Investing Classics: Quant & Systematic InvestingRules, Factors, and the Limits of the Machine

Factors are the bedrock of quant. In 1992/1993 Fama and French used value (low price-to-book) and size (small caps) to explain most of the cross-section of stock returns, later adding profitability and other quality factors; momentum was documented by Jegadeesh and Titman in 1993. Tellingly, this quote comes from Fama—father of the efficient-market hypothesis—admitting momentum is the "premier anomaly" his own model cannot explain. Why do factors persist? Usually two explanations: risk compensation (cheap value stocks are more fragile, so the premium is pay for bearing risk) or behavioral bias (investors systematically chase winners and shun boring bargains). The key: a factor is not a free lunch—it demands that you not flee during long stretches when it fails.

Value's "lost decade" is the best teacher. From 2010 to 2020, the classic long-short value portfolio (buy cheap, sell dear) suffered one of the worst drawdowns in history, trailing growth for years; even AQR, famed for quant, saw its flagship strategies struggle through 2018–2020, and founder Cliff Asness took to writing essays defending the faith. Just as many declared "value is dead," it roared back from late 2020 into 2021. A factor validated over decades can still make you question your sanity for ten straight years—which is exactly the hardest part of factor investing: sound in math, brutal on the nerves.

Limits: factors decay once published (anomaly returns fall by roughly 58% after publication, McLean & Pontiff 2016); crowding erodes the premium; many "factors" are just data-mined noise. Misuse: treating short-term performance as proof of validity, or stacking so many factors you overfit.

• Does this factor have an economic or behavioral rationale independent of the data?
• Does it hold across different markets, eras, and asset classes—not just one slice of US equities?
• Can I endure it lagging for 5–10 years? When it lags, will I double down or capitulate?
• After trading costs and turnover taxes, how much of the paper premium survives?

Sharpe's "arithmetic of active management" is an identity, not an opinion: the sum of all investors' holdings is the whole market, so before fees active and passive earn the same, and after fees active as a group must lag. This dictates the logic of the three layers—Index: give up seeking an edge, accept the market return, drive costs to the floor; Factor: use rules and diversification like an index, but tilt systematically toward a characteristic—the middle road between active and passive; Stock-picking: concentrate, which demands a genuine informational or analytical edge—concentration without edge merely amplifies randomness. The essential difference among the three is: where does your edge come from?

The SPIVA scorecard confirms Sharpe year after year: over the past 20 years roughly 90% of US large-cap active funds trailed the S&P 500. The most dramatic proof was Buffett's 2007 million-dollar bet—that an S&P 500 index fund would beat a basket of hedge funds over a decade. The 2008–2017 tally: the index fund returned +125.8% cumulatively; his opponent's fund-of-funds about +36% (only ~2.2% annualized). Fees and the arithmetic of active management quietly crushed some of Wall Street's smartest.

Limits: an index is not risk-free—cap weighting makes you buy ever more of a few giants (the US "Magnificent Seven" once made up a large share of the S&P), especially dangerous at a bubble top. Factors can lag for years; stock-picking remains optimal for those who truly have an edge (Buffett himself is the counterexample). Misuse: forcing stock-picking without the ability, or using "I understand it" to mask having no edge.

• Do I go active/stock-picking on an edge I can articulate, or just because I can't stomach the average?
• Are my turnover and fees quietly eating all of my excess return?
• Is my index holding already highly concentrated in a handful of mega-caps?
• Of the three layers, which actually fits my time, temperament, and edge?

This line captures the essence of quant: an edge can be razor-thin, as long as you bet often enough, independently enough, and with enough discipline. Medallion does not predict that a given stock will rise; it places repeated bets on vast numbers of tiny, very-short-horizon signals, letting the law of large numbers turn a sliver of win-rate into staggering compounding. Renaissance hires mathematicians, physicists, and codebreakers—almost no one with a Wall Street background—because it wants scientific method, not market intuition. And the price of this edge: signals decay fast and must be constantly renewed; capacity is severely limited, and every extra dollar dilutes returns.

Per Gregory Zuckerman's The Man Who Solved the Market, Medallion returned ~66% annualized gross and ~39% net from 1988–2018 (charging a 5% management fee plus 44% performance fee—the industry's most expensive). But two details matter most: first, it closed to outside money in 1993 and capped its size at roughly $10 billion—the edge does not scale. Second, Renaissance's public funds (e.g. RIEF) have been mediocre, even losing heavily in 2020. The same firm, the same geniuses, and the public products still cannot reproduce the flagship's magic—which is the loudest warning of all: you cannot buy that edge.

Limits: Medallion is a survivor; for every one of it, thousands of quant funds died. In the "quant quake" of August 2007, many equity market-neutral quant funds (including Goldman's Global Alpha) were trampled within days as crowded positions unwound, taking huge losses. Misuse: mistaking "quant can make money" for "I can make money with quant," treating a singular case as a replicable path.

• Is the track record I envy a replicable method, or a non-replicable exception?
• If the edge were really that good, why did it close and cap its capacity?
• Am I looking at survivors, or have I also counted the dead peers?
• Could crowding—many people using the same signal—make it backfire on everyone one day?

Harvey's warning strikes quant's fatal flaw—multiple testing: try a few hundred variants and pure luck will hand you one with a beautiful backtest, but it has only fit noise; he therefore argues a factor's t-stat threshold should rise from 2.0 to above 3.0. For an individual, the honest division of labor is: what you can do is systematize discipline—rule-based rebalancing, using factor ETFs for diversified exposure, mechanical position sizing, stripping emotion out of order entry; what you cannot do is mine genuine alpha out of price data. Engineers should be especially wary: the better you are at machine learning, the easier it is to train noise into a "signal" in a domain this low signal-to-noise and prone to leakage.

The "replication crisis" in factor research gives cold, hard numbers. Hou, Xue, and Zhang's 2020 Replicating Anomalies systematically tested 452 published anomalies; about 65% failed to replicate (with microcaps excluded, the failure rate is even higher). Combined with the ~58% post-publication decay noted above, the conclusion is clear: a strategy with a paper Sharpe of 2 often approaches 0 live. It's not that the market changed—the "edge" in the backtest was, more often than not, an overfitting illusion from the start.

Limits: individual quant has real value—it fights behavioral bias rather than trying to beat institutions. Misuse: treating a backtest as an "experiment" (a backtest can only falsify, never prove), or tuning parameters until the curve looks pretty and then going live. And don't forget transaction costs, taxes, and slippage will eat most of the paper return.

• Does the strategy still hold on out-of-sample data it has never touched?
• How many variants did I try in total? Have I discounted my Sharpe for multiple testing?
• Does it have an economic rationale independent of the backtest, or is it pure curve-fitting?
• After real costs, taxes, and slippage, does any excess return remain?
• Do I do quant for discipline, or a fantasy of copying RenTec?