Factor Investing: What the Research Says About Premiums, Decay, Quant Strategies and Alternatives

This article summarises published research. It is general information, not financial advice, and it doesn't know your situation, your tax position or what is available to you. Nothing here recommends a fund, product or strategy.

VettedVids rates videos and books by what people report happened after using them. Research is different: we don't score it. We summarise what each study found, what it studied and its limits, and we show disagreements rather than settle them. Many of the core papers come from researchers at firms that sell factor, trend-following or alternative strategies, most often AQR; we say where. Every claim below links to its source. The same findings, with their limits, are in the Research tab of our factor investing page.

The factor evidence

Size and value. Fama and French found that firm size and book-to-market equity together captured much of the variation in average US stock returns, while the link between market beta and average return was flat (Fama & French 1992). Their 1993 paper turned this into a factor model: market, size and value factors for stocks plus maturity and default factors for bonds, which together appeared to explain average returns on stocks and bonds (Fama & French 1993). The main tests used US stocks from roughly 1963 to 1990 and were in-sample; the later papers below test how durable they are.

Profitability and investment. A five-factor model capturing size, value, profitability and investment patterns described average US stock returns better than the three-factor model, and once profitability and investment were added, the value factor became redundant in the authors' sample (Fama & French 2015). The authors report the model fails for small stocks that invest heavily despite low profitability, and the "value is redundant" result is sample-specific. Separately, in mainly US stocks from 1963, gross profits divided by assets predicted the cross-section of average returns about as well as book-to-market; profitable firms earned higher average returns despite higher valuation ratios, and controlling for profitability markedly improved value strategies, especially among large, liquid stocks (Novy-Marx 2013). Later "quality" and profitability factors use different definitions, so these results don't transfer one-for-one.

Momentum. Buying US stocks with strong past returns and selling those with poor past returns produced significant positive returns over 3- to 12-month holding periods, not explained by systematic risk, and part of the excess return reversed over the following two years (Jegadeesh & Titman 1993). The sample ran from 1965 to 1989, gross of trading costs, and momentum strategies trade heavily.

Value and momentum across markets. Value and momentum return premia appeared in eight markets and asset classes, including stocks in several regions, country indices, bonds, currencies and commodities. Value and momentum were negatively correlated with each other, and each was correlated across asset classes (Asness, Moskowitz & Pedersen 2013). Sample periods differ by market and end around 2011, and the authors were affiliated with AQR, an investment manager that runs value and momentum strategies.

Low beta, and a dispute over how to build it. Frazzini and Pedersen found high-beta assets had low risk-adjusted returns in US equities, 20 international equity markets, Treasuries, corporate bonds and futures, and that a leveraged long-low-beta, short-high-beta ("betting against beta") factor earned significant positive risk-adjusted returns (Frazzini & Pedersen 2014). Novy-Marx and Velikov argued that the factor's reported performance comes largely from non-standard construction choices that effectively equal-weight stocks and tilt toward very small companies, profitability and investment (Novy-Marx & Velikov 2022). The strategy needs leverage and shorting, and Frazzini and Pedersen are AQR-affiliated.

Where the returns come from. Long positions made up almost all of size profits, about 60% of value profits and about half of momentum profits. The value premium shrank with firm size and was weak among the largest stocks, and momentum showed no reliable link to size (Israel & Moskowitz 2013). The study used 86 years of US data plus about 40 years from four international equity markets and five asset classes; the authors are affiliated with AQR.

Risk or mispricing? The papers lean different ways. Jegadeesh and Titman found momentum returns not explained by systematic risk (Jegadeesh & Titman 1993). Asness, Moskowitz and Pedersen read the correlation of value and momentum across asset classes as evidence of common global risks (Asness, Moskowitz & Pedersen 2013). McLean and Pontiff attributed about 32 percentage points of the post-publication decline in 97 predictors to investors trading on the published findings (McLean & Pontiff 2016). None of these papers settles the question.

How much of the evidence survives scrutiny

• Returns weakened after publication. Across 97 published return predictors, portfolio returns were 26% lower out-of-sample and 58% lower after publication. The authors attribute about 32 percentage points of the decline to investors trading on the published findings, with the rest an upper bound for data mining (McLean & Pontiff 2016). These are US cross-sectional predictors, gross of costs, and decay varies widely by predictor.
• A higher statistical bar. Given hundreds of factors tested on the same data, Harvey, Liu and Zhu argue the usual significance threshold is too low and that a new factor should clear a t-statistic above 3.0. They argue that most claimed research findings in financial economics are likely false (Harvey, Liu & Zhu 2016). This is a methodological argument: the right threshold depends on assumptions about how many tests were run, which Jensen, Kelly and Pedersen dispute.
• Most anomalies failed stricter tests. In US stocks, with microcaps reduced by using NYSE breakpoints and value-weighted returns, 65% of 452 anomalies failed a t-value hurdle of 1.96, rising to 82% at a multiple-testing hurdle of 2.78. Even anomalies that replicated were much smaller than originally reported (Hou, Xue & Zhang 2020). The conclusions depend on the value-weighting and breakpoint choices, and the authors promote their own competing q-factor model.
• A counter-argument: most factors do replicate. Using a Bayesian model of factor replication, Jensen, Kelly and Pedersen found that most asset pricing factors can be replicated, cluster into 13 themes, and work out-of-sample in a new data set covering 93 countries. They argue the large number of factors strengthens rather than weakens the evidence (Jensen, Kelly & Pedersen 2023). The conclusions rest on the Bayesian prior and method, co-author Pedersen is an AQR principal, and the data are public.

The two replication studies reach different answers partly through method: one depends on value-weighting and breakpoint choices, the other on its Bayesian prior (Hou, Xue & Zhang 2020; Jensen, Kelly & Pedersen 2023).

Putting factors to work

• Judging a fund by its factor exposures. In a sample free of survivor bias, common factors (including one-year momentum) and expenses almost fully explained persistence in US equity mutual fund returns from 1962 to 1993. The only unexplained persistence was among the worst funds continuing to underperform, and the results did not support skilled managers (Carhart 1997). The paper introduced the four-factor model commonly used to judge factor funds.
• Costs and crowding. Israel and Moskowitz found little evidence that size, value and momentum returns changed with trading costs or institutional ownership over time (Israel & Moskowitz 2013). Looking at 97 published predictors rather than these three factors, McLean and Pontiff attributed about 32 percentage points of the post-publication decline to investors trading on the findings (McLean & Pontiff 2016). Several of the premiums above were measured gross of trading costs, including momentum, which trades heavily (Jegadeesh & Titman 1993).
• Valuations and a "smart beta crash". Research Affiliates argued that, net of changes in valuation levels, many factor returns were much lower than recent performance suggested, and that performance-chasing inflows had pushed valuations up. They warned of a reasonable probability of a "smart beta crash" as factor-tilt strategies grew popular (Arnott, Beck, Kalesnik & West 2016; Arnott et al. 2016, SSRN). This is practitioner research from a firm then selling fundamental-index strategies, and AQR's 2017 paper disputes how overvalued factors were.
• Timing factors. Asness and co-authors found that timing styles by their value spreads looked promising alone but added mostly extra value exposure, and gave disappointing results inside a portfolio that already held value (Asness, Chandra, Ilmanen & Israel 2017). A later century-long study by AQR researchers of value, momentum, carry and defensive premia across six asset classes found that premia vary over time and are mildly predictable, with little relation to macroeconomic risks and some evidence of overfitting; but significant profitability eluded a range of timing strategies once realistic data lags and trading costs were counted (Ilmanen, Israel, Moskowitz, Thapar & Lee 2021; working paper, 2019). Both come from AQR researchers and use different signals; neither tests a retail fund.

We found no study in this set that tests how large a factor tilt investors actually hold through long periods of underperformance.

Quantitative and rules-based strategies

• Momentum crashes. Momentum strategies had strong average returns but suffered infrequent, persistent strings of large losses. These crashes were partly forecastable: they occurred in panic states after market declines, when volatility was high and the market rebounded. A dynamic version that scaled exposure using forecasts of momentum's mean and variance roughly doubled the static strategy's Sharpe ratio in the authors' tests (Daniel & Moskowitz 2016). The data cover US stocks from 1927 plus international equities and other asset classes, and the dynamic strategy is a backtest.
• Trend following. Each of 58 liquid equity index, currency, commodity and bond futures showed returns persisting for 1 to 12 months and partly reversing later, and a diversified time-series momentum portfolio did best in extreme markets (Moskowitz, Ooi & Pedersen 2012). Rebuilding a similar strategy back to 1880, Hurst, Ooi and Pedersen found it was consistently profitable over the 137 years studied (Hurst, Ooi & Pedersen 2017). That long history is a simulation built from historical prices, not fund returns, and the authors were AQR principals when the papers were published.
• Backtest overfitting. Testing many strategy variants and reporting only the best leads to overly optimistic performance expectations. The deflated Sharpe ratio adjusts for the number of trials and for non-normal returns (Bailey & López de Prado 2014), and a companion paper argues that backtests which do not report how many configurations were tried cannot be properly assessed (Bailey, Borwein, López de Prado & Zhu 2014). These are statistical methods rather than a test of any one strategy, and the adjustment needs the true number of trials, which outside readers rarely know.

Alternatives and private markets

• Private equity against public markets. Kaplan and Schoar found average private equity fund returns net of fees roughly equal to the S&P 500, with wide dispersion and strong persistence across a firm's successive funds (Kaplan & Schoar 2005). Using newer Burgiss data on nearly 1,400 US buyout and venture funds, Harris, Jenkinson and Kaplan found buyout funds outperformed the S&P 500 by 20% to 27% over a fund's life (more than 3% a year), while venture funds outperformed public equities in the 1990s but lagged in the 2000s (Harris, Jenkinson & Kaplan 2014). Results swing with the data source (Venture Economics showed lower returns) and the benchmark, and an S&P 500 comparison does not adjust for buyout leverage or small-company exposure.
• Private equity fees and reported performance. Phalippou and Gottschalg estimated average net-of-fee fund performance at 3% a year below the S&P 500 (6% after risk adjustment), with fees of about 6% a year and reported performance inflated by valuations of unsold investments (Phalippou & Gottschalg 2009). Phalippou later found private equity returns roughly matched public equity indices since about 2006, with average net multiples of about 1.55 to 1.63 (about 11% a year), while an estimated US$230 billion in performance fees went largely to a small number of people (Phalippou 2020). The 2009 paper uses older funds and is disputed by later data; the 2020 paper is a single-author critique that discusses industry rebuttals; and IRR, multiple and public-market-equivalent methods give different answers.
• What hedge fund investors kept. Over a 22-year sample of hedge funds, the effective incentive fee rate was about 50% rather than the headline 20%, because about 60% of the gains on which incentive fees were paid were later offset by losses (Ben-David, Birru & Rossi 2026). The working-paper version estimated investors kept 36 cents of every dollar earned above a risk-free hurdle, before risk adjustment (Ben-David, Birru & Rossi 2020). The sample is funds in commercial hedge fund databases (5,917 funds in the working paper), and one author disclosed a past role as a hedge fund general partner.
• Commodity futures. An equal-weighted index of fully collateralized commodity futures from July 1959 to December 2004 earned about the same return and Sharpe ratio as US equities, was negatively correlated with stock and bond returns, and was positively correlated with inflation (Gorton & Rouwenhorst 2006). A ten-year follow-up by the same research group found the conclusions largely held out-of-sample (Bhardwaj, Gorton & Rouwenhorst 2015). The index is equal-weighted and rebalanced monthly; investable commodity funds weight and roll differently, and correlations rose temporarily in the 2008 crisis.

We found no study in this set on private credit, other real assets or the endowment model.

Hedging and tail risk

• Covered calls. Index covered calls earned their risk-adjusted returns mainly from equity exposure plus a short-volatility premium, but also carried an uncompensated "equity reversal" exposure that made up about a quarter of risk for little reward. The short-volatility part had a realized Sharpe ratio near 1.0 but contributed under 10% of risk (Israelov & Nielsen 2015). The decomposition is the authors' own model, covers US equity index covered calls, and was published while the authors were at AQR.
• Protective puts. In simulations and a study of the CBOE S&P 500 5% Put Protection Index, typical protective put programmes were quite ineffective at reducing drawdowns compared with simply holding less equity. Unless purchases and expiries happened to line up with drawdowns, puts could increase drawdowns and volatility per unit of expected return (Israelov 2018). This covers typical rolling put programmes on US S&P 500 options, not every tail-hedge design; some practitioners argue that rules for selling puts early change the results.

Expected returns

• Forecasting the equity premium. Reexamining dividend yields, earnings yields, book-to-market, interest rates and other predictors, Welch and Goyal found the forecasting models had failed both in-sample and out-of-sample over the prior 30 years and would not have helped an investor time the market (Welch & Goyal 2008). Campbell and Thompson responded that forecasts with simple sign restrictions outperformed the historical average, though by a small margin (Campbell & Thompson 2008). Both use the US equity premium with data to 2005, and the debate is over small out-of-sample gains.
• Building an estimate. Ilmanen describes expected-return judgments as mixing historical averages, finance theory, forward-looking indicators such as yields and valuation ratios, and discretionary views. He argues forward-looking indicators have a better forecasting record than historical averages, but that forecasts are noisy and tactical tilts should be a small part of a portfolio's risk (Ilmanen 2012). This is a CFA Institute Research Foundation monograph drawn from the author's book, not a peer-reviewed study, and the author has long worked at AQR.

How we put this together

We searched the academic and practitioner literature on factor premiums, their replication and decay, quantitative strategies, private markets, hedge fund fees, options hedging and expected returns, and kept findings that answer a question an experienced investor faces: which premiums hold up, how much survives costs and publication, and what alternatives and hedges deliver after fees. Where authors work for firms that sell related strategies, we say so. Each finding is listed with its sample and limits in the Research tab of the factor investing page, alongside rated books, research sites, free factor data and tools. Last checked 17 October 2026.