What an ETF really costs

Imagine, for teaching purposes, two ETFs replicating the same index. The first declares annual costs of 0.12%, the second 0.20%. The conclusion seems immediate: the first costs less. Then you look at the returns. In a given year the first lags the benchmark by 0.29%, the second by 0.18%. The TER was not false; it was only part of the story.

An ETF’s cost does not live in a single number. Some items are declared in advance, others emerge from the fund’s activity, others still depend on the exchange, the broker, the currency and the investor’s behaviour. Confusing them produces two opposite errors: ignoring real costs, or attributing to the product expenses that belong to the operation or to the account.

Four levels, four questions

In the industry this perimeter has a name: TCO — Total Cost of Ownership. It is broader than the sum of the TER and the fund’s transaction costs, because it also includes the costs borne by the investor, such as broker commissions and the buy-sell spread.

No level substitutes for the others. The TER belongs mostly to the first; the tracking difference photographs the net effect of the fund’s costs and of the other components of replication; spreads and commissions belong to the fourth.

TER: what it really says

TER stands for Total Expense Ratio. In European commercial communication the term is often used as shorthand for the fund’s ongoing costs, expressed as an annual percentage of the assets. Regulatory documents may use different labels — for example ongoing costs, ongoing charges, or administrative and management costs — which are not always perfectly interchangeable.

The economic principle is simple: those costs are taken from the fund’s assets and are progressively reflected in the NAV. The investor does not normally receive a separate annual invoice.

If an ETF has a recurring cost of 0.20%, it does not mean the manager deducts €20 from every €10,000 in one go on 31 December. The cost accrues over time and is embedded in the value of the unit. The €20 example is useful for grasping the order of magnitude; it does not describe the daily accounting, nor does it guarantee that the final gap to the index will be exactly 0.20%.

The denominator changes. An annual percentage applies to the assets, which vary with markets, flows and distributions. That is why multiplying the initial capital by the TER is not enough to reconstruct precisely the economic amount borne during a year. The document’s date also matters: a cost reduction announced mid-year, a newly launched class or a temporary expense waiver can make the current number and the previous full-period return poorly comparable.

The TER does not contain everything. The SEC notes, in the bulletin updated on 23 July 2025, that even funds presented as “zero expense” can generate indirect costs, and that the expense ratio does not necessarily include the transactions the fund carries out on the underlying securities.

In the European market, PRIIPs documents expose broader categories separately, including the portfolio’s transaction costs estimated under the regulatory methodology. Adding ongoing costs and PRIIPs transaction costs produces a declared overall cost broader than the TER. It still does not produce the investor’s absolute total cost: the unit’s spread, broker commissions, currency conversion and personal taxation remain outside, among other things.

The useful question is therefore not “which is the true number?”, as if all the others were false. It is: which perimeter does this number cover, at which date, and with which methodology?

Tracking difference: the result of replication

The tracking difference compares the fund’s return with the benchmark’s over the same period.

Tracking difference = fund return − index return

With this convention, a tracking difference of −0.25% means the fund returned 0.25 percentage points less than the index. Some sources use the opposite sign, calculating index minus fund. Before comparing two values, check the convention.

Example: the total return index returns 7.40% and the ETF 7.17% over the same interval. The tracking difference is:

7.17% − 7.40% = −0.23%

This −0.23% is closer to the fund’s economic experience than the TER alone, because it embeds the net effect of many components. It is not, however, an accounting decomposition: it observes the result, it does not automatically identify the causes.

Why it can differ from the TER

The distance between fund and index can reflect:

• recurring costs;
• taxes and withholdings on the portfolio’s income;
• costs of buying and selling the underlyings;
• sampling, optimisation and rebalancings;
• uninvested cash and dividend reinvestment lags;
• net securities-lending revenues;
• the economic terms of a swap;
• valuation, timing and calendar differences;
• changes in the index’s composition;
• corporate actions and rounding;
• differences between the observed class and the declared benchmark.

Some items normally worsen the replication; others can offset part of the costs. Net securities-lending revenues, for example, can narrow the gap, but they introduce an activity and a risk that deserve separate analysis. A tracking difference better than the TER does not prove the fund is free, nor that every source of revenue is risk-free.

The benchmark must be comparable

Before the calculation, at least five checks are needed:

• Right index: not a comparative benchmark different from the replicated index.
• Right version: price, net return or gross return.
• Consistent currency: or a conversion carried out with the same methodology.
• Identical period: same start and end dates.
• Right class: accumulation and distribution require series consistent with the income.

An ETF that looks far from a gross return index can be perfectly consistent with the corresponding net return version, because the two benchmarks embed different tax assumptions. The wrong comparison produces a false tracking difference with many decimals.

Tracking difference is not tracking error

The two terms describe different objects.

Technically, the tracking error is the standard deviation of the differences between the periodic returns of the fund and the index, generally annualised. The frequency of the observations and the calculation period must be declared.

A fund can sit roughly 0.20 points below the index every year: negative tracking difference but low tracking error. Another can close near the index on average while alternating positive and negative gaps: modest average tracking difference but higher tracking error.

Neither measure is, on its own, a universal grade. For an indexed replication, stability and gap are central information; for an active ETF, deviation from the benchmark can be part of the mandate, not an execution defect.

Declared, calculated, observed

A tracking difference can come from different paths:

• declared by the issuer, with its own period and methodology;
• calculated from official series of the fund and the index;
• derived from returns published in documents or charts;
• estimated from market data, when a traded price is compared with an index.

These paths are not equivalent. The exchange price embeds premium, discount and microstructure, as seen in chapter 3; the NAV represents the fund; a rebased chart can have limited precision; a declared figure can refer to a financial year different from the rolling twelve months.

The correct editorial rule is to keep together value, period, source, sign convention and method. Without these metadata, two equal numbers can mean different things, and two different numbers can both be correct.

The portfolio’s internal costs

An ETF must turn a methodology into real positions. This activity produces frictions that do not coincide with the management cost.

Transactions on the underlyings. When the fund buys or sells securities it can incur spreads, commissions, transaction taxes and market impact. Turnover helps gauge how often the portfolio changes, but it does not measure the cost on its own: trading very liquid government bonds is not the same as trading high-yield bonds or frontier-market shares. The cause of the turnover also matters: an index can change composition; the fund can receive flows; a sampled replication can be optimised; an active ETF can change positions at the manager’s choice.

Full replication and sampling. Buying every component removes one source of theoretical deviation, but it can be inefficient when the index holds thousands of securities or illiquid instruments. Sampling avoids some transactions but introduces model risk and composition differences. There is no automatic equation “full replication = lower costs”.

Swaps and other derivatives. In a synthetic replication the return can come from a swap contract. Part of the economic cost can be embedded in the contract’s terms without appearing as a separate fee paid by the reader. Quality must be assessed together with collateral, counterparty, resets and actual performance — not through the TER alone.

Securities lending. The fund can temporarily lend some securities and earn revenues. After the share retained by intermediaries or the manager, the net benefit can offset replication costs. The positive figure must not be isolated from counterparty risk, collateral and the revenue-sharing policy.

Taxation inside the fund. Before the return reaches the investor, dividends and interest can suffer withholdings in their countries of origin. The fund’s domicile, treaties, replication structure and the nature of the underlying can change the result. This level is distinct from the reader’s personal taxation, which depends on the country of residence and will be covered in the tax chapters. Saying that two ETFs have the same TER therefore does not mean they have the same internal tax efficiency.

The costs borne by the investor

Chapter 4 separated spreads, book depth, impact and commissions. Here those costs enter the overall picture.

One-off and recurring costs. An annual cost of 0.20% repeats as long as the capital stays invested. A €5 commission, or half the spread, materialises when you trade. Comparing them requires amount, frequency and horizon. On a €1,000 purchase, a fixed €5 commission equals 0.50% upfront; on €10,000 it equals 0.05%. The nominal cost is identical; the incidence is not.

If an overall entry-and-exit cost is 0.40% and the investment lasts ten years, simply dividing it by ten gives only a first approximation of 0.04% per year. It ignores time, compounding, changes in value and asymmetric costs — but it avoids comparing a one-off expense directly with an annual percentage as if they were homogeneous.

Currency conversion. The listing’s currency neither creates nor removes the portfolio’s currency exposure, as seen in chapter 3. It can, however, produce an operational cost if the broker must convert money to trade that line. The conversion cost belongs to the relationship between investor, account and trading venue — not to the fund’s TER.

Custody, platform and advice. Account fees, custody costs, percentage platform fees and advisory compensation can weigh on the investor’s return. They must not be attributed to the ETF, but they cannot be ignored when the personal net result is calculated.

Personal taxation. Taxes on distributions, capital gains, wealth or transactions depend on tax residence, vehicle, account and national rules. There is no single “European tax cost”. That is why the series devotes separate chapters to taxation and does not compress it into a universal indicator.

How much a small difference weighs

Annual differences compound. Suppose, purely for teaching purposes, an initial investment of €10,000, a constant 6% gross return and no other flows. With a hypothetical total cost of 0.20%, the simplified net return is 5.80%; with a cost of 0.60%, it is 5.40%. After twenty years:

The example is not a forecast: returns and costs are not constant, and the linear subtraction is a simplification. It does show, however, why a few recurring tenths deserve attention over long horizons.

The conclusion is not “always pick the lowest TER”. A different exposure, a fragile replication, an expensive listing or a different tax treatment can dominate the apparent advantage. First verify that the products really do the same job; then compare the costs.

How to read Rebalix correctly

Rebalix separates declared data from measured values. The cost published by the issuer keeps its source and documentary date; the tracking difference, when available, is not presented as a synonym of the TER.

Declared costs: from the TER to the PRIIPs perimeter. In the snapshot of 2 September 2026, the live register holds 3,778 classes. The TER is available for 3,771 classes, 99.8%; transaction costs taken from the KID are available for 3,211 classes, 85.0%, together with the documentary date.

In the public screener Rebalix shows both the TER bands and those of the declared overall cost, calculated as the sum of TER and PRIIPs transaction costs and rounded to two decimals. It is a broader comparison, not a promise of total cost: it does not include the expenses that depend on the listing, the broker or the investor’s tax position.

Tracking difference: calculated, declared or withheld. For 1,973 classes Rebalix calculates a tracking difference over the rolling twelve months with the convention used in this chapter: fund return minus index return. The calculation starts from explicit total return series when available; for accumulation classes it can use the NAV, while for distribution classes it rebuilds the total return by reinvesting the archived distributions.

The figure passes automatic checks before being written. If the gap exceeds ±3 percentage points, it is treated as a possible semantic error — for example a non-comparable index or currency — and discarded. Three control cases also prevent process changes from producing results incompatible with already verified examples.

For 2,136 classes a tracking difference declared by the issuer is also available, drawn from four documentary families: SPDR, Amundi, iShares and Xtrackers. On the fund page, if the calculation is missing, the declared value can be shown as a fallback with a label making its provenance explicit.

For active ETFs and leveraged, inverse or VIX-linked products the annual comparison is structurally excluded: in those cases the gap can measure discretion or compounding, not the quality of an ordinary passive replication. The criterion is stated in the methodology.

When the dash is a data point

For 471 classes Rebalix keeps a note that prevents the publication of an ambiguous figure. The main reasons are:

• 216 distribution classes with neither a total return series nor sufficient distributions in the archive: using the price alone would understate the fund;
• 70 cases with a stalled index;
• 41 calculations discarded because the gap exceeds 3 percentage points;
• around 34 hedged classes where the feed provides an unhedged index: the difference would mostly measure the currency;
• cases where the issuer changed the index series and the tracking difference is suspended until the comparison is consistent.

When the series are not comparable, the absence of the number is more informative than artificial precision. Rebalix does not automatically turn the TER into a tracking difference, and does not reconstruct a certain cause from a simple observed gap.

Three real cases

The page must allow different questions to be answered: which recurring cost does this class declare? On which document, and at which date? Is there a tracking difference declared by the issuer? Is there a tracking difference calculated from comparable series? Which period, benchmark and convention were used? Does a methodological note prevent publishing the figure without ambiguity?

The same discipline applies to costs: declared figure and measured behaviour remain distinct. The first describes the documentary perimeter; the second observes the result of the replication.

A total-cost checklist

• Identity: do the compared products really replicate the same index and the same version?
• Class: are currency, distribution policy, hedging and domicile consistent?
• Document: which cost indicator is shown, and at which date?
• Perimeter: what does that number include, and what does it exclude?
• Replication: what was the tracking difference over comparable periods?
• Stability: is the gap regular, or does it vary a lot?
• Portfolio: can turnover, internal taxation, securities lending or derivatives explain differences?
• Trading: are spread, depth, commissions and currency conversion compatible with amount and frequency?
• Horizon: how much do one-off costs weigh against recurring ones?
• Tax residence: which costs depend on the product, and which on the individual investor?

Ten mistakes to avoid

• “The TER is the total cost.” It is an important indicator, not the whole chain.
• “Tracking difference and TER must coincide.” Replication embeds several forces.
• “A negative tracking difference means bad management.” Benchmark, period and causes come first.
• “Tracking error and tracking difference are synonyms.” One measures variability, the other the return gap.
• “The fund with the lower TER is always cheaper.” Only equivalent products can be compared meaningfully.
• “A commission-free ETF is free.” Fund costs, spreads and possible indirect costs remain.
• “Low volume is an annual cost.” It is market information, not a recurring percentage.
• “The listing’s currency measures the currency risk.” The operational conversion cost and the portfolio’s exposure are distinct.
• “A declared figure and a calculated one are interchangeable.” Source and methodology are part of the data.
• “If the number is missing, you can estimate it from the TER.” The absence may stem from an invalid comparison.