Quantitative Investing: What Individuals Can Use, and Why Most Backtests Mislead

Quantitative investing means picking investments with written rules applied to data. A fund manager might rank thousands of stocks by valuation and price trend, forecast returns with a regression, or trade price gaps between related securities, and each step can be repeated by a computer and tested on past prices. Most of what is sold as quant, from satellite-image data to microsecond execution, needs money and infrastructure that individual investors do not have, so this guide sticks to what an individual can use and how to tell whether a rule works.

The more useful skill turns out to be skepticism about backtests. A backtest is a simulation of what a rule would have earned on past data, and it is the main selling tool for every systematic product. The sections below cover the funds that are available, then four ways backtests mislead, with numbers computed from public data. Factor definitions and long-run factor results are in our factor investing guide, and this one focuses on testing and costs.

What individuals can actually use

• Factor ETFs. As of September 2026, the iShares MSCI USA Momentum, Quality, Value, and Min Vol factor ETFs each list an expense ratio of 0.15%, against 0.03% for the iShares Core S&P 500 ETF. On $10,000 the extra 0.12 percentage points cost $12 a year. The momentum fund alone had about $20.4 billion in assets on September 28, 2026, so this is not a niche product. Other firms sell similar funds; compare index rules and fees, not names.
• Rules-based index funds and direct indexing. Our index fund guide and direct indexing guide cover the low-cost end. Direct indexing platforms can add factor tilts inside a separate account.
• Do-it-yourself tests. Kenneth French's data library publishes monthly factor returns back to 1926 for free. That is enough to test a factor idea against what academics found, though not to trade it.
• Rebalancing rules. Setting a written schedule for portfolio rebalancing is a small quant decision that most investors benefit from.

Statistical arbitrage, which trades price gaps between related securities, and machine learning models on alternative data sit mostly inside hedge funds; see our hedge fund strategies guide for how they are structured and paid. The public evidence on what individuals earn from them is thin, so this guide makes no return claims for them.

Four ways a backtest misleads

Look-ahead bias

A backtest uses information that was not public on the trade date. Company accounts, for example, are released weeks or months after the period they describe, and are sometimes revised later. Academic factor construction avoids this with a delay. In French's six-portfolio value-and-size sorts, book value for the fiscal year ending in year t-1 is used for portfolios held from July of year t, so a December 2020 balance sheet is not used until July 2021. A home-built test that matches December results to January prices would look better than any real portfolio could have done.

Survivorship bias

A backtest that starts from today's list of stocks or funds leaves out the ones that failed or were bought out. Rules like "buy the cheapest 10%" then look safer than they were, because many cheap stocks are cheap for good reasons and some went to zero. Ask whether the data includes delisted securities with their final returns, and whether index membership is taken as of each past date.

Overfitting and multiple testing

Try enough rules and one will fit the past by luck. Bailey and coauthors report that with five years of daily data and 45 or more independent variations of a strategy, the best one is likely to show a Sharpe ratio of 1.0 or more even when nothing works. Harvey, Liu, and Zhu wrote in 2014 that given hundreds of published factors, the usual t-ratio cutoff of 2.0 makes no statistical sense, and that a new factor should clear about 3.0.

Illustration: we simulated strategies with no real edge (true average return of zero), independent of each other, with normally distributed monthly returns. For each row we drew N strategies, recorded the best annualized Sharpe ratio among them, and repeated that 400 times.

Two lessons follow from the table. Longer histories help a lot, and the number of tries matters as much as the history length. The count that matters includes every variation you tried and dropped, not just the one you present.

The same arithmetic gives a plain rule for significance. A zero-skill strategy passes a t-statistic of 1.96 about 1 time in 20, so 14 independent tries give even odds that one passes and 45 give a 90% chance. At the 3.0 cutoff, a pass by luck is about 1 in 370: it takes 257 tries for even odds and 852 for 90%. How long a true edge takes to show also depends on its size. A t-statistic equals the Sharpe ratio times the square root of the years of data, so a real Sharpe of 0.5 needs about 16 years to reach 2.0 and 36 years to reach 3.0. For comparison, the US market's excess return in French's data had a Sharpe ratio of 0.40 from 1927 through 1993 and 0.60 from 1994 through August 2026.

Costs and decay

A paper portfolio trades at closing prices with no spread, commission, or price impact. Real trading pays all three, and the bill rises with turnover.

Illustration: the annual drag from trading equals the share of the portfolio replaced each year, times two (one sale and one purchase for each replaced dollar), times the all-in cost per trade. The costs below are round assumptions, not measured figures.

For scale, French's momentum factor earned about 4.3% a year from 1994 through August 2026, before any costs, as a long-short paper portfolio. A strategy that replaces 300% of its holdings a year at 0.30% a trade gives up 1.8 points, or about 42% of that premium. At 0.50% a trade it gives up 3.0 points, and at 600% turnover and 0.30% it gives up 3.6 points, or about 84%. Small-company stocks cost more to trade than large ones, and the more money a rule manages, the more its own trades move prices.

A worked example: momentum before and after the paper

Momentum is the case where a backtest famously worked: buy stocks that rose most over the past year (skipping the latest month) and sell those that fell most. The academic case for it appeared in the early 1990s. Using French's momentum factor with monthly data through August 2026, we split the record at January 1994.

The full 1927 to August 2026 record still shows 6.0% a year compounded and a t-statistic of 4.51. But the recent record is weaker than most product brochures suggest. Since 1994 the premium does not clear the 2.0 bar, and since 2008 it is essentially zero before costs. The 1994 split is our choice and was not fixed in advance, and a difference of 4.5 points between the two halves is only about 1.3 standard errors, so noise could explain it. We can't tell from this data whether crowding, higher trading costs, or bad luck is behind the fade. Value, measured the same way, went from 5.4% a year (t = 3.48) before 1994 to 1.8% (t = 0.89) after, and the size factor from 2.9% to 0.3%, though value has recovered since 2021, as our factor guide covers.

The practical reading is modest. A factor can be real and still lose for a decade, and a decade is short relative to how long it takes to prove a premium exists.

Rules of thumb before you trust a systematic strategy

1. Ask how many variations were tested before this one was chosen. If the answer is unknown, treat the record as a hypothesis.
2. Look for a reason the premium exists that does not depend on the data: compensation for risk, a constraint that forces others to trade, or a known behavioral error. A pattern with no story attached is more likely noise.
3. Hold back data. Fix the rule before looking at the last portion of history, then test once. A second look at the same holdout uses it up.
4. Demand costs in the record. Ask for the strategy's turnover and the trading-cost assumption. For funds, use the return after fees and the fund's actual turnover from its prospectus.
5. Check for a live record. A fund with only a backtest and no history of real trading has not been tested against costs, taxes, or its own inflows.
6. Size it so a lost decade is survivable. With a core market fund and 10% to 30% of stocks in tilts, being wrong costs a few tenths of a point a year.

Who should skip it: investors who would sell after a few bad years, anyone who can't state their own rule for selling, and anyone paying performance fees for strategies they can't inspect. A plain index fund is the comparison point every systematic product has to beat after fees, and for many investors that comparison ends the search. For automated portfolios that use simple rules, see our robo-advisor guide.

This guide is for informational purposes only and does not constitute investment advice. Factor statistics are computed by us from the Kenneth R. French Data Library (US long-short paper portfolios, monthly, through August 2026), before costs and taxes. Simulations and cost tables are illustrations with stated assumptions. Fund fees are as of September 2026. Past performance does not guarantee future results. Consult a qualified financial advisor before changing your portfolio.