Factor Investing with ETFs • Samuel Lee, ETF Strategist
Outline • Model-based investing 101 • Factor-based view of the world • Practical factor investing • Presenter: Alex Bryan • Parting thoughts • Shameless product pitch
Human foibles • Overconfidence • Most people think they’re above-averagedrivers, lovers, socializers, workers, etc. • Privilege vivid, resonant ideas and thoughts over statistical facts • Over-reliant on narrative thinking
Models usually beat intuition • Paul Meehl, “Clinical vs. Statistical Prediction: A Theoretical Analysis and a Review of the Evidence”, 1954 • Found that in 20 studies, overall simple models beat trained experts • William Grove, et. al., “Clinical Versus Mechanical Prediction: A Meta-Analysis”, 2000 • Of 136 studies, 46% favored models, 48% tied, 6% favored humans • Philip Tetlock, “Expert Political Judgment”, 2006 • 20-year forecasting study on 300 experts; simple prediction models won.
Imitating the best • Top investors are model-driven. • Warren Buffett: buy when price < intrinsic value with margin of safety • Jeremy Grantham: buy when current valuation < historical averages • Ray Dalio: Economy is a transaction-based machine driven by short-term debt cycle, long-term debt cycle, productivity growth • Jim Simons: ??? Black box models
Data-snooping • Definition: Testing many different models to obtain a desired outcome.
The problem with scienceas practiced • Many studies turn out to be false. • John P.A. Ioannidis, “Why Most Published Research Findings are False”, 2005 • Argues current study methodologies don’t have enough statistical power and are biased to false positives. • Pharma giant Bayer couldn’t replicate 2/3rds of 67 studies it tried to replicate (2011) • Amgen couldn’t replicate more 90+% of 53 “landmark” papers in cancer research (2012)
Skewed incentives • Publication bias: Statistically significant positive findings more likely to be published. • Encourages data-snooping by researchers to get published.
Quantitative strategies: witchcraft? • Even worse incentives to data-snoop. • Joel M. Dickson, et. al., “Joined at the Hip: ETF and Index Development”, 2012 • Back-tested equity indexes beat U.S. market 12.25% annualized in back-tests, returned -0.26% annualized live.
Hallmarks of a good back-test • From most to least important: • Strong economic intuition • Intellectually honest source. Credibility! • Simple and transparent methodology. • Large sample size that spans many decades and many countries. • Economically and statistically significant results. • Then find multiple independent researchers coming to the same conclusion!
Academia has a decent record in producing models • R. David McLean and Jeffrey Pontiff, "Does Academic Research Destroy Stock Return Predictability“, 2013. • Independently replicated 82 characteristics purported to predict stock returns. • Of 72 that could be replicated, returns after study on average decayed 35%--10% from statistical bias, 25% from arbitrage.
The big two, value and momentum • Value – Tendency for stocks cheap by fundamental measures to outperform stocks expensive by such measures. • Usually defined as having low price/book. Low price/earnings, price/cashflow, price/sales and the like also work. • Momentum – Tendency for performance to persist. • Relative momentum assets are ones with highest relative 12-month returns. • Time-series momentum assets have positive 12-month returns.
How a value strategy usually works • Define a universe of stocks. • Calculate a valuation ratio, most commonly price/book, for each stock. • Sort stocks. • Select a basket of the “cheapest”. • Hold for one year and repeat.
High book/price (value) stock portfolios have outperformed Source: French Data Library
How a stock-based momentum strategy usually works • Define a universe of stocks. • Calculate past return, usually 12 months, excluding the last month. • Sort stocks. • Select a basket of the highest-return. • Hold for one month and repeat.
High momentum stock portfolios have outperformed Source: French Data Library
Why believe these back-tests? • Strong economic intuition. • Value effect: Investors overextrapolate recent trends, leading to stocks overshooting to up or downside. • Momentum effect: Investors underreact to new information, and herd into same stocks/positions. • Identified by numerous credible and independent researchers. • Simple and transparent methodology. • Found in almost every country studied (except for Japan). • Statistically and economically significant.
What is a factor? • Acharacteristic that explains an asset's returns. • Risk factors: • Market/economic growth • Inflation • Duration • Illiquidity • Behavioral “factors”: • Value? • Momentum
Factors : assets :: nutrients : foods • In factor theory, an asset’s expected returns are derived wholly from its exposure to various risk factors. • Assets are bundles of factors. • Risk factors represent unique and different kinds of “bad times.”
What is a (linear) factor model? • A theory about what explains an asset’s returns • Usually takes the form of a linear relationship
Original factor model: CAPM • Asset’s expected returns determined by its covariance with market. where is the expected return of the asset, is the risk-free (cash) rate, is the asset’s beta to the market, and is the market return
Lots of strict assumptions • Everyone rational • Risk averse • No one can influence prices • No transaction costs • Everything tradable • Investors only care about return and standard deviation • Everyone agrees on correlations, expected returns of all assets • One period • Investors can lend and borrow at risk-free rate
Implications • All investors would own market portfolio. • Only differences are portion of cash + market portfolio + leverage. • Market portfolio is “mean-variance efficient”—no other combination can produce a superior risk-adjusted return
Linear regression • Statistical procedure to estimate linear relationship between two sets of data. • Often interpreted as cause-effect relationship: x causes y. • CAPM is based on simple linear regression! • Example: How well does the market’s monthly excess return predict Fidelity Magellan’s FMAGX excess return?
CAPM regression • Model parameters (betas) cannot be directly observed. They must be estimated. • CAPM regression: Where is a variable representing the excess return of the asset at time , is the intercept, is the asset’s estimated beta to the market, is the market’s excess return, and is a noise term.
CAPM doesn’t work! • Black, Jensen, Scholes, “The Capital Asset Pricing Model: Some Empirical Tests”, 1972. • Created portfolios of stocks sorted by CAPM beta. • Found that CAPM beta did not predict excess returns linearly, i.e., securities market line was too flat. • Low-volatility anomaly: Higher beta != higher returns
Multi-factor models • Fama-French model: Added “size” and “value” to CAPM. • Fama and French argued size and value predicted and explained stock returns. • Carhart model: Added “momentum” to Fama-French model. • Jegadeesh and Titman “discovered” momentum effect. Carhart added it to FF.
Carhart model • CAPM plus three new “factors”: value, size and momentum: Where (or “high minus low”) is the return of the value-factor-mimicking portfolio, (or “small minus big”) is the return of the size-factor-mimicking portfolio, and (“up minus down”) is the return of the momentum-factor-mimicking portfolio.
Splitting the equity universe by size and value • Portfolios are intersections of size and value breakpoints, formed yearly at June end. • Market-cap breakpoint determined at June end. • B/P breakpoint determined previous fiscal year book/end of last year market-cap—can be up to 1.5 years stale! Median market cap Big value Small value 70th B/P percentile Big neutral Small neutral 30th B/P percentile Small growth Big growth
Size factor construction • SMB, or small minus big, is the average return of three small portfolios minus average return of three big portfolios.
Value factor construction • HML, or high minus low, is the average return of two value portfolios minus the average return of two growth portfolios.
Splitting the universe by size and momentum • Portfolios are intersections of size and momentum breakpoints, formed monthly. • Market-cap breakpoint determined at June end. • Prior 2-12 returns are the 12-month returns of stocks, excluding most recent month. Median market cap Small up Big up 70th prior (2-12) percentile Small medium Big medium 30th prior (2-12) percentile Big down Small down
Momentum factor construction • UMD, or up minus down, is the average return of two winner portfolios minus the average return of two loser portfolios.
Momentum is the strongest; size is the weakest Source: French Data Library
Factor models raise the benchmark for active managers • A manager’s excess returns attributable to a factor is no longer “alpha” • Studies demonstrating active managers can’t outperform use factor models • Factor models have a hard time detecting statistically sig. alpha • Factor research is ongoing
Quality/profitability • Firms with high profits, low leverage, low earnings variability, high payouts persistently outperform • “Gross profitability” defined as (revenues – cost of goods sold)/assets • Robert Novy-Marx, “The Other Side of value: The Gross Profitability Premium”, 2013 • Gross profitability is strong as value factor, but negatively related.
Splitting the equity universe by size and gross profitability • Portfolios are intersections of size and profitability breakpoints, formed yearly at June end. • Market-cap breakpoint determined at June end. • GP/A breakpoint determined previous fiscal year book/end of last year market-cap—can be up to 1.5 years stale! Median market cap Small profitable Big profitable 70th GP/A percentile Big neutral Small neutral 30th GP/A percentile Small unprofitable Big unprofitable
Profitability factor construction • PMU, or profitable minus unprofitable, is the average return of two profitable portfolios minus the average return of two unprofitable portfolios.
Quality stocks took a beating after the Nifty Fifty craze Source: Robert Novy-Marx Data Library
At what price quality • Jeremy Siegel, “Valuing Growth Stocks: Revisiting the Nifty Fifty”, 1998 • Argues high-quality “Nifty Fifty” stocks of early 1970s warranted high valuations