Systematic and Quantitative Investing
Contents
Trevor L Taylor’s approach to investing grew out of systematic and quantitative models, and he now applies that background as a Miami private investor and the founder of Fortius Capital. Systematic investing means making decisions by written rules that can be tested on data, rather than by a fresh call each time. Quantitative investing is the family of methods that build those rules from statistics, mathematics and computing. The two overlap heavily, and together they now run a large share of the money invested in public markets.
What makes an approach systematic
A systematic investor commits in advance to what will be bought, sold and held under stated conditions. The rule might be as simple as owning an index only while it trades above a long average, or as complex as a model that weighs thousands of signals at once. What matters is that the rule exists before the decision, can be written down, and can be checked against history. A discretionary investor may reach similar conclusions, yet each decision depends on the reasoning of the moment and is much harder to test.
The academic roots
Much of the field rests on research into how prices behave. The committee that awarded the 2013 Nobel memorial prize in economics to Eugene Fama, Lars Peter Hansen and Robert Shiller summarized a paradox: prices are very hard to predict over days or weeks, yet their broad course can be foreseen over three to five years1. That combination invites a rules based approach. If short term moves are mostly noise, discipline matters more than prediction, and any lasting edge is likely to come from patterns that show up across long periods and many securities.
Factors and the data behind them
Researchers describe those patterns as factors: characteristics such as cheapness relative to earnings, company size, or recent price momentum that have been associated with differences in returns. Kenneth French publishes the Fama and French three and five factor return series and many related data sets2 at no cost, and they have become a standard starting point for anyone testing a factor rule.
The best test of a factor is data its discoverers never saw. William Goetzmann and Simon Huang examined momentum on the stock market of Imperial Russia3, a market far removed from the modern US data on which most factor research was built, to see whether competing explanations for the effect held up. Out of sample checks like this one separate durable patterns from accidents of a particular data set. A factor that works in one country and one era may simply reflect how that market happened to unfold, while one that appears across continents and centuries has a stronger claim to being real. Even then, a factor can weaken once it becomes widely known and heavily traded, because buyers compete away the advantage.
Trend rules and simple signals
Not every systematic method needs a supercomputer. Trend following, one of the oldest approaches, buys assets that are rising and steps aside from those that are falling. CNBC describes the 200 day moving average, a common trend gauge, as a way to look beyond daily trading noise to the larger market trajectory4. Simple rules like this one have the advantage of being transparent and hard to overfit, and the disadvantage of being widely known. Trend rules also tend to lose money in choppy, sideways markets, when prices cross the average back and forth, and to earn most of their gains during a few long, sustained moves. Investors who use them have to accept long dry spells in exchange for protection during deep declines.
Data and computing power
At the other end of the spectrum, large quantitative firms treat investing as a data science problem. The Wall Street Journal described how scientists at Two Sigma, with backgrounds in fields such as astrophysics and linguistics, program machines to mine newswires, weather reports and social media for investment clues5. The more data a firm uses, the more care it needs to avoid finding patterns that are coincidence, which is why testing on fresh data is central to the craft.
The Renaissance example
The best known quantitative investor is the mathematician Jim Simons. The Wall Street Journal recounted how he started trading from a strip mall office and turned to math and computers to eliminate the emotional ups and downs of investing6, wanting models that would earn money without constant attention.
CNBC reported that the flagship Medallion fund at his firm, Renaissance Technologies, returned 66 percent a year before fees since 19987, and that his rule of thumb was to remove all emotion. Bloomberg noted that Medallion was open only to roughly 300 employees, about 90 of them PhDs8. Few other firms have come close, and the record is often cited as proof that markets contain patterns, along with a warning about how hard they are to find.
When models crowd and break
Systematic methods have their own failure modes. When many firms use similar signals, their positions can unwind together. Amir Khandani and Andrew Lo found that quantitative long and short equity funds suffered unprecedented losses beginning on August 6, 20079, consistent with a rapid deleveraging of similarly built portfolios. Markets that change character can also confound models trained on the past. Bloomberg reported that during the pandemic, wild markets threw off the investing models10 of even Renaissance and Two Sigma, which posted losses in several funds.
Discipline as the real product
The deepest benefit of a systematic approach may be behavioral. A rule written in calm conditions protects an investor from decisions made in fear or excitement, the same idea behind the entries on position sizing and checklists in high stakes work. Rules still need human thought: deciding which hypotheses deserve testing, how large positions should be, and when a model has genuinely stopped working rather than merely hit a bad stretch. The honest test of any quantitative process is what it does under stress, when following the rules is uncomfortable and abandoning them feels reasonable. Discipline in those moments is the real output of the model.
