Key takeaways
- The FCA regressed UK large-cap active fund performance on charges over 2006 to 2015. Net of fees the coefficient was -1.1864 and significant; gross of fees it was -0.2406 and not.
- Global large-cap funds gave the same answer: -1.2320 net of fees at a p-value of 0.007, against -0.3106 gross of fees at 0.501.
- Morningstar ranked US equity funds by expense ratio at the end of 2010. Over the next five years the cheapest fifth, averaging 0.65%, had a 61.82% success ratio; the priciest, at 2.20%, had 20.44%.
- That gap widened with the horizon, from 39.03 points over three years to 40.16 over four and 41.38 over five.
- The asset-weighted expense ratio across all US funds was 0.32% in 2025, against 0.80% in 2006, which is a shrinking variable to predict with.
Fees predict fund returns after costs, and manager skill not at all
You're looking at two funds in the same category. One charges 0.65% a year, the other 2.20%. Those were the average expense ratios of the cheapest and priciest fifths of US equity funds in Morningstar's data at the end of 2010. Does the expensive one earn the difference back?
On the evidence, no. The shape of the failure is the interesting part.
When the Financial Conduct Authority regressed the performance of UK active equity funds on their charges over 2006 to 2015, it ran the test twice. Measured before fees came out, the relationship was flat and statistically invisible. Measured after fees came out, it was negative, significant, and close to one point of return lost per point of charge.
That pairing has a plain reading. Expensive managers weren't worse at picking shares. They were charging more for the same shares, and the charge came off the top.
The FCA ran the regression twice, and only one version found anything
The regulator's June 2017 annex is titled Assessing the relationship between the price and performance of retail equity funds in the UK. It takes annual returns in excess of benchmark as the dependent variable and the ongoing charges figure, the OCF, as the independent one. The OCF is the headline annual cost of a fund, and here it's the "clean" version. The FCA removed "the commission (i.e. the distribution fee) received by the platform from the asset manager" so that pre-RDR bundled share classes could sit alongside post-RDR clean ones.
Gross of fees, across three equity categories, nothing came out. The UK large-cap coefficient was -0.2406 with a p-value of 0.687. Global large cap was -0.3106 at 0.501. Europe large cap was actually positive, 0.4075, at 0.642. None of these are distinguishable from zero.
Net of fees, two of the three turned negative and significant. UK large cap came in at -1.1864, p-value 0.044, on 905 annual observations from 151 funds. Global large cap came in at -1.2320, p-value 0.007. Europe large cap stayed insignificant at -0.4522.
The FCA's own summary is careful: "there has been no clear linear relationship between the gross performance of retail active equity funds in the UK and the level of on-going charges figure in the period examined." On net returns it says only that it found "some evidence that the impact of higher fees for more expensive active funds on average has meant that they underperformed cheaper active funds."
Read those two tables together and you get an arithmetic result rather than a behavioural one. A coefficient near -1.2 on net returns, sitting on top of a coefficient near zero on gross returns, is what pure subtraction looks like in a regression output.
Morningstar's quintiles say something stronger than the FCA's regression
Russel Kinnel's Predictive Power of Fees, published by Morningstar in May 2016, sorted funds into expense ratio quintiles inside each category and then asked what happened next. Its opening line is not hedged: "The expense ratio is the most proven predictor of future fund returns."
For US equity funds over the five years from January 2011 to December 2015, the success ratios ran 61.82%, 47.87%, 39.26%, 29.63% and 20.44%, cheapest quintile to priciest. The average expense ratios of those quintiles, measured at the end of 2010, were 0.65%, 1.01%, 1.26%, 1.59% and 2.20%. The chart above plots that first row.
A success ratio is stricter than a return comparison. It asks what share of funds both survived the period and beat their category, so a fund that was merged away counts as a failure rather than vanishing from the sample. Kinnel notes that about 20% of the funds were merged away, "making 40% the average success ratio point."
The plain returns move the same way. Those five quintiles returned 10.91% a year down to 8.88%. Fit a line through the five quintile averages and the slope is about -1.30: each extra percentage point of expense ratio came with roughly 1.30 points a year less return. That's steeper than the FCA's -1.1864, and steeper than one-for-one.
A persistence table by horizon, and why the gap gets wider
Kinnel ran the same test over three, four and five years. Here are the US equity success ratios by expense ratio quintile, all measured to the end of 2015.
| Horizon | Cheapest | 2nd | 3rd | 4th | Priciest | Cheapest less priciest |
|---|---|---|---|---|---|---|
| 3 years, from 2013 | 67.88% | 55.17% | 46.98% | 39.12% | 28.85% | 39.03 points |
| 4 years, from 2012 | 64.69% | 52.64% | 42.09% | 34.74% | 24.53% | 40.16 points |
| 5 years, from 2011 | 61.82% | 47.87% | 39.26% | 29.63% | 20.44% | 41.38 points |
Two things move as the window lengthens. Every quintile's success ratio falls, and the distance between the ends grows, from 39.03 points to 41.38. Kinnel's explanation for why the levels sit where they do is attrition: "If no funds were merged away, then the overall success rates would average something close to 50%."
The ordering never breaks. In every row, each quintile beats the one above it in cost. That monotonicity is the part the FCA's UK sample never produced, and it's worth asking why.
Why the two studies disagree, and which one you're standing in
Three differences do most of the work.
The first is dispersion. A regressor with no variance can't predict anything. The FCA found that "the majority of funds cluster within a narrow price range but often deliver very different levels of return." In 2016 it put the average OCF for actively managed UK equity funds at 0.90% and for passive ones at 0.15%. Morningstar's US quintiles spanned 0.65% to 2.20%, a spread of 1.55 percentage points. There was simply more fee variation to work with. The FCA also narrowed its own range on purpose: it says "we have not included index-tracker funds", because a tracker's gross return is set by the index rather than by a manager.
The second is survivorship. Kinnel's success ratio deliberately counts the dead. The FCA's regression uses funds present in its dataset, and it says so plainly about one of its charts: "This sample of funds will be likely to have on average outperformed their peers as the constraints imposed have introduced survivorship bias." Kinnel's reason for counting them is that "higher-cost funds are much more likely to fail and be merged away", so excluding them flatters the expensive end.
The third is what's being measured. The FCA regressed excess return over benchmark. Kinnel measured a fund against its category peers, and combined it with survival. Those are different questions and they don't have to agree.
None of that makes either result wrong. It means how well fees predict fund returns depends on how much fee variation exists where you're shopping, and on whether you count the funds that disappeared.
One-for-one is the finding, not a boring by-product
William Sharpe's 1991 note The Arithmetic of Active Management sets out two propositions that "depend only on the laws of addition, subtraction, multiplication and division." Before costs, the return on the average actively managed dollar equals the return on the average passively managed dollar. After costs, it's less.
Feed that into a regression and it predicts exactly what the FCA found. If fees buy nothing, the coefficient on gross performance sits at zero and the coefficient on net performance sits at -1. The FCA's -1.1864 and -1.2320 are a little steeper than that, and Morningstar's -1.30 steeper again, but the story is a subtraction with a small amount of extra drag on top.
The FCA measured the drag directly. Active funds sold in the UK "on average, outperformed benchmarks before charges were deducted, but underperformed benchmarks after charges on an annualised basis by around 60bps." Gross skill existed. It just didn't survive the invoice. That is the same arithmetic set out in our piece on the SPIVA scorecard, arriving from a different direction.
The strongest objection: the fee may be the price of skill, not the absence of it
Jonathan Berk and Richard Green's 2002 NBER paper Mutual Fund Flows and Performance in Rational Markets makes the serious case against reading any of this as a verdict on managers.
Their model has money chasing performance and active management getting harder at scale. A skilled manager attracts inflows until the extra size erodes the edge, and the fee captures what's left. In equilibrium, "investments with active managers do not outperform passive benchmarks because of the competitive market for capital provision, combined with decreasing returns to scale." Net alpha goes to zero for the skilled and the unskilled alike. Calibrating to fund flows and survival rates, they find the data "consistent with the vast majority (80%) of active managers having at least enough skill to make back their fees."
That reframes the FCA's flat gross line. A gross coefficient of zero isn't evidence that expensive managers are bad. Under Berk and Green it's the predicted result of a market where fees track skill closely enough to cancel it out.
The objection lands on interpretation, not on outcome. Whether the fee is a rent or a fair price for genuine ability, the investor's net return is what's left after it. On that number the two accounts agree.
Where fees beat past performance is that you know the fee in advance
Vanguard's March 2022 paper The case for low-cost index-fund investing ranked active equity funds available in the UK by excess return over the five years to December 2016, then followed them for the next five years to December 2021. Of the 539 funds in the top quintile, 26.16% were still in the top quintile five years later, and 18.18% had been merged or liquidated.
The bottom of the table is more striking. Of the 540 funds in the lowest quintile, 20.78% stayed in the lowest quintile and 36.18% were merged or liquidated. Vanguard's own reading: "persistence was actually stronger among the underperforming managers than those that outperformed."
The FCA reported the same asymmetry. It found "little evidence of persistence in outperformance in the academic literature, but there is some evidence of persistent underperformance", and added that "our analysis provides some evidence of persistence of relatively poor performance."
Set that beside the fee. Last decade's top-quartile return tells you very little about the next decade. The expense ratio is printed on the factsheet, applies from the day you buy, and is one of the few inputs to a net return you can read before committing money. That asymmetry, rather than any claim about manager quality, is why fund charges and performance keep getting studied together. Our note on performance chasing covers what happens to money that uses past returns as the signal instead.
Where the fee signal is still doing work in 2026
The regressions above stop in 2015. Fees predict fund returns in fresher data too, and Morningstar's Active/Passive Barometer is where that shows up. Its edition for the twelve months to June 2026 covered 9,226 funds holding about USD 29 trillion, roughly 67% of the US fund market. Just over 40% of active funds survived and beat their average passive peer over that year, and across all categories "an investor selecting an active manager at random would face a 60% probability of them underperforming their average passive peer."
The fee cut still separates the field. "Across 16 of the 20 categories, the choice of an active manager in the lowest fee quintile improved the chances of randomly selecting an active fund that outperforms." In US large blend and mid-cap value the cheapest funds' success rates are "more than double those of their most expensive active peers." In diversified emerging markets, US real estate and intermediate core bond, the cheapest quintile clears 50%.
The same report puts the 10-year success rate for active US large-growth funds at 5%, with an excess-return distribution skewed to the downside. Cost is not the only thing that varies by category, and in some categories it barely rescues the odds.
What this evidence cannot tell you
Start with the regressions themselves. The line I fitted through Morningstar's US equity quintiles runs through five points, not five thousand funds. Quintile averages smooth away the variation that makes fund selection hard, so the fit is far tighter than any individual investor's experience. The FCA's fund-level regressions are the honest version, and they're messier. Two of its three equity categories cleared significance on net excess returns and Europe did not, and the regulator notes that "our results vary according to different sampling methodologies."
The windows are short and lucky. Kinnel's tables cover 2011 to 2015, one stretch of a US bull market. The FCA's cover 2006 to 2015 in three equity categories. Neither is a forecast, and neither is world data.
The sources aren't neutral either. Vanguard sells index funds. Morningstar sells fund ratings and runs its own funds, and its fee study states plainly that "Morningstar research has demonstrated that fees are a reliable predictor of future returns." The FCA is the disinterested party here, and it's also the one that found the weakest relationship.
And the variable is shrinking. Morningstar's 2026 fee study puts the asset-weighted expense ratio across US mutual funds and ETFs at 0.32% in 2025, down from 0.34% in 2024 and 0.80% in 2006. Passive funds sit at 0.10%. Over 15 years the cheapest decile halved its fees, the median fund's fee "came down 31% during that time, and the priciest decile declined by just 18%." A predictor works on dispersion, and dispersion at the cheap end is nearly exhausted. Our breakdown of the total cost of owning a fund shows how much of what you actually pay never appears in the expense ratio at all.
What would change the conclusion
If fee dispersion collapses, the predictor loses its input. Fees predict fund returns only for as long as funds charge measurably different fees. Morningstar expects fees at the top of the market to keep falling faster than fees at the bottom, so "the gap between the most expensive and least expensive funds could eventually converge." Two funds charging 0.07% cannot be told apart by price.
If the gross coefficient ever turns positive and significant, the whole reading flips. That would mean expensive managers really do generate more before costs, and the question becomes whether they charge more than they add. None of the seven gross-of-fee coefficients the FCA reports came close, but the sample was three equity categories over ten years.
If the expensive end re-widens, the signal gets sharper again rather than weaker. Morningstar reports that a few high-cost launches pushed the average cost of new mutual funds back above 1% in 2025, "for the first time since 2015". It attributes dearer new ETFs separately, to "the proliferation of complex high-cost strategies making their way into the ETF wrapper". New dispersion at the top would restore the range the old studies had.
The number to watch isn't a study. It's the gap between what your own funds charge and the cheapest comparable option in the same category, because that gap is the entire size of the effect described above. LedgerTouch shows the weighted cost of a portfolio alongside its returns; a factsheet and a calculator get you the same figure. What neither can do is tell you which manager is skilled, and on the evidence above, neither can the price tag.