Quant Investing: An Honest Definition From Someone Who Does It

Quant investing is the practice of buying and selling securities according to rules derived from data, tested on history before any money moves, and executed without discretionary override. That is the whole definition. The three load-bearing words are derived, tested, and executed, and nearly everything that fails in this field fails at one of those three steps.

One number reframes the whole subject. Value, the most documented stock-selection rule in finance, paid an average premium of 4.2 percentage points a year across 98 years of US data. Over the last fifteen of those years, it paid minus 2.7. Both figures come from the same public dataset, the Kenneth French library, computed the same way. A field that publishes its own losing streaks in that much detail is at least honest about its raw material. Whether any of it is usable by an individual investor is the real question, and it deserves a straight answer instead of a sales page.

That answer is what this note attempts. For context: I run quantitative research for institutional clients through Quantalytics, and the backtesting and Monte Carlo engines I use are ones I wrote myself in Python. What follows is a description of the field from the inside, including the parts that do not flatter it.

Systematic versus discretionary: where the line actually sits

A discretionary investor gathers information, forms a judgment, and decides. A systematic investor decides once, in advance, by writing a rule, and then lets the rule decide every time after that. The difference is not intelligence, and it is not data. Good discretionary managers read the same filings and run the same screens a quant does. The difference is where the final decision lives. In quantitative investing it lives in the rule, and the human's job is to build and test that rule, not to second-guess it on a Tuesday afternoon because the market feels heavy.

A quant strategy worth the name has three ingredients. Not two. Three.

1. A hypothesis with an economic reason. Why should this pattern exist, and who is on the other side of the trade? Cheap stocks may outperform because investors overreact to bad news. Winners may keep winning because information spreads slowly. If the only justification is "it worked in the past," you do not have a hypothesis. You have a coincidence with a chart.

2. A backtest with honest accounting. Test the rule on history you did not use to invent it, subtract realistic trading costs, and count every version of the rule you tried, not just the one that looks good. This last part is where most published "strategies" quietly cheat, and it gets its own section below.

3. A rule that runs without override. The backtest measured what the rule does, including its behavior in crashes. The moment you override it, you are running an untested strategy with zero history. Discipline is not a personality trait here. It is a methodological requirement: the evidence only applies to the rule that was actually tested.

Is quant trading the same as quant investing?

Same discipline, different clock speed. Quantitative trading usually refers to shorter horizons, from weeks down to microseconds, where the edge comes from execution, speed, and market microstructure. Quant investing refers to horizons of months to decades, where the edge comes from persistent cross-sectional patterns and disciplined exposure. The three ingredients above apply to both. The practical difference for you: parts of the slow end are accessible to individuals, and none of the fast end is. More on that at the bottom.

What the factor canon actually paid

The academic core of quantitative investing is the factor literature, and its reference dataset is public: the Kenneth French data library at Dartmouth, which has made factor returns freely available for decades, with history back to 1926, updated monthly. For this piece I downloaded the library's annual factor files (June 2026 vintage) and computed the averages myself rather than quoting someone's summary. Here is what the canon paid, measured as the arithmetic average of annual long-short returns.

Click to enlarge

Three honest readings of that table. First, the market itself — the premium for simply holding stocks rather than T-bills — is the biggest number on the board, and it is the one factor you can capture with a single index fund at near-zero cost. Second, the classic long-short premia have been much weaker recently: value averaged minus 2.7 points a year from 2010 through 2024, and size has been roughly nothing. Third, these are paper portfolios, long one basket and short another, rebalanced without any trading costs. Nobody collects these numbers in full.

There is also a risk warning hiding in the momentum row. Its 98-year average of 8.7 points includes 2009, when the momentum portfolio lost 83.8% in a single year as the junk rally crushed anyone short the previous losers. Factor premia are not annuities. They are compensation, in part, for tolerating exactly that kind of episode.

Is the value premium dead?

Nobody knows, and be suspicious of anyone who claims certainty in either direction. What the data does show is that long droughts are not new: value lost 10.8 points in 1998 and 28.6 in 1999, right before one of its best decades. It also lost 46.9 points in 2020, then gained 25.6 and 25.7 in 2021 and 2022. The honest statement is that factor premia are episodic, they can disappear for a decade and return violently, and a strategy built on them only works for investors who can hold through the drought. That behavioral hurdle, not the math, is the hard part.

The industry is a spectrum, and only one end is open to you

"Quant" covers everything from an index fund with a tilt to the most secretive trading operation in history. It helps to see the whole spectrum at once.

The far end of that table deserves its own paragraph, because it distorts expectations for the entire field. Renaissance Technologies' Medallion fund returned roughly 66% a year before fees and 39% after fees, on average, from 1988 through 2018, according to the figures reported in Gregory Zuckerman's book The Man Who Solved the Market. An analysis by Bradford Cornell puts the compound gross figure at 63.3% a year over those 31 years. The fees themselves tell you how unreachable this is: 5% of assets plus, eventually, 44% of profits, and people happily paid. Medallion has been closed to outside investors for decades and deliberately caps its own size, because the strategy's capacity is limited.

Medallion proves that the ceiling of quantitative investing is extraordinarily high. It proves exactly nothing about what you or I can do, because the ingredients (petabytes of cleaned data, dozens of PhDs, microsecond execution, and a capped, closed fund) are not for sale. Anyone offering you "Medallion-like returns" has told you everything you need to know about them.

Backtest overfitting is the trap that eats beginners

Here is the mechanism that separates real quantitative work from astrology with spreadsheets, and it is worth understanding in plain language.

Market data is mostly noise. If you test one trading rule on that noise, its measured performance is a mix of any true edge plus luck. If you test a thousand variations (moving-average lengths, RSI thresholds, entry and exit combinations), the luckiest variation will look brilliant even when no variation has any true edge at all. The search itself manufactures the beautiful equity curve. You did not discover a strategy. You selected the largest of a thousand coin-flip streaks.

The chart below makes this concrete with a toy model: strategies with zero true skill, evaluated on five years of data. The expected in-sample Sharpe ratio of the best variation grows steadily with the number of variations tested, past 1.1 by 100 combinations and past 1.7 by 10,000. The out-of-sample Sharpe of that "winner" stays exactly where the true skill is: zero.

Click to enlarge

What is the Probability of Backtest Overfitting?

Bailey, Borwein, López de Prado, and Zhu formalized this in a 2015 paper as the Probability of Backtest Overfitting, or PBO: the probability that the strategy configuration you selected because it was best in-sample goes on to perform below the median of all the configurations you tried when taken out of sample. In other words, the probability that your winner was a fluke of the search rather than a real edge. A high PBO means the backtest told you more about your optimization process than about the market.

This is not abstract to me. My master's thesis in quantitative finance at HEC Lausanne, titled "On The Relevance Of Optimizing Technical Indicators On The US Stock Markets," applied the Bailey et al. framework to optimized technical-indicator strategies on US equities. The sobering general finding, consistent with the broader literature: strategies built by optimizing indicator parameters rarely survive the test. The configurations that win in-sample routinely fall back into the pack out of sample, which is precisely what you would expect if most of their measured edge was selection luck. Writing that thesis is why I built my own backtesting engines with overfitting checks wired in, and why a gorgeous equity curve now makes me more suspicious, not less.

The defenses are unglamorous: fewer free parameters, an economic reason for the rule before any optimization, data held out and touched only once, and honest counting of every variation tried. None of this fits in a YouTube thumbnail, which is why so little of the retail "quant" content you will encounter does any of it.

Click to enlarge

A 3 percent paper edge dies at realistic trading costs

Even a genuine edge has to survive the cash register. Here is a deterministic example, every assumption stated, that you can recompute on a napkin.

Suppose a strategy has a real gross edge of 3 percentage points a year over a benchmark returning 7%, so 10% gross. Suppose it turns the portfolio over 200% a year, meaning every position is replaced twice, which is mild by the standards of signal-driven stock strategies. Now the accounting:

The 3-point edge is gone, and the strategy now trails the plain index fund it was built to beat. Over 20 years on $100,000, the backtest promised $672,750; the net result is $369,282, against $386,968 for the benchmark. When I built this table I chose deliberately moderate numbers: half a percent per trade is realistic for an individual trading mid-liquidity names once slippage is counted, and plenty of retail strategies churn far more than 200%.

Click to enlarge

This is why professional quant shops obsess over execution and why turnover is a design constraint, not an afterthought. Institutional desks trade at a few basis points; you do not. The arithmetic points one way: for an individual, only rules that survive being traded rarely clear the cost hurdle.

What an individual investor can genuinely use

None of what follows is advice to buy anything. It is a list of documented, readable approaches that apply the systematic discipline at costs an individual actually faces.

Factor exposure through low-cost funds. A cap-weighted index fund captures the largest premium on the factor table at a near-zero expense ratio, and buying the S&P 500 is itself a rules-based act. Factor ETFs extend this: their screening rules are printed in the prospectus, which makes them the rare quant product whose rule you can read before owning it. The quant mindset applies to what you pay, too, because fund fees compound exactly like the cost drag above.

Contribution and rebalancing rules. A standing instruction to invest every month regardless of headlines, and a written rule like "rebalance when an asset class drifts 5 percentage points from target," are quantitative strategies in the full sense: derived from a hypothesis, testable on history, executed without discretion. They are unglamorous, and they are the portion of the field with the best ratio of evidence to cost.

Risk monitoring on documented indicators. Watching a small set of public macro series through a fixed lens, rather than reacting to headlines, is another rule you can write down. That is the design behind the recession radar in the lab, which combines a cycle-armed yield-curve signal with the Sahm rule, and behind my research note on macro crash signals. Whether a signal should change anything about a portfolio is a separate question; knowing what the data says beats guessing.

The discipline itself. The cheapest quant tool available is a text file where you write the rule before you act on it. "I will sell when X" written in advance is a strategy. The same sentence composed during a drawdown is an emotion with a timestamp.

What retail cannot do, and the warning signs

The flip side of the spectrum table. High-frequency trading requires colocation, exchange fee schedules, and infrastructure measured in nanoseconds; the game is decided before your order leaves your broker. Short-horizon statistical arbitrage dies, for an individual, on the cost arithmetic in the section above, plus borrow fees on the short side. These are not skill gaps you can close by studying. They are structural.

And because "quant" now sells, a red-flag list, two items of which appear almost verbatim on the SEC's investor-education checklist of fraud warning signs:

• A Sharpe ratio above 2 with no audited track record. Backtests are not track records. A live, audited history is the only evidence that survives this article's overfitting section.
• "Guaranteed" or "risk-free" returns. Both phrases appear on Investor.gov's red-flags checklist for a reason. Real quant returns, including Medallion's, come with drawdowns.
• Marketing built entirely on a backtest chart. Ask how many variations were tested. Silence is an answer.
• A refusal to describe the rule even in outline. Legitimate systematic funds describe their style (trend, value, carry) while keeping parameters private. "Proprietary algorithm, trust us" describes nothing.
• Urgency and scarcity. A real edge with real capacity limits is not sold to strangers through ads. Medallion, the most capacity-constrained strategy known, solved its capacity problem by returning outside money, not by recruiting more of it.

The honest pitch for quant investing was never secret formulas. It is more modest and more useful: rules you can read, tested on data you can check, executed with a discipline that does not depend on mood. The parts of that pitch that work at retail scale are cheap, slow, and public. Everything about the field that is fast, secret, or expensive is either out of reach or, when offered to you anyway, a warning sign.