AQAI QuantAI research lab for systematic strategies

Automated analysis

This analysis was drafted by our research engine and has not been checked by a human editor. It may contain errors. It separates the paper’s own results from our tests, and any figures called ours come from our own backtest.

Our automated analysisOur backtest

Frazzini and Pedersen's 70bp BAB figure ends in 2012

Stulz's forty-year review adds no tests; our 2020 to mid-2024 large-cap run lost 1.81% after commissions.

2026-09-29 · 5 min read · Factor Investing — Betting Against Beta · US stocks

Reviewing: Finance at the intersection of academia and practice: from 1986 to 2026 · René M. Stulz · Read it on openalex

Our backtest of this idea

Our automated quick test, not the paper's

Monthly SPY-Beta-Neutral US Stock Long–Short

Backtest period 2020-01-01 to 2024-07-01 · hypothetical, net of modelled costs

Why these figures are not the paper's (1)

Our own audit found this run does not follow the paper faithfully

  • The article's reported US and Swiss historical evidence covers 1926–2012 and 1984–2012, respectively.: Backtest eligible US stocks over 2020-01-01..2024-07-01 as requested. (invalidates: The reported US 70-basis-point monthly excess return and Swiss 61-basis-point monthly excess return cannot be treated as predictions for this backtest.)

These are our findings about our own implementation, not criticisms of the paper. Read the figures below as a description of what we ran.

Jan 2020Total -1.8%Jul 2024
Sharpe
-0.03
Total Return
-1.8%
Max Drawdown
-19.0%
CAGR
-0.4%
Volatility
11.6%
Beta vs SPY
0.12
Trades
3,363

Stulz's 70 basis points a month betting-against-beta figure is too thin to size. It comes from Frazzini and Pedersen (2014), whose US stock sample runs from 1926 to 2012. Stulz's own discussion of costs and published anomalies gives a desk good reasons to test the trade before trusting that return.

The trade inside the retrospective

Stulz wrote for the fortieth anniversary of Financial Markets and Portfolio Management. His account covers four changes in investments since 1986: indexing, factor models beyond the CAPM, behavioral finance and limits to arbitrage, and private markets. He runs no new test. His evidence comes from cited sources, including several of his own. PWL Capital data put passive funds at 30% of US stock mutual fund and ETF assets in 2015 and almost 55% now. From 2015 to 2024, active funds saw $2.4 trillion in outflows while passive funds received $5.8 trillion. An updated count puts US public corporations at 4,010 in 2024, down from a 1996 peak of 8,025. Crunchbase counts 862 US unicorns in February 2026, compared with fewer than 10 in 2010.

The factor discussion supplies the trade. Stulz writes that "the CAPM had a central role in investment discussions in 1986, whether in academia or practice. It no longer has a central role in 2026." Black, Jensen and Scholes (1972) found the security market line crossing the vertical axis above the risk-free rate; its slope was often insignificantly different from zero. Low-beta stocks earned more than the CAPM predicted, while high-beta stocks earned less. Betting against beta (BAB) buys the former and shorts the latter. Fama and French later said beta "does not seem to help explain the cross-section of average stock returns." Stulz presents BAB as a case where the 1972 finding has largely held up under further testing.

What sits behind the 70 basis points a month?

Frazzini and Pedersen's US result is 70 basis points a month over 1926 to 2012, roughly 8.4% a year on simple annualization. Stulz also gives a Swiss result, a nod to the journal's home: 61 basis points a month over 1984 to 2012. He calls them "significant risk-adjusted returns." In the paragraph that reprints them, we found no account of beta estimation, relative leg sizing, rebalancing frequency, or shorting and financing costs.

Those choices determine the book a trader can hold.

Stulz raises the broader cost problem himself. He writes that most anomaly studies do not show profitability after implementation costs. Those costs change with economic conditions, he adds, allowing trading to erase most of an anomaly's profit at some times and leave it at others. US trading costs have fallen sharply over forty years, though that says nothing about short borrow or financing.

His limits-to-arbitrage discussion supplies another warning. After intermediary losses, levered investors "might be forced to close positions precisely when the arbitrage opportunities appear particularly large." A levered long-short BAB book fits that description, though Stulz makes no explicit connection. The hedge fund industry, which he says exists to exploit arbitrage trades, expanded from under $100 billion in 1986 to more than $3 trillion in the US alone in 2026.

Four hundred fifty-two anomalies later

BAB appears alongside the factor zoo, an uneasy place for a trading claim. Hou et al. (2020) examined 452 anomalies, and a Google Scholar search for stock market anomalies returns 356,000 results. McLean and Pontiff (2016) find that many published anomalies lose profitability after publication. Stulz notes the proposed explanation: asset managers trade on published papers. Harvey et al. (2016) observe that the anomalies were mined from the same database, so the findings are not independent discoveries.

Even familiar factors have long lean spells. Stulz cites extended unprofitable periods for value and size, the best-known anomalies. He also cites Bessembinder et al. (2021), who find that the number of factors needed to explain returns varies with economic conditions.

The dates matter most to a current BAB trader. Frazzini and Pedersen's sample stops in 2012; their paper appeared in 2014. A trade placed today falls wholly in the post-publication period in which McLean and Pontiff find erosion. Stulz offers a possible counterforce. Citing Haddad et al. (2025), he argues that indexing, now almost 55% of US stock fund and ETF assets, may reduce demand elasticity for stocks. Mispricings would then require larger price moves to draw in capital. For a crowded large-cap book, I put more weight on decay. A clean post-2014 result across a broad universe, net of borrow and financing, would change my mind.

Our 2020 to mid-2024 run

We built and backtested our own BAB version. The following figures are ours and cover 2 January 2020 to 1 July 2024.

Each year we refreshed a universe of the 100 largest US stocks by capitalization. Each month we estimated beta to SPY using 252 daily returns and excluded names with beta at or below 0.1. We bought the lowest-beta 30% and shorted the highest-beta 30%, equal-weighting stocks within each leg. Each leg targeted 0.5 of estimated SPY beta. Gross exposure was capped at 4.0, positions at 10%, and trades filled at the monthly close. We charged commissions of four tenths of a cent a share, subject to a $1 minimum.

It lost money. Total return was -1.81%, CAGR -0.41% and Sharpe -0.03. Maximum drawdown reached 18.97%, with volatility of 11.56%.

The cited US return comes to about 8.4% a year on simple annualization. Our -0.41% a year trails it by roughly nine points, but the setups differ substantially. Frazzini and Pedersen's figure is an 86-year excess return for US stocks from 1926 to 2012; Stulz's text states no cost assumption for it. Our figure covers 54 months and 100 mega-caps, net of commissions.

Our return also leaves costs out. We charged no short borrow for roughly 30 short names and no financing on gross exposure up to 4.0, so the true result is worse. We could not verify that the annual universe lists were available at each formation date. The resulting survivorship risk has an unknown sign.

This is one automated pass, with no claim to settle the authors' result. The 100-name universe, legs of about 30 stocks, equal weights and 4.5-year window are our choices. Our loss fits post-publication decay. It also fits the ordinary time variation Stulz describes; 54 months cannot distinguish the two.

Testing on the desk's own costs

For a trader, Stulz adds no new BAB evidence. He repeats 70 and 61 basis points, then moves on. His useful contribution is to put a CAPM failure he says has largely survived beside two warnings: most anomaly studies do not establish profits after implementation costs, and many published anomalies fade. A desk already running a low-beta tilt learns nothing new here. One considering the trade should test it after 2014, with borrow and financing charged.

Our backtest stops at 2024-07-01, and everything after that date is deliberately left untouched so the same strategy can be checked out of sample later.

How our backtest worked

The steps the code we ran actually executed, from its strategy card. Ours, not the paper's — it is one automated implementation of the idea, not the authors' own.

On the first trading session of each year, refresh the eligible top-100 capitalization universe using records available by formation; target zero for holdings no longer eligible.
On the first trading session of each month:
  Using adjusted closes through the preceding session, pair each stock's daily returns with SPY returns over 252 trading days.
  Exclude stocks with fewer than 200 paired returns, invalid beta, or beta <= 0.1.
  Rank by estimated beta, breaking ties by symbol_key. Buy the bottom floor(0.30 × count) and short the disjoint top floor(0.30 × count); stay flat if either leg is empty.
  Set equal dollar weights within each leg so estimated SPY-beta exposure is +0.5 long and -0.5 short.
  Scale all target weights together to respect 4.0 gross leverage and a 10% absolute per-position cap.
  Execute changes at that session's adjusted close. Skip orders without an actual execution close; retry eligible-universe exits when one becomes available.
New positions earn returns only after execution.