Multi-Factor Funds: The Tilt Is Smaller Than the Label

Key takeaways

• Regressed on the Fama-French five factors plus momentum, five well-known US multi-factor funds showed value loadings between -0.02 and 0.327, against 1.0 for the academic factor itself.
• Vanguard's fund carried the largest tilt of the five, at 0.455 on size and 0.327 on value, with t-statistics of 10.86 and 9.03 across 101 monthly observations.
• After iShares moved its fund onto a STOXX index in June 2022, its measured loadings fell to -0.075 on size and 0.053 on value, and R-squared against the six factors reached 0.991.
• STOXX caps that index at 100 basis points of ex-ante tracking error and a predicted beta of 0.98 to 1.02, which bounds the tilt before a single stock is selected.
• Priced at the 1963 to 2026 average premia, the measured tilts are worth 0.10% to 3.62% a year gross. Priced at the premia the funds actually lived through, they were worth -0.53% to 0.59%.

If you buy a fund with "multifactor" in its name, how much factor exposure are you getting? For most of the funds sold under that label, a fraction of a unit. I pulled the monthly returns of five well-known US multi-factor funds and regressed them on the Fama-French five factors plus momentum, using Kenneth French's data library, which currently runs through July 2026. The value loadings came out between -0.02 and 0.327. The size loadings ran from 0.007 to 0.455.

A loading of 1.0 would mean the fund moves point for point with the academic factor. None of these is close to that, and none of them claims to be in its filings. The finding isn't mis-selling. It's about scale. The tilt inside a mainstream multifactor product is typically a tenth to a third of the thing the research papers measure, so the expected reward scales down by the same fraction.

How to read a factor loading

A factor regression asks one question. In a month when the market beat cash, cheap stocks beat expensive ones, and small beat large, how much of the fund's return does each of those three moves explain? The coefficient on each factor is the loading. A value loading of 0.327, which is what the Vanguard fund shows, means it behaved as though roughly a third of the money sat in the long-short value portfolio Fama and French build, with the rest tracking the market.

Those portfolios are long and short at once. French's library defines the profitability factor as "the average return on the two robust operating profitability portfolios minus the average return on the two weak operating profitability portfolios", and the others follow the same pattern. No fund you can buy holds that. A long-only fund can lean toward the long leg; it cannot hold the short leg at all. That constraint alone caps how much factor exposure a long-only vehicle can carry, before anyone writes an index rule.

The sample runs from each fund's first full month to July 2026, the last month in French's file. Returns come from monthly dividend-adjusted prices, converted to excess returns over the one-month Treasury bill rate in the same file.

What five multi-factor funds actually load on

Here are the six-factor loadings, with the iShares Core S&P 500 ETF in the last row as a control. It targets no factor at all, so its numbers are the baseline a tilt has to beat.

Vanguard's is the outlier, and it's the only one of the five that is actively managed rather than index-tracking. Its prospectus, dated 27 March 2026, says the fund "employs an active management approach" using "a quantitative model". The size loading of 0.455 and value loading of 0.327 are the largest in the table by a distance, and both are strongly measured, at t-statistics of 10.86 and 9.03.

At the other end, GSLC's value loading is -0.020 and its size loading is -0.054, both effectively nothing, while its profitability loading of 0.104 is the one real tilt. Its prospectus, dated 29 December 2025, describes an index built to give exposure to "value", "momentum", "quality" and "low volatility". Three of those four have a counterpart in the Fama-French model, and all three come out under 0.11. JPUS is the most broadly tilted of the index trackers, at 0.162 on size, 0.153 on value and 0.265 on profitability. OMFL's only meaningful loading is 0.324 on the investment factor, and its momentum loading is negative at -0.114, with a t-statistic of -2.11, which is an odd result for a fund whose index is built to rotate between factor exposures. The investment factor OMFL loads on is one of the two Fama and French added in 2015, and on the test portfolios it was built for it cut average pricing errors by 7.9 basis points a month.

Look at the control row before judging any of them. The S&P 500 itself carries a value loading of 0.017 and a profitability loading of 0.053. Subtract that baseline and LRGF's value tilt against the index is 0.073 and GSLC's is negative, while JPUS's is 0.136 and VFMF's is 0.310.

The index rulebook caps the tilt before a single stock is chosen

Why so small? For the index trackers the answer is written into the methodology, and you can read it. The STOXX Index Methodology Guide describes the STOXX Equity Factor family, which LRGF tracks, as "constructed by maximizing the index exposure to a multi-factor alpha signal while satisfying a set of constraints intended to closely track their parent indices". The constraints are the operative half of that sentence.

STOXX lists them in a table. Predicted beta must sit "Between 0.98 and 1.02". Active sector exposures must stay "Within 2% of Parent Index". Ex-ante tracking error has a "Maximum 100 bps". Turnover is capped at "Maximum 5% one way per quarter". The active signal exposures themselves are bounded: "0.2 < Quality < 0.5", "0.2 < Momentum < 0.4", "0 < Value < 0.4". A portfolio held within 100 basis points of tracking error against the STOXX USA 900 cannot deviate far enough from it to produce a large Fama-French loading. The optimiser maximises the tilt, but it does so inside a box that is deliberately narrow.

The holdings show the same thing. The iShares factsheet for LRGF, dated 30 June 2026, lists NVIDIA at 7.08% and Apple at 6.15% of the fund, a price-to-earnings ratio of 28.18x and a price-to-book ratio of 5.70x across 292 holdings. The factsheet for the S&P 500 tracker on the same date lists NVIDIA at 7.51%, Apple at 6.58%, a P/E of 30.17x and a price-to-book of 5.59x across 504 holdings. The fund built for "value" among other things is 6.6% cheaper on earnings than the index and slightly more expensive on book value. Its three-year standard deviation is 13.19% against the index's 13.05%.

One fund changed index and its factor exposure went to roughly zero

LRGF gives an unusually clean natural experiment. A supplement filed with the SEC on 31 March 2022 announced that the fund would be renamed from the iShares MSCI USA Multifactor ETF to the iShares U.S. Equity Factor ETF, and that its underlying index would change from the MSCI USA Diversified Multiple-Factor Index to the STOXX U.S. Equity Factor Index, "expected to be implemented on or around June 1, 2022". Splitting the regression at that date gives two different funds.

Under the MSCI index, from May 2015 to May 2022, the loadings were 0.118 on size, 0.089 on value and 0.185 on profitability, with an R-squared of 0.964. Under the STOXX index, from June 2022 to July 2026, they were -0.075 on size, 0.053 on value and 0.001 on profitability, and R-squared rose to 0.991. Market beta went from 0.923 to 0.988. The profitability tilt, the largest one the fund had, disappeared entirely.

The fund still describes itself as tracking an index "designed to maximize exposure to five target factors: momentum, quality, value, low volatility and size", in a summary prospectus dated 28 November 2025. Both statements are true at once. The index does maximise, and the maximum available inside a 100 basis point tracking-error budget is close to nothing. A word like "maximize" tells you the direction of the optimisation, not its magnitude. Only the regression tells you the magnitude, and the second sample is only 50 months long.

Factor premia by decade, and what a 0.1 loading is worth

A loading is half the arithmetic. The other half is what the factor paid. Averaging the monthly series in French's file and annualising gives the table below, in percent a year. It's the record of six long-short portfolios, gross of every cost of holding them.

Multiply each fund's loadings by the full-sample premia and you get the chart accompanying this piece, the estimated style contribution in percent a year. VFMF comes out at 3.62% and JPUS at 2.63%, because they carry real tilts. LRGF lands at 0.56%, GSLC at 0.49% and OMFL at 0.16%. The S&P 500 control, which targets nothing, scores 0.10%. Those figures are gross: LRGF charges 0.08% a year, GSLC 0.09%, VFMF and JPUS 0.18% each and OMFL 0.29%, against 0.03% for the index tracker.

So a 0.09 value loading, at the value premium's 63-year average of 3.59% a year, is worth about 0.32% a year before fees, and the fee difference against a plain index fund eats roughly a sixth of that. That is the real size of the decision, and it is a long way from the headline premia the factor literature reports.

Historic premia are not forecasts

The decade rows exist to make that point unmissable. Value paid 7.84% a year in the 1970s and lost 2.35% a year in the 2010s. Size paid 9.57% a year in the stub period from July 1963 to December 1969 and 7.06% a year in the 2000s, then lost money in four of the other five decades. Momentum paid 13.55% a year in the 1990s and 1.01% a year in the 2000s. These are realised averages over finite samples, not expected returns, and nothing here forecasts the next decade.

Recent history makes the point sharper. Over the exact months each fund has existed, the premia were mostly absent. From October 2015 to July 2026, size averaged -1.58% a year, value 0.32% and momentum 1.57%, while the market paid 13.26%. Repricing each fund's loadings at the premia of its own sample rather than at the long-run averages, VFMF's style contribution falls from 3.62% to -0.03% a year, JPUS's from 2.63% to 0.59%, and OMFL's from 0.16% to -0.53%. The exposure was real and the premium wasn't there to collect. The same gap between backtest and live record shows up directly in factor ETF returns, and in what happened to the value premium after it was published.

The case for small loadings

The strongest objection to everything above is that the small loadings are the product working, not failing. A fund that held a 1.0 value loading would look nothing like the index, and the tracking error would be enormous. Most investors who buy a core equity fund have a job for it, and that job is to be the equity allocation. A tilt that can trail the S&P 500 badly for a year gets sold at the bottom, which converts a paper premium into a realised loss. STOXX's 100 basis point tracking error cap is a behavioural constraint as much as a risk constraint.

There's a cost argument too. The academic factors are gross of trading costs, and their turnover is high, which is exactly the problem documented for the momentum premium. JPUS reported portfolio turnover of 25% of average portfolio value in its most recent fiscal year, and OMFL 50%. A long-short implementation of the same signals at full strength would trade far more than that, pay borrowing costs on the short leg, and charge a multiple of the 0.18% these funds cost. Whether the extra exposure survives the extra cost is an empirical question, and a regression on fund returns doesn't answer it.

The honest version of the argument is that the funds do roughly what their rulebooks say, and the rulebooks are conservative on purpose. What the marketing language obscures is the size of the effect. "Maximize exposure" and "a 0.053 value loading after the index switch" both describe LRGF's STOXX index accurately.

What this regression cannot tell you

Several things. The samples are short: 101 to 135 months, one country, and a stretch in which the market paid 12.08% to 13.26% a year and the size premium was negative. A period that favoured large growth stocks is a hard period in which to measure a small tilt, and the standard errors reflect that. The alpha estimates are the weakest numbers in the exercise: LRGF's is -0.96% a year with a t-statistic of -1.12, VFMF's -0.23% with a t-statistic of -0.18, and OMFL's +0.92% with a t-statistic of 0.39. None of those is distinguishable from zero.

The six-factor model is itself a choice. Loadings are defined relative to it, so a fund tilting toward a characteristic the model doesn't span, low volatility being the obvious one here, shows up with small loadings on everything and a lower R-squared. OMFL's R-squared of 0.863 and JPUS's of 0.906 both hint at variation the model isn't capturing. Monthly data over eight to eleven years also cannot detect a loading that moves around, which is precisely what OMFL's index is designed to do.

Fund returns come from monthly dividend-adjusted prices rather than official NAV series, so they carry any premium or discount to NAV at each month end. And the LRGF split assumes the index change took effect on 1 June 2022, the date the SEC filing gives as expected, not a confirmed completion date.

What would change the conclusion

Three things would. If a fund published loadings this large against its own stated benchmark and the tracking-error constraint were relaxed, the tilts would widen and the arithmetic in the chart would scale up with them, which is what a fund like VFMF already shows. If the size and value premia returned to anything like their 1970s or 2000s levels, when value paid 7.84% and 7.95% a year, a 0.327 loading would be worth several times what it is worth now, and the gap between the funds in the table would matter far more than the gap in their fees. And if a longer post-2022 sample for LRGF showed the loadings recovering, the natural experiment above would turn out to be a 50-month artefact rather than a description of the index.

Until one of those happens, the number that settles the question about a multifactor product isn't in the brochure. It's the loading, and you can compute it from the fund's monthly returns and French's file, both of which are free. LedgerTouch runs the same calculation across a whole account rather than one fund. The comparison that matters is against what a plain index already delivers, which is where equal weight factor exposure makes the same point from the other direction: the cheapest source of a size tilt may turn out to be a rule about weighting rather than a fund about factors.