Vanguard's FTSE 250 UCITS ETF charges 0.10% a year. Over the twelve months to 30 June 2025 it returned 9.96% while the FTSE 250 Index returned 10.15%. The shortfall was 0.19 percentage points, roughly double the headline fee.
The same annual report shows Vanguard's FTSE All-World UCITS ETF charging 0.22% and trailing its index by only 0.04 points. Neither figure is an error. That's what happens when you measure a fund against what it was meant to deliver rather than against its price list.
Four things live in the gap. The spread and price impact a fund pays when it trades. The turnover the index forces on it. The cash it can't keep invested. And the securities-lending revenue that hands some of the money back. The arithmetic of fund fees compounded over 30 years assumes one clean annual charge. This is what that charge leaves out.
The fee and the shortfall are different numbers
Vanguard's Irish-domiciled ETF range publishes fund and index returns side by side in its annual report, so nothing here needs estimating. Five equity funds, year to 30 June 2025:
| Fund | Annual management fee | One-year tracking difference |
|---|---|---|
| FTSE 100 UCITS ETF | 0.09% | -0.13 |
| FTSE 250 UCITS ETF | 0.10% | -0.19 |
| FTSE All-World UCITS ETF | 0.22% | -0.04 |
| FTSE All-World High Dividend Yield UCITS ETF | 0.29% | +0.14 |
| S&P 500 UCITS ETF | 0.07% | +0.16 |
The ranking is close to inverted. One caveat lands straight away: the S&P 500 fund's +0.16 isn't clean. A footnote in the report says the fund's return is adjusted by 15% for US dividend withholding tax while the index return is adjusted by 30%. That's a genuine advantage of the Irish domicile, but it's a tax-treaty effect rather than an efficiency one. Withholding tax is a quiet cost of owning foreign shares, which is a separate question from how much of a portfolio's equity should sit abroad.
Set that fund aside and the pattern still holds. The 0.22% fund tracked more than four times as closely as the 0.10% fund. The 0.29% fund beat its index outright.
Most of a trading cost isn't the spread
Ask what it costs a fund to trade and most people say the bid-ask spread. The measured answer looks different.
A 2009 working paper for the Center for Retirement Research at Boston College reconstructed the trading of the 100 largest US domestic equity mutual funds held in defined-contribution plans, using NYSE quote and trade data for 2004 to 2008. For the middle quintile of funds, price impact accounted for about three-quarters of total trading cost. Brokerage commissions were roughly one-sixth. Bid-ask spreads were about one-twentieth.
Price impact is the market moving against you while you fill. Buy enough of something and you push its price up before you're done. It doesn't show up on any contract note, which is exactly why it goes missing from cost comparisons.
Size mattered in that sample. Ranked by assets, median total trading cost fell from a range of 0.70% to 1.15% in the two smallest quintiles to 0.40% to 0.56% in the three largest. Reported turnover averaged 48% across the five years, and the authors note that understates real trading, because the published turnover ratio takes the smaller of purchases or sales. Trading costs are one of the standing explanations for the pattern in twenty years of SPIVA data on active fund underperformance.
Three caveats. It's a working-paper draft rather than a journal article. The sample is American and actively managed. And it ends in 2008, when spreads and impact were unusually wide.
Not everyone agrees the numbers are that large. Andrea Frazzini, Ronen Israel and Tobias Moskowitz examined $1.7 trillion of live executions from one large institutional manager across 21 developed markets, from August 1998 to June 2016. Their estimate of market impact is just under 9 basis points on average for trades completed within a day. About 85% of it is permanent: only 1.26 basis points reverse over the next 24 hours. They conclude that real trading costs are "an order of magnitude smaller than previous studies suggest".
The contrast they draw is stark. At a trade size of 10% of daily volume, a standard model built on aggregated exchange data implies 223.31 basis points of cost. Their live trades cost 32.34.
Read the fine print before treating that as settled. Their manager trades patiently, using limit orders designed to provide liquidity rather than demand it. Long-short strategies make up 73% of the sample, and large-cap portfolios account for $1.65 trillion of the $1.7 trillion traded. The authors were affiliated with the firm whose data this is. An index fund that has to trade at the close on an index effective date has none of that patience available to it.
Turnover the index chooses for you
An index fund's turnover isn't really a management decision. When a stock joins or leaves the benchmark, every mechanical tracker buys or sells it, on the same day, in the same direction.
Antti Petajisto measured what that habit costs. Studying S&P 500 and Russell 2000 index changes from 1990 to 2005, he found additions rose 8.8% and 4.7% respectively between announcement and effective day, while deletions fell 15.1% and 4.6%. Indexers systematically buy after the move up and sell after the move down.
He named the recurring drag the index turnover cost and put a lower bound on it: 21 to 28 basis points a year for the S&P 500, and 38 to 77 basis points for the Russell 2000. The peak came in 2000, at 65 to 82 basis points for the S&P 500 and 232 to 463 for the Russell 2000, the year the gap between addition and deletion premia reached 32.2% and 37.4%.
Two caveats, both important. Those are explicitly lower bounds. And the sample stops in 2005, with premia already falling after 2000, so nobody should read 38 to 77 basis points as a current figure for a small-cap tracker.
The mechanism is what survives, and it explains why turnover bites harder in some corners than others. Vanguard's own accounts show the spread. Its FTSE 100 fund paid £955,394 in transaction fees and commissions against £4,659,277 of management fees. Its FTSE 250 fund paid £4,484,806 in dealing costs against £2,550,306 of management fees, or 1.8 times the fee.
Honesty about that comparison: most of the FTSE 250 fund's dealing wasn't index-driven. It took in £912.8m of subscriptions and paid out £1,042.1m of redemptions on a fund that ended the year at £2.31bn. Investor flow, not index rules, did most of the trading. Every rebalance is a trade too, which is part of why the choice between annual rebalancing and 5% threshold bands is a cost question as much as a risk one.
Cash drag is real, small and asymmetric
Funds hold cash. Dividends arrive daily and go out quarterly. Subscriptions land before they can be invested. Whatever sits in cash isn't tracking anything.
The cleanest measurement of this is old but exact. Edwin Elton, Martin Gruber, George Comer and Kai Li studied the original S&P 500 tracker, the SPDR trust, from February 1993 to the end of 1998. Its net asset value underperformed the S&P 500 by 28.4 basis points a year. Of that, 18.45 points were the expense ratio. The remaining 9.95 points came from one structural quirk: the trust's legal form required dividends received to sit in a non-interest-bearing account until the quarterly payout.
They cross-checked it. With a dividend yield near 2.2% and index returns averaging about 22.2% a year over that stretch, the arithmetic loss from not reinvesting on receipt works out at roughly 10.2 basis points, close enough to the 9.95 they measured directly.
Then the asymmetry. In the four quarters where the S&P 500 fell, the SPDR beat the index, because holding dividends in cash is the right side of the trade in a falling market. Regressing the return gap on the index's quarterly return produced an R-squared of 0.99. Cash drag isn't a fixed leak. It's the market's return multiplied by the share of assets that missed it.
Three things have shrunk it since. The SPDR expense ratio was cut to 12 basis points shortly after the sample ended. Dividend yields are lower than 2.2%. And accumulating share classes reinvest income rather than parking it. The effect is smaller than it was in 1996, but the arithmetic hasn't changed.
Lending the portfolio out pays some of it back
Index funds lend their holdings to short sellers and keep most of the fee. It's the one line here that runs in the investor's favour.
Vanguard's Irish ETF range earned $13,518,041 of net securities-lending income in the year to 30 June 2025, after $1,134,721 of lending-agent fees. Set against $22,684,814 of transaction fees and commissions across the same funds, lending covered 60% of the explicit dealing bill.
Fund by fund it varies enormously. The FTSE All-World fund earned $5,427,750, or 94% of the $5,752,698 it paid in transaction fees. The FTSE 250 fund earned £309,448 against £4,484,806 of dealing costs, under 7%.
The reason is what borrowers actually want. On 30 June 2025 the FTSE Japan fund had 4.90% of net assets out on loan. The S&P 500 fund had 0.02%. Demand to short large US stocks was close to nil, and the fund with the lowest headline fee in the table therefore had the least to offset it with. That's another reason a cheap fee and a good outcome aren't the same thing, and it interacts with the overlap between broad funds in the same ten mega-caps.
Lending isn't free of risk, and the terms are a design choice. Collateral in this programme is limited to high-quality sovereign debt, and stood at £20,997,815 against £16,068,820 of FTSE 250 stock on loan, about 31% over-collateralised. The report also states that no collateral was re-used during the year. A programme accepting weaker collateral, or reinvesting cash collateral for extra yield, would carry a different risk profile for the same headline revenue.
The strongest objection: tracking difference already contains all of it
Here's the case against everything above. If spreads, impact, index turnover, cash drag and lending revenue all end up in the fund's return, the tracking difference nets them out already. Compare that one number and ignore the components. Anything else is decoration.
It's a good argument and it's mostly right about what tracking difference measures. It's wrong about what tracking difference can see.
Start with the index itself. Petajisto describes the index turnover cost as borne by index funds and by the indexes they track. The benchmark's return is computed as though the constituent change happened at the closing price, so the premium the market pays to buy an addition is inside the index. A fund tracking it perfectly reports a tracking difference of zero and still bore the cost.
Then there's what minimising the number costs. Marshall Blume and Roger Edelen showed that an S&P 500 indexer trading at the opening price the day after an index change was announced, rather than at the close on the effective date, would have added an average of 19.2 basis points a year from 1995 through 2000. The catch is that the standard deviation of that strategy's annual tracking error is 23.9 basis points, against 2.8 basis points for the largest indexer at the time. To keep the reported number tight, indexers give up return. Tracking difference isn't a neutral observation of cost. It's partly a product of being managed to.
Third, the number moves with the window. Vanguard's FTSE 250 fund shows -0.19 over one year and -0.10 annualised over five. Same fund, same index, different answer. And it can be flattered by things unrelated to efficiency, as the withholding-tax footnote on the S&P 500 fund shows.
Fourth, it stops at the fund's net asset value. Your own dealing spread when you buy the ETF, the platform fee on top, and the premium or discount to NAV at the moment you trade all sit outside it.
Where that leaves the objection: as a single summary of what a fund did to your money last year, tracking difference beats the expense ratio by a wide margin, and comparing it is sensible. What it can't tell you is why, whether it repeats, or what it left outside the measurement.
What would change the conclusion
Better disclosure of implicit costs would settle much of this. The Financial Conduct Authority's 2017 policy statement on workplace pension costs requires firms to calculate transaction costs by the slippage method, comparing the execution price with the price when the order reached the market. The FCA adopted it precisely because explicit costs are easy and implicit ones aren't, while conceding there is "no perfect way of calculating implicit transaction costs". If one consistently computed number were published for every fund, none of the reconstruction above would be necessary.
A modern replication of the price-impact work would change the size of these estimates, if not their direction. The two best measurements disagree by an order of magnitude, and both are ageing: the fund-level data stops in 2008, the live-execution data in 2016. Commissions have fallen, market structure has changed, and index funds are a far larger share of trading than they were.
Index turnover would matter less if index rules changed. Petajisto's mechanism depends on every tracker trading the same stock on the same day. Rules that phase changes over several days, or that transition at volume-weighted prices, weaken it directly.
And securities lending could run the other way. It currently offsets a meaningful share of dealing costs. If short interest fell, or if collateral rules tightened, that offset would shrink and tracking differences would widen without a single fund changing anything it does.
The figure worth carrying away is the one from the table. A 0.10% fee produced a 0.19-point shortfall and a 0.22% fee produced a 0.04-point one, in the same year, at the same firm, in funds run by the same people. Headline charges rank price. They don't rank outcomes.