What Quant Trading Actually Is (And Isn't)

Quant trading is rules. Pre-defined, written-down, testable rules. The kind that tell a computer “if X happens, do Y” without ever asking a human for permission. That’s it. That’s the whole thing.

That definition has consequences for the common myths and for the arithmetic that decides whether a strategy is worth your time.

Quant trading is rules, not intuition

A quant trader differs from a discretionary trader in one basic way. A discretionary trader watches the tape, reads news, calls their broker, and makes a gut call. A quant trader wrote the call down as code last month. The code runs. The position gets taken. There’s no “I have a feeling” at 9:34 AM.

A working definition: quantitative trading is any strategy where entry and exit decisions are made by a pre-specified algorithm, with zero discretion at the moment of execution.

The algorithm can be stupid. “Buy every Monday, sell every Friday” is a quant strategy. So is “buy stocks whose 20-day moving average crossed above their 60-day moving average.” So is a neural net trained on 200 features of order book data. The defining feature is that the rules came first and got backtested before any real money moved.

What it isn’t (let me be direct)

It’s not magic. There is no secret formula. No “the quants have an edge you don’t.” Most quant funds underperform the S&P 500 over a 10-year window. The few that don’t are running infrastructure that costs tens of millions a year and teams of 50+ PhDs.

It’s not “an AI that trades for you” the way Reddit thinks. A Large Language Model cannot generate alpha. It can summarize earnings calls, sure, and that’s sometimes useful as a feature in a model, but the model itself is the quant part, not the LLM. Anyone telling you they’re “running GPT-4 on the markets” is either selling a course or describing a sentiment-score feature, not a strategy.

It’s not guaranteed. Past performance does not predict future results. Backtest overfitting is real. Regime change is real. If someone promises you returns, they’re selling.

It’s not a get-rich-quick scheme. The math is brutal.

The four flavors that actually work

Most strategies you’ve heard of fit into one of four buckets. Some funds blend them.

Systematic trend following is the oldest. Buy what’s going up, short what’s going down, across futures markets. Managed futures and CTAs live here. Works because trends persist, eventually, even though most of the time they don’t. Long time horizons, low frequency, often called “the simplest strategy that still works.”

Statistical arbitrage is pairs trading on a budget. Find two securities that historically move together, bet on the spread reverting when they diverge. Pure mean-reversion. Hard to scale, brittle to regime shifts, but elegant.

Factor investing is buying baskets with shared characteristics: value (cheap stocks), momentum (recent winners), quality (high ROE), size (small caps), low volatility. Academic roots in Fama-French (1993) and Carhart (1997). It’s the reason your robo-advisor owns “Value ETFs.”

ML-driven alpha is what people actually mean when they say “quant” today. Use gradient-boosted trees, neural nets, or transformer models on engineered features to predict returns. This is where the work is, and where most retail attempts fail. More on that in a minute.

Why bother? Scale, discipline, removing emotion, backtestability

Four reasons people go systematic:

1. Scale. A human can watch 5 stocks. A script can watch 5,000. If your edge is small, you need volume to make it matter.
2. Discipline. You’ll never override the rules at 2 PM on a Tuesday when your $40k position is down 15% and your gut says “cut it.” The rules don’t have a gut.
3. Removing emotion. Revenge trading, FOMO, panic selling: these are the actual killers of retail P&L. Code doesn’t revenge trade.
4. Backtestability. You can run a strategy over 10 years of data in 30 seconds. You can run it over 100 parameter combinations overnight. A discretionary trader gets one shot per setup per year.

Why most retail quants lose money

If it’s so easy to backtest, why isn’t everyone rich? Because the four horsemen of quant failure:

Transaction costs eat your edge. We’ll do the math in a second.

Overfitting to backtests. You tested 200 momentum variants. The one that worked best on 2010-2020 data is the one most likely to be noise. The literature calls this “multiple testing bias” (López de Prado, Advances in Financial Machine Learning, 2018, Ch. 11–12).

Capacity limits. A 0.5% edge on $100k of illiquid micro-caps is real money. The same edge on $100M is gone. Your own orders move the price against you. Most retail strategies hit this wall if they ever work.

Ignoring regime change. The 200-day MA crossover worked beautifully in 2009-2019. It got crushed in 2022. Markets change character. Strategies don’t adapt unless you make them.

The arithmetic is brutal

Consider a realistic scenario. You’re a retail trader with $100k. You find an edge: 0.3% expected return per trade. You trade 250 times a year (about once per trading day). Sounds great, right?

That’s a 75% annual return. Phenomenal. Why isn’t everyone doing this?

Because transaction costs. On US equities, retail pays roughly 0.05% round-trip (commissions + half-spread + slippage on a small order). On 250 trades:

Net P&L: $75,000 −$12,500 = $62,500. Still great.

But that assumed you actually have a 0.3% edge. If your edge is smaller (say 0.1% per trade, which is more realistic for many strategies):

Subtract $12,500 in costs and you've got $12,500 left. That’s a 12.5% return after costs. Not bad. But if your edge is 0.05%, a number that sounds tiny but is actually plausible for many “I backtested this” strategies:

Costs: $12,500. Net: zero. You broke even after fees. And that’s before taxes, before slippage, before the bad fills you’ll actually get.

This is why edges are tiny in practice and execution is everything. Chan (Quantitative Trading, 2008) makes the broader point that most academic alpha signals are below 0.1% per day. The chapter attribution to Ch. 1 is approximate, since the claim is a recurring theme of the book rather than a single quoted passage. The whole game is not having a worse edge than your costs.

51% a thousand times beats 90% three times

A useful mental model has saved me from a lot of bad trades:

I’d rather have a 51% edge executed 1000 times than a 90% edge executed 3 times.

Why? Because the 90% edge happens so rarely that the variance kills you. Three trades means three outcomes, and any of them could go wrong. The 51% edge, repeated many times, lets the law of large numbers shrink the variance around the expected value. The expected number of wins is 0.51 × 1000 = 510 — a thin edge compounded over many independent trials, not a streak of certainties.

A common misread: 0.51^1000 ≈ 0 is the probability of winning every single bet, which is essentially zero. That tells you almost nothing about whether the strategy compounds wealth. What actually matters is whether the expected log growth is positive:

where μ is the per-trade expected return and σ is the per-trade return standard deviation. For repeated 51% bets to reliably compound, you need three conditions the simple model hides:

1. Positive expected payoff per bet. 0.51 × win − 0.49 × loss > 0. If wins and losses are symmetric, the 51% edge does compound. If losses are larger than wins (a “bets that pay $1 to win $1” framing is wrong if you’re risking more than you can make), it doesn’t.
2. Independence across bets. Sequential bets in the same strategy, on the same asset, in the same regime, are correlated. Variance shrinks slower than 1/n when bets cluster, so the “large numbers” intuition understates the path.
3. Finite bankroll and ruin risk. Even with a positive edge, a long streak of losses early on can wipe you out before the edge shows. Position sizing and a max-drawdown cap are what turn “positive expected value” into “reliably compounds wealth.”

So the right framing is: repeated small edges are valuable because the expected value converges and the variance shrinks with sample size — not because any particular streak is guaranteed. You get rich slowly and reliably instead of broke quickly and excitingly.

This is why quants obsess over sample size, not signal magnitude. A noisy signal you can trade every day is more valuable than a clean signal that fires once a quarter — provided the payoff ratio, bet independence, and position sizing are honest.

Career paths if you go deeper

If you keep going, the industry looks like this:

• Quant researcher: develops the alpha. Math, statistics, finance PhDs. Reads papers, runs experiments, ships signals.
• Quant developer: turns the researcher’s notebook into production code. Strong engineering, knows Python/C++/Rust, owns the backtest infrastructure.
• Quant trader: runs the book live. Watches for breakages, monitors risk, intervenes when models misbehave. Often a researcher who went client-facing.
• Risk manager: the adult in the room. Says no. Tracks exposure, drawdowns, leverage. Boring until it isn’t, then indispensable.

Most quant shops expect you to be fluent across at least two of these. The hardest skill to hire for is the researcher-developer hybrid who can both find alpha and ship it without bugs.

Read this if you want to go deeper

• Chan, E. (2008). Quantitative Trading: How to Build Your Own Algorithmic Trading Business. Wiley. ISBN 978-0470284889. Chapter 1 is the best one-page defense of systematic trading I’ve read.
• López de Prado, M. (2018). Advances in Financial Machine Learning. Wiley. ISBN 978-1119482086. Chapter 1 frames the multiple-testing problem that breaks most retail quants.

Next up: the Python stack you’ll actually use, starting with the 20% of numpy and pandas that does 80% of the work.