Factor Investing: A Systematic Approach, From Fama-French to Machine Learning

Factor investing is a systematic approach to picking stocks. You buy companies for a trait you can measure, like cheapness or recent price strength, and skip the forecast.

The trait is called a factor. A rule scores every stock on it, buys the top names, and repeats on a schedule. Nobody sits in a room deciding they like the CEO.

I build systematic models for advisers, so I live with rules like these. Here's how a factor rule runs and the five factors you'll meet. Then the dated stretches where each one failed, and where machine learning fits.

What is factor investing, in plain terms?

Start with a universe of stocks. Score every name on one measurable trait. Buy the highest scores. Rebalance on a fixed calendar and repeat.

That's the whole discipline. One rule, applied without discretion, for a long time.

The academic spine came from Eugene Fama and Kenneth French. Their 1992 paper showed that size and book-to-price explained much of the spread in stock returns. In 1993 they built the three-factor model: market, size, and value. In 2015 they added profitability and investment to make five. Momentum came separately, from Narasimhan Jegadeesh and Sheridan Titman in 1993.

What does "a systematic approach to factor investing" mean?

It means building a portfolio from measurable risks, chosen by rule. It's also the subtitle of Andrew Ang's 2014 textbook from Oxford University Press. The full title is Asset Management: A Systematic Approach to Factor Investing. Ang taught at Columbia and later ran factor strategies at BlackRock.

The book's core idea is simple. Asset classes are bundles of deeper risks. Ang compares factors to nutrients and assets to the foods that carry them. Two foods can look different and deliver the same nutrients. A bond fund and a stock fund can both be loaded with the same economic-growth risk.

So the book argues for allocating to factors directly. Its anchor case is Norway's sovereign wealth fund. After 2008, Ang co-wrote a review for Norway's Ministry of Finance. It found that much of the fund's active return came from systematic factor exposures.

That's the practical point for an adviser. Much of what gets sold as manager skill is factor exposure you could buy more cheaply.

How does factor investing work, step by step?

1. Define the universe. Usually a broad index, like the S&P 500.
2. Define the metric. Book to price for value. Trailing twelve-month return for momentum. Profitability and debt for quality.
3. Score and rank. Every name gets a number.
4. Select and weight. Take the top slice, often a fifth or a tenth. Weight by score, by market cap, or equally.
5. Rebalance on a calendar. Monthly is common for momentum, annual for value. Faster tracks the trait better and costs more to trade.

Notice what's missing. No step asks what the market is doing. A value screen in March 2020 looked exactly like one in March 2019.

What are the main factors an adviser will run into?

• Value. Stocks that are cheap against book value, earnings, or cash flow.
• Size. Smaller companies over larger ones.
• Momentum. Stocks that rose over roughly the past year, usually skipping the latest month.
• Quality. Profitable companies with low debt and stable earnings.
• Low volatility. Stocks whose prices move less than the market's.

Each has decades of research behind it. Each has also spent long stretches losing to a plain index fund.

The list is short on purpose. Rob Harvey of Dimensional Fund Advisors described the vetting problem on Behind the Ticker. "There are hundreds of factors or premiums, whatever you want to call it, that are out there in the market," he said. Dimensional held size and value for years before it added profitability in 2012.

How is machine learning used in factor investing?

Machine learning swaps the single hand-built score for a model. The model reads dozens or hundreds of stock traits at once. It learns which combinations predicted next month's returns, then ranks stocks on that forecast.

The reference text is Machine Learning for Factor Investing by Guillaume Coqueret and Tony Guida, published in 2020. It walks through tree models, neural networks, and how to test them without fooling yourself.

The best-known study is Shihao Gu, Bryan Kelly, and Dacheng Xiu's 2020 paper in the Review of Financial Studies. They tested many methods on decades of US stocks. Trees and neural networks beat linear regression at predicting returns. Every method agreed on which signals mattered most: price trends, liquidity, and volatility.

That's the uncomfortable part. The fancy models mostly rediscovered momentum and low volatility.

The risk is overfitting. Campbell Harvey, Yan Liu, and Heqing Zhu counted more than 300 published factors in 2016. They argued most were statistical flukes, and proposed a much higher bar for calling one real. A flexible model can fit noise faster than a human can.

What does a factor failure actually look like?

Momentum in 2009 is the cleanest case on record.

The S&P 500 closed at 676.53 on March 9, 2009. That was the bottom. A momentum rule held staples and defensive names. It owned almost no banks or homebuilders, because they'd been crushed for eighteen months.

Then the market turned and the most damaged names rose hardest. Kent Daniel and Tobias Moskowitz measured it. From March through May 2009, the decile of past losers rose 163%. Past winners rose 8%.

The signal wasn't broken. A trailing window simply can't see a sharp reversal coming.

Value ran the same trap on a longer clock. It trailed growth in most years from 2007 through 2020.

Plenty of clients quit value after 2020. The ones who stayed held it through 2022.

Why do investors combine factors?

Because they fail at different times. Here are calendar returns for funds tracking the MSCI USA factor indexes against the S&P 500.

A different factor led in each year. Min vol held up best in 2022 and lagged badly after. Value was last in 2024 and first in 2025.

Cheap stocks are also sometimes cheap for a reason. Harvey described Dimensional in early 2020, when airlines and cruise lines screened as value. Momentum kept them out. "That's a nice way for us to avoid value traps," he said.

Elena Khoziaeva, co-CIO of Bridgeway Capital Management, does the same in small caps. She screens out negative-momentum names, which she called "fallen knives" (at 26:57).

When does factor investing stop working?

The drought outlasts the client. Premiums are measured in decades. Advisers get judged in quarters. Catherine LeGraw of GMO pointed to "the number of deep value managers who've gone out of business in the past five years, more than any other time in history" (at 14:10).

Crowding shrinks the premium. When a factor gets popular, money chases the same names. Prices adjust before the rule fires.

Every provider defines the factor differently. Jon Clements of Market Desk Research gave a sharp example. Two stocks both rise 30% in six months. One climbs steadily, the other jumps on one earnings week. Most momentum rules rank them the same. Read the methodology, not the fact sheet.

The rule answers the wrong question in a drawdown. A factor screen picks what to own inside the equity sleeve. It never decides how much equity to hold. From February 19 to March 23, 2020, the S&P 500 fell 34%. Long-only factor portfolios fell with it, fully invested by design.

How is factor investing different from smart beta?

Same idea, two names. Factor investing is the academic term. Smart beta is the product label, and it stretched to cover almost anything that isn't cap-weighted. More in our smart beta guide.

Where does a systematic risk overlay fit?

THOR's work sits next to factor investing. A factor rule scores what to own. An adaptive model scores how much to be invested, shifting exposure as trend and volatility conditions change. The approach seeks to reduce drawdowns.

Many advisers run both. The low volatility factor is where the two overlap most, and volatility targeting is the sizing side on its own.

Definitions

• Factor. A measurable stock trait that has historically helped explain return differences.
• Factor premium. The extra return a factor has earned over the market across a long period.
• Fama-French three-factor model. A 1993 model explaining returns through market, size, and value.
• Momentum. Ranking stocks by trailing price strength, usually twelve months skipping the latest one.
• Value trap. A stock that screens cheap and keeps falling because the price reflected a real problem.
• Factor zoo. The hundreds of published factors, most of which likely won't hold up out of sample.
• Overfitting. A model that fits past noise and fails on new data.

Educational content only. This is not investment advice and not a recommendation regarding any security. Index figures are measured on funds tracking each index using adjusted closes, and are rounded. Managers quoted are speaking about their own approaches on Behind the Ticker. Investing involves risk, including possible loss of principal. Past performance is not indicative of future results.