Index Funds, Active-Investor Selection, and the Re-Sorting of Factor Premia: An Extension of Stambaugh (2014)

Summary

Two jobs a factor premium can pay for

Index mutual funds and ETFs held approximately 54% of US equity assets by the close of 2024, up from less than 5% in 1995 (Figure 1). Does that much indexing starve prices of information? Many people think so. Robert Stambaugh argued the opposite in his 2014 presidential address to the American Finance Association: index funds mostly replace noise traders, so the mispricings that active managers feed on get smaller, and active management shrinks with them.

We carry his argument down to individual factors. On our account a factor premium pays active capital for one of two jobs, either absorbing the overreactions of noise traders or working out firm-specific fundamentals before the market does. Value, measured by HML, leans on the first. A low price-to-book ratio is easy to see and often comes from panic selling. Quality, measured by QMJ, depends more on the second, since spotting durable earnings and modest debt ahead of time takes steady research.

As passive money pushes noise traders out, the absorption premium should shrink a lot. The information premium should shrink a little. In the current version of the paper, quality compresses at roughly one-fifth the rate of value.

A pair of facts from 2010-2024 first drew us in. HML earned an average close to zero. Over the same years QMJ ran a Sharpe ratio above 1.0, and the literature has mostly treated the two as separate puzzles, where we think they may share a cause.

What the simulator shows

Much of the evidence comes from a calibrated simulator of the monthly factor panel for 1963-2024. A caveat belongs up front: its true slopes are set to the real-data estimates, -3.2 for value and -0.7 for quality, so the simulator tests whether the estimators recover what was put in. Here they do. Regressing monthly factor returns on the growth of the passive share gives an annualised HML slope of -3.228 and a QMJ slope of -0.816 (Table 2), and Figure 3 plots the same relation year by year.

Both series show a structural break in a Quandt-Andrews test, HML in 1993 and QMJ in 1998. For HML the shift in the mean is roughly three times larger. Figure 2 tracks trailing ten-year Sharpe ratios: HML's turns sharply negative in the 2010s, and QMJ's softens but stays positive.

Next comes a cross-section of 8,000 simulated stocks sorted into quintiles by passive ownership (Table 4, Figure 5). Value earns +6.79 per z-score in the least indexed quintile and +3.28 in the most. Quality barely moves, between +4.70 and +5.51.

Over 200 simulator seeds the HML slope comes out negative every time, 100.0% of runs. For QMJ the share is 77.5% (Figure 7). A smaller effect is harder to detect in one sample of this length, so the lower share reflects limited power in a single sample.

Checking against real data

Version two adds a replication on Ken French factors and 13 factor ETFs. With the ICI measure of passive share, the HML slope is -3.18. A quality proxy built from French's profitability and investment factors gives -0.57. Is the difference more than noise? Across 2,000 paired bootstrap draws the gap between the two slopes is -2.610, with a 95% interval running from -5.042 to -0.192, so zero lies outside it.

Funds give a second view. The Russell 1000 value-minus-growth spread, IWD less IWF, ran at +0.20 a year in 2003-2009, -3.09 in 2010-2019 and -8.40 in 2020-2025. Abroad, where indexing is less widespread, the EAFE counterpart stayed close to flat after 2005: -0.07, against -2.93 for the US spread over the same stretch.

Where the evidence is thin

One of the three passive-share measures does not cooperate. An ETF-based proxy gives an HML slope near zero, which we read as a weak-instrument result, because the proxy is built partly from the same prices it is meant to explain. Early windows have the same problem: before 1990 the passive share hardly moved, and both slopes are unstable.

The paper also carries a leftover. Its closing limitations paragraph still describes the empirical work as synthetic, a sentence written before the real-data section was added. The model has only two trader types; a continuum of skill would map more cleanly to fund-flow data.

None of this contradicts evidence that passive ownership makes individual stock prices less informative. Average skill among remaining active investors can rise while single-stock informativeness falls. If the idea is right, a long drought for value and a steady run for quality are one event seen from two sides.

Who this is for: Researchers and quants studying factor premia, and journalists following the debate over passive share and the so-called death of value.