Arnott, Harvey, Kalesnik, and Linnainmaa (2019) identify three fundamental blunders that compromise the promise of factor investing. The paper examines 15 popular factors and 33 additional factors over July 1963 to June 2018.
Blunder 1: exaggerated return expectations
Factor premia historically earned by paper portfolios overstate what investors should expect going forward, for several reasons:
Data mining and the factor zoo. Harvey, Liu, and Zhu (2016) document over 400 factors published in top academic journals. Most deliver statistically significant in-sample returns, but many are likely spurious discoveries from combing through finite historical data. Chordia, Goyal, and Saretto (2017) examine 2.1 million equity-based trading strategies and find only 17 that survive multiple-testing adjustments, none of which have meaningful economic rationale.
Post-publication decay. Factor performance displays a clear breakpoint at the end of each factor’s original sample period. The average return in the 10 years after the end of the in-sample period is less than half that of the prior 10 years. McLean and Pontiff (2016) document this degradation systematically, and 12 of 97 published anomalies fail to replicate even in-sample.
Crowding. As investors pile into known factors, the mispricing disappears. Backtest results do not reflect the market impact of capital flowing into the strategy. Post-publication, factor strategies attract capital that compresses the premium.
Trading costs. Novy-Marx and Velikov (2016) show that almost no long-short factor with turnover exceeding 50% has any return left after transaction costs. Hou, Xue, and Zhang (2017) find that 64% of 447 factors fail to deliver significant alpha once micro-cap stocks (bottom 2% by market cap) are excluded.
Valuation dependence. A factor backtest may look impressive if it begins when the factor is cheap and ends when it is expensive. This valuation change inflates historical returns but is unlikely to repeat.
Blunder 2: underestimated tail risk
Factor returns deviate far from normality:
- Excess kurtosis is positive for all factors examined, often considerably so
- momentum has a skewness of -1.41 and excess kurtosis of 11.38; its worst month (-24.3%) was an 8-sigma event under a normal distribution, expected once in 4.1 quadrillion years
- Operating profitability’s worst month (-24.3%) has similar tail extremity
- 11 of 14 individual factors had worst months that would be expected less than once in 2,000 years under normality
Investors who rely on standard risk management tools that assume normality will be surprised by the magnitude of drawdowns. The momentum strategy crashed 44% from peak in 2009, with a 24% drop in April 2009 alone.
Blunder 3: illusory diversification
Forming portfolios of factors does not eliminate tail risk as much as expected:
- The worst single month for a portfolio of the 6 most popular factors is -16.0%, nearly identical to the average worst month of the constituent factors (-17.0%)
- Excess kurtosis for factor portfolios is similar to that of individual factors
- Factor correlations are time-varying and tend to spike during stress periods
- Five-year rolling market betas of factor portfolios varied widely, dipping below -1.0 in the early 2000s and then rising toward zero during the 2008-2009 crisis
- An investor who believed their long-short factor portfolio was “market-neutral” discovered during the GFC that the portfolio had a beta indistinguishable from 1.0
Recent factor performance (2003-2018)
In the 15-year period from July 2003 to June 2018, largely out-of-sample for most factors:
- Four of the six most popular factors had near-zero or negative average returns
- Not a single factor (besides the market) delivered a statistically significant excess return
- Three factors had significant CAPM alpha only because of substantial negative betas, requiring leverage to exploit
- Approximately 75% of pre-2003 alphas evaporated in the 2003-2018 period
- value earned only 0.15%/yr (vs. 4.15%/yr full sample); investment earned 0.34%/yr (vs. 4.32%/yr full sample)
Implications
The paper does not conclude that factor investing is broken, but argues that investors must:
- Adjust return expectations for current factor valuations rather than extrapolating historical averages
- Use risk management tools that account for fat tails and time-varying correlations
- Demand economic rationale for any factor, not just statistical significance
- Account for realistic trading costs and market impact
- Recognize that diversification benefits from multi-factor portfolios can vanish precisely when they are most needed