Over their first five years, specialised exchange-traded funds that track narrow themes lost about 30% in risk-adjusted terms. That figure comes from Itzhak Ben-David, Francesco Franzoni, Byungwook Kim and Rabih Moussawi, whose study of the ETF market covers launches from 2000 to 2019. Broad-based ETFs over the same window delivered a four-factor alpha of roughly minus 0.24% a year, which is about what they charge. Specialised ones delivered minus 3.24% a year across their whole lives, and minus 6% a year in the five years right after launch.
The size of the loss isn't the interesting part. The timing is. These funds don't lag mainly because they're expensive, though they are: the median specialised ETF charged 58 basis points against 35 for a broad-based one. They lag because of when they arrive. The authors show the underlying stocks are already richly priced on launch day, and the run-up in the index before launch reverses afterwards.
Launches follow attention, and attention peaks
Ben-David and his co-authors split their sample by what happened before each fund existed. ETFs launched after the strongest pre-launch returns and the most positive media sentiment did worst, with a monthly four-factor alpha of minus 0.53%. Those launched after weaker pre-launch returns and cooler coverage lost only 0.20% a month. Same structure, similar fees, very different outcomes. The separator was how hot the theme was at the moment the money showed up.
Two further details sharpen the picture. By the end of 2019 specialised ETFs held 18% of industry assets but generated about 36% of industry fee revenue, so the incentive to keep launching them is obvious. And 30% of the specialised ETFs in the sample had been liquidated by the end of 2019, against 18% of broad-based ones. Products built around a moment tend not to outlive it.
One nuance cuts against the simple story. After the first five years, the risk-adjusted underperformance of specialised ETFs shrinks and becomes statistically indistinguishable from zero. The damage is concentrated in the window when the flows are newest. That's consistent with a valuation that was stretched at launch and then normalised, rather than with a permanently bad product.
The mutual fund version of the same result
The ETF finding is recent. The underlying pattern is old. Andrea Frazzini and Owen Lamont examined US equity mutual fund flows and stock holdings from 1980 to 2003, and used flows as a proxy for retail sentiment towards individual stocks. Stocks held by funds receiving heavy inflows underperformed stocks held by funds seeing outflows by between 36 and 85 basis points a month, or roughly 4% to 10% a year, depending on the flow horizon measured.
Their worked example is 1999. Investors sent $37 billion to Janus funds that year and only $16 billion to Fidelity, despite Fidelity managing three times the assets at the start of the year. That allocation loaded retail portfolios with the technology names Janus held. By 2001 the same investors pulled about $12 billion out of Janus and added $31 billion to Fidelity, after the damage.
Frazzini and Lamont are careful about the cost to actual investors, which is much smaller than the long-short spread. They put it at seven to nine basis points a month — enough to cut Sharpe ratios by 11% to 17%. Their comparison is the useful one: at the time, the average mutual fund expense ratio of about 1% a year worked out to eight basis points a month. Reallocation decisions cost roughly what the entire fund management industry cost. We've written separately about how fund fees compound over a thirty-year horizon, and the two drags are the same order of magnitude.
Simulating the chase
Flow data tells you what investors did. A simulation tells you what a rule would have produced. Vanguard built one in July 2014 using 3,568 active US equity funds over the ten years to December 2013. The chasing rule bought funds with above-median three-year returns, sold any fund that fell below median, and reinvested in that year's top-20 performers in the same style box. The buy-and-hold rule bought a fund and sold only if it closed. The exercise generated more than 40 million return paths.
Buy-and-hold produced the higher median return in all nine Morningstar style boxes. In large blend it was 6.8% a year against 4.5% for chasing. In mid-cap blend the gap was widest, 8.9% against 4.9%. The narrowest was mid-cap value, 9.2% against 7.6%. Across the grid the shortfall from chasing ran from 1.6 to 4.0 percentage points a year.
That's a mechanical rule, not a survey of real behaviour, and the rule is deliberately aggressive. But it isolates one thing cleanly: selecting on recent three-year performance and acting on it destroyed return in every style box tested, over a decade that included both a crash and a long recovery.
What the aggregate flow data shows
Morningstar's annual Mind the Gap study measures the same idea from the other end. It compares the dollar-weighted return of more than 25,000 US funds and ETFs with their time-weighted total return. For the ten years to December 2024 the average dollar earned 7.0% a year while the average fund earned 8.2%, a gap of 1.2 percentage points, or roughly 15% of the total return on offer.
The distribution across categories is where performance chasing shows up. Sector equity funds had the widest gap at 1.5 points, with the average dollar earning 7.0% against the funds' 8.5%. Allocation funds had the narrowest at 0.1 points, capturing nearly 97% of what the funds produced. International equity lost 1.1 points, US equity 0.6.
Morningstar added a cash-flow-volatility cut in 2025 that maps the effect directly. Sorting funds by how erratic their flows were, the quintile with the steadiest flows had a 0.8-point gap and the quintile with the most erratic flows had a 1.8-point gap. Tracking error ran the same way: the funds hugging their benchmark showed a 0.9-point gap, the funds furthest from it 1.6 points. More deviation and more transacting, wider gap.
The 2023 edition put a sharper number on the sector-fund problem. Over the decade to December 2022, sector equity investors gave up more than four percentage points a year to poorly timed flows. Morningstar's own comment was that sector funds are particularly prone to performance chasing, with investors piling in after a strong showing and leaving when the theme cools.
The strongest counter-argument
Critics have gone after exactly this evidence, and they have a case. In May 2026 the Financial Analysts Journal published a paper by Jon Fulkerson, Bradford Jordan, Timothy Riley and Qing Yan with a blunt title: bad timing does not cost investors 15% of their funds' returns. Working with the same sample Morningstar used, they conclude that poor timing by mutual fund investors costs 0.10% a year, not 1.2%.
The objection is about what dollar-weighting measures. A time-weighted return assumes a lump sum invested on day one and left alone. Real money arrives gradually. In a market that rises more often than it falls, money that arrives later simply earns less of the period's return, regardless of anyone's judgement. The internal rate of return records that as a shortfall. It's arithmetic, not error.
Morningstar has tested this itself and the result is awkward for the behavioural reading. In the 2023 study the firm simulated an investor contributing equal monthly amounts to each category group — a schedule with no discretion in it at all. The gaps under that hypothetical were negative for every single category group, and wider than investors' actual gaps in every case except nontraditional equity. A perfectly disciplined saver would have shown a bigger gap than real investors did.
The 2023 active-versus-passive result points the same way. Index funds had wider gaps than active funds in every category group, and the average dollar in active US stock funds beat the average dollar in passive ones, even though passive funds earned the higher total return. Nobody thinks index investors are worse market timers than active ones. The explanation Morningstar gave was asset growth: money moved en masse into passives during the decade, so a large share of those assets simply weren't present for the earlier, more profitable stretches.
Morningstar's 2025 report says as much in its own introduction. It warns that even sensible habits like investing part of every paycheque or rebalancing on schedule will open a gap, and that the findings shouldn't be read as a parable of dumb money. The study also notes the truism that for the market as a whole there can be no gap, because one investor's shortfall is another's surplus. If you want the fuller treatment of that measurement question, we've covered what the behaviour gap does and doesn't measure separately.
Why the launch evidence survives the objection
The measurement critique lands hard on aggregate gap statistics. It lands much more lightly on the ETF and mutual fund flow studies, because those don't rely on dollar-weighting at all. Ben-David and colleagues form calendar-time portfolios and run factor regressions on fund returns. Frazzini and Lamont sort stocks on flow-based ownership and measure subsequent returns. Vanguard's simulation compares two rules on the same time-weighted return series. None of these can be explained by money arriving late in a rising market.
That's the distinction worth holding on to. Whether the average investor loses 1.2 points a year or 0.10 is genuinely contested. Whether assets bought at the peak of a theme's popularity subsequently disappoint is much better established, and it's established with methods the arithmetic objection doesn't touch.
Persistence data explains why. S&P Dow Jones Indices tracks whether good funds stay good. As of December 2024, none of the top-quartile large-cap US equity funds from 2022 held that rank through the next two years, against 6.25% expected under random chance. Only 9% of above-median large-cap funds stayed above median in each of the two following years, versus about 25% under randomness. Over five consecutive years, just 2.4% of large-cap funds stayed in the top half. Buying last period's winner is buying something the data says rarely repeats — a point that sits alongside the SPIVA record on active funds against their benchmarks.
What would change the conclusion
Three things would.
First, a different sample window. Chris Brightman, Feifei Li and Xi Liu at Research Affiliates studied ETFs launched between 1993 and 2011 and found post-launch returns were virtually flat. Ben-David's team replicated that and traced the difference to two choices: Brightman and co-authors required funds to survive 36 months, which drops the worst cases, and their sample ended before the thematic boom. Change the window and the effect shrinks. A study starting in 2020 would cover a very different set of launches.
Second, the fate of a fund category can swamp the flow effect. Morningstar reports that of 1,259 alternative funds in its 2025 study, 711 were later liquidated. Including or excluding those dead funds moves the alternative-category gap from 0.1 points to 0.8. Some of the smaller cells are noisy in the same way: the fourth fee quintile of alternative funds shows a 5.7-point gap, which reflects a small pool of assets more than a behavioural pattern.
Third, if flows themselves stopped predicting anything, the mechanism would be gone. Frazzini and Lamont argue that the dumb money effect works partly because companies issue shares into strong demand, increasing supply exactly when sentiment is high. If issuance stopped responding to demand, or if the value effect that the dumb money result is tied to disappeared, the relationship would weaken. Their own data shows no reliable effect at horizons shorter than three months, so this isn't a short-horizon trading signal in any case.
What the evidence supports is narrower than the headlines suggest. Money arriving into a strategy after a strong run has, on average, earned less than the strategy did. The clearest version of that is at product launch, where the loss is roughly 6% a year for five years and then stops. The aggregate investor-return gap is real but its size is disputed, and a large share of it is the arithmetic of contributions rather than a record of panic. Pre-commitment is the practical response either way, which is why an investment policy statement written before the decision does more work than any flow chart can.