, negative years included. No market forecast is embedded and none is asked of you: the horizontal axis is the assumption. The point it makes is mechanical: the gap between two fee levels is not the fee difference, it is the fee difference compounded on an ever-larger base.

5. What no ranking measures

Every table above was compiled in one macro configuration and will be read in another. That is the structural blind spot of the genre. A fee comparison is regime-proof: 0.02% costs less than 0.0945% in every state of the world. A performance-based ranking is not: it encodes the regime it was computed in. The funds that topped return screens in the disinflationary 2010s were long-duration and growth-tilted precisely because that regime rewarded those exposures; 2022 re-priced the same funds to the bottom of the same screens without a single fee changing. The magnitude of such re-ranking is not subtle: global developed equities swung from about minus 18.1% in 2022 to plus 23.8% in 2023 (iShares URTH factsheet, March 2026), a 42-point reversal that reshuffled every trailing-return table it touched. A ranking compiled in December 2022 and one compiled in December 2023 would crown different funds for reasons that have nothing to do with any fund’s quality.

Where the environment sits is a measurement question. The live regime dashboard currently reads a transition state: mixed cyclical signals, no clear growth-inflation direction (Eco3min classifier, June 2026 reading). The base rates for what that implies come from history: how asset classes behaved in each regime compiles the record since the 1970s, and matching vehicles to the macro regime translates it from asset classes to the wrappers that hold them. Read together, they answer the question rankings dodge: not which fund won lately, but which exposures the current configuration has historically rewarded and punished. A transition state answers cautiously, which is itself information: it favors the criteria that hold in every regime, cost and structure, over the ones that shine in one. It also dates this page’s own tables in advance: the categories above are ranked by assets gathered during a specific rate and growth configuration, and the next configuration will redistribute those flows. The numbers will be revised each January; the reason they need revising is the subject of this section.

6. How to read a commercial ranking

Ranking pages are an economic product, and their business model is observable. Most operate on affiliate agreements: the site earns a commission when a reader opens an account or invests through a link. The mechanism does not make the content false; it shapes the frame. A methodology that needs a winner will produce one, because a page concluding “the candidates are near-substitutes; the data cannot separate them” generates fewer clicks on partner links than a podium does.

Three checks extract the useful part of any ranking. Whether the criteria are published and verifiable, or summarized in a proprietary score whose weights are editorial choices. Whether the data is dated, since an undated fee is a rumor. And whether the universe is complete, because inclusion is the quietest filter: a comparison restricted to funds with referral agreements is a shortlist wearing a ranking’s clothes. Selection bias compounds the frame: screens built on past returns systematically harvest the winners of the regime just ended, the exact tilt that section 5 prices. Which regime just ended, and which one is starting, is the question the equity and ETF pillar on cycles and monetary regimes is organised around.

None of this requires imputing bad faith. It requires reading a ranking the way the tables above are meant to be read: as inputs with sources and dates, feeding a method the reader owns.

A falsifiability test compresses the three checks into one question: what result would this methodology be unable to produce? A scoring system whose weights are editorial can justify nearly any podium after the fact; a fee-and-size table can be contradicted by a single screenshot of an issuer page. The more a ranking’s conclusion depends on unpublishable judgment, the less information its order contains. Survivorship adds a quieter distortion: closed funds vanish from trailing-return tables, so the visible field always looks more successful than the field that existed when past readers made their choices.

7. The instruments that game the screens

Volume and return screens reliably surface products built for neither patient holding nor benchmark fidelity. Leveraged and inverse funds top short-term return tables after every sharp move, mechanically, since a daily multiple of a rally is a bigger rally; how inverse ETFs decay quantifies what the same daily reset does over weeks. Their presence at the top of a screen is a property of the screen, not evidence about holding-period returns.

The subtler screen distortion is structural. ETFs and mutual funds report performance under the same rules but distribute taxes under different mechanics, so an after-tax comparison can invert a pre-tax one; ETFs versus mutual funds, mechanically traces the in-kind machinery behind the difference. A ranking that mixes the two wrappers without saying so is comparing gross figures across products whose net figures diverge by construction.

The common thread in both cases is that screens measure what is easy, not what is held. Trading volume measures traders; short-window returns measure the regime; expense ratios measure the one variable issuers publish to the basis point precisely because it photographs well. None of those is the holding-period, after-tax, after-inflation return of an actual saver, which no screen publishes because no two savers share one. The tables in section 3 stay useful precisely by claiming less: they are the stable inputs, and the missing variables belong to the reader.

8. FAQ

What does “best ETF” mean in measurable terms?

Nothing directly: superlatives about the future are not observable. What can be measured and dated are the published expense ratio, realized tracking difference, assets under management and quoted spread. Factual superlatives within those metrics exist, such as the lowest published fee of a category on a given date, and they carry a source rather than a promise.

Why can a higher-fee ETF end up ahead of a cheaper one?

Because the fee is one input into the realized tracking difference. Securities-lending revenue, index sampling choices, withholding tax efficiency and transaction costs all enter the final gap between fund and index. Annual reports regularly show funds tracking tighter than their fee alone would predict, and occasionally finishing ahead of their benchmark.

How are commercial ETF rankings typically financed?

Mostly through affiliate arrangements: the publisher earns a commission when readers open accounts or transact through outbound links. The model rewards pages that produce a clear winner and penalizes inconclusive verdicts. It does not automatically falsify the content, but it explains why methodologies favor podiums over the finding that leading funds are near-substitutes.

Which sources publish verifiable ETF cost data?

Issuer pages and prospectuses publish expense ratios; annual and semi-annual reports publish realized tracking against the benchmark; exchanges publish quoted spreads and volumes; fund databases aggregate fees and assets with a reference date. Each figure is dated at the source, which is what separates a verifiable number from a ranking’s summary score.

9. From tables to decisions

The data on this page ages on purpose: every figure carries its date, and the January revision will replace them. What does not age is the structure of the field the figures describe. Passive vehicles keep absorbing flows, which changes what markets are, a shift examined in how passive flows shape market structure. And the fund-level comparison remains the smallest decision in the chain: sizing, mixing and holding exposures across conditions is the layer treated in portfolio construction across market regimes, and the vehicle-by-vehicle view continues in the 2026 panorama of investment vehicles. Rankings will keep promising the shortcut. The tables above are slower and hold up better: numbers, sources, dates, and a method that stays yours.

Last updated — 15 September 2026