So You Want to Be a Quant
What is the sound of one hand clapping?
I recently did a two-part podcast series (Part 1, Part 2) with Simon M. of The Algorithmic Advantage, which inspired me to write this post.
The Byrds had a song: “So You Want to Be a Rock ‘n’ Roll Star.” The advice was straightforward. Get a guitar. Learn to play. Find some people who’ll tell you you’re great. Then get ready to lose everything that made you interesting in the first place.
The advice for aspiring quants is similar. Except for the guitar.
If you want to trade quantitatively — if you want to build strategies, manage risk, understand markets at the level where you can actually make money — you need to learn to program.
Not because you’ll write much code. In the age of AI, the code increasingly writes itself. I can describe a strategy in English and have working Python in minutes. That part of the job is rapidly going to zero.
No, you need to learn to program because programming is the only reliable way to learn computational thinking. And computational thinking is the actual skill.
What do I mean by computational thinking? Stephen Wolfram has written extensively about this, but the short version is: it’s the ability to think in terms of processes, not just states. To ask not “what is the answer?” but “what is the procedure that generates the answer, and what are its failure modes?” To understand that a system can follow perfectly deterministic rules and still produce behavior you cannot predict without running it. (If that last bit sounds familiar, it’s because it’s computational irreducibility — the central concept in my book and arguably one of the most important ideas in modern science.)
Wolfram would add — and he’s right — that you should learn a high-level language like Mathematica or the Wolfram Language, where the abstractions are close enough to thought that you’re expressing what to compute, not wrestling with how to allocate memory. The less distance between the idea and the code, the more the programming teaches you to think. But the specific language matters less than the main point: learn to program. Period.
A physicist thinks about what’s true. A mathematician thinks about what must be true. An economist thinks about what’s optimal. A computational thinker asks: what happens when you actually run it?
That question — what happens when you actually run it? — is the question that separates people who understand markets from people who have theories about markets.
Theories are beautiful. Backtests are seductive. Models are comforting. And then you run it. With real money. In real time. Against other agents who are adapting to you as you adapt to them. And you discover that the map is not the territory. That it was never the territory. That the entire skill of trading is navigating the gap between the two.
Hacker Wisdom
The hacker culture at MIT figured this out decades ago. Not about trading — about computing itself. They encoded their hard-won insights in a peculiar literary form: koans.
If you haven’t read the koans from the MIT AI Lab, you should (be patient, the page takes time to load!). They were composed by Danny Hillis (who would later build the Connection Machine) and concern legendary figures like Marvin Minsky, Gerald Sussman, and Tom Knight. They’re short. They’re funny. And they contain real technical insight wrapped in Zen Buddhist packaging.
Here’s my favorite. A novice was trying to fix a broken Lisp machine by turning the power off and on. Tom Knight, seeing what the student was doing, spoke sternly: “You cannot fix a machine by just power-cycling it with no understanding of what is going wrong.” Knight turned the machine off and on. The machine worked.
This is not a story about hypocrisy. It’s a story about the difference between doing something without understanding and doing the same thing with understanding. Knight power-cycled the machine because he understood what was going wrong. The novice power-cycled it instead of understanding what was going wrong. Same action. Completely different epistemology.
Trading is full of exactly this distinction. Here’s one.
The Master Trader and the Losing Position
A novice was adding to a losing position in a biotech stock that had dropped 20% on an FDA delay.
The Master Trader, seeing what the novice was doing, spoke sternly: “You cannot fix a losing trade by making it bigger with no understanding of why it went against you.”
The Master Trader added to the position.
The position recovered.
Jim Morrison opened “The Soft Parade” by screaming: “You cannot petition the Lord with prayer!”
You cannot petition the market with backtests. But you must do them — just as the devout must pray. The key is to understand what you are doing when you backtest. A backtest is not a petition. It is not “I ran the numbers, therefore the market owes me this Sharpe ratio.” A backtest is a diagnostic — it tells you something about your assumptions, not about the future. Confuse the two, and you are praying for profits.
Some Trading Koans
In that spirit, I offer more koans about the quantitative trading arts. The Master Trader is a figure of longstanding reputation. The Quant is a recent PhD. The Risk Manager is the person who works either for the regulators, in which case he is worse than useless, or for the fund, in which case he is merely useless.
The Quant and the Backtest
A novice came to the Master Trader and said: “I have backtested my strategy over twenty years of data. It has a Sharpe ratio of 3.2 and never had a losing year.”
The Master Trader said: “How many parameters does it have?”
“Only seven,” said the novice. “Each one is necessary.”
“And how many did you try before settling on these seven?”
The novice was silent.
“With enough parameters,” said the Master Trader, “I can fit an elephant. With one more, I can make it wiggle its trunk.”
The novice said: “But this is not a curve fit. I have an economic rationale for each parameter.”
“Ah,” said the Master Trader. “Then you found the rationale before or after you found the parameter?”
The novice was enlightened.
The Risk Manager’s Koan
A student asked the Risk Manager: “What is the worst loss our portfolio can sustain?”
The Risk Manager said: “Our Value at Risk is two percent.”
The student said: “Then we cannot lose more than two percent?”
The Risk Manager said: “No. I said that is the Value at Risk.”
“What is the difference?”
“Value at Risk is the amount you will lose on an ordinary day when you happen to lose money. It tells you nothing about extraordinary days.”
“Then what protects us on extraordinary days?”
“Nothing that can be expressed as a single number.”
The student said: “Then what good is the number?”
The Risk Manager said: “It satisfies the regulators.”
The student was enlightened.
Drescher and the Toaster (for a Trading Desk)
A disciple of another sect — a wealth advisor — once came to the Master Trader as he was reviewing his morning positions.
“I would like you to fill out this risk tolerance questionnaire,” said the outsider, “because I want you to manage your portfolio appropriately.”
The Master Trader took the questionnaire and fed it into the shredder, saying: “I wish the shredder to have the appropriate risk tolerance, too.”
Debug the Strategy
A Quant came to the Master Trader in great distress. “My strategy worked perfectly for three years and has now lost money for six consecutive months. I have checked the code line by line and there are no bugs.”
The Master Trader said: “The code is not the strategy.”
“What do you mean? The code implements the strategy.”
“The strategy,” said the Master Trader, “includes all the other funds running the same code.”
The Quant was enlightened.
The Nature of Alpha
A novice asked the Master Trader: “Where does alpha come from?”
The Master Trader said: “From other traders’ mistakes.”
“Then what happens when the other traders stop making that mistake?”
“Then your alpha was never yours. It was merely on loan.”
“How do I find alpha that is truly mine?”
“You are asking the wrong question,” said the Master Trader. “Ask instead: whose mistake am I?”
On Simplicity
A young Quant, freshly arrived from a machine learning program, presented his model to the Master Trader. It used an ensemble of gradient-boosted trees, a transformer for sequence modeling, and a graph neural network for cross-asset dependencies.
The Master Trader examined it carefully and said: “What does the model do when the Fed raises rates by 75 basis points for the first time in 28 years?”
“It has never seen such an event in training,” admitted the Quant.
“And what does your model do when it encounters something it has never seen?”
“It extrapolates from the nearest—”
“It guesses,” said the Master Trader. “Your model is a very expensive way to guess.”
“Then what should I use?”
The Master Trader showed him a chart with a single moving average on it.
“That’s it?” said the Quant. “A child could do that.”
“Yes,” said the Master Trader. “And it will work on the day nothing else does, precisely because a child could do it.”
Moon Instructions (for a Trading Desk)
A junior trader came to the senior trader and said: “I have found an arbitrage. The ETF is trading at a discount to its net asset value. I will buy the ETF and short the underlying basket.”
The senior trader said: “What is the carrying cost of the short?”
“Negligible.”
“What is the redemption mechanism?”
“I am not an authorized participant.”
“Then who will close the discount?”
“Market forces.”
“You have described a trade that requires someone else to do the work and give you the profit,” said the senior trader. “This is not arbitrage. This is hope.”
The Empty Order Book
A novice was studying the order book, watching bids and offers flicker and vanish.
“The book is full of lies,” he said to the Master Trader. “Ninety percent of these orders will be cancelled before they execute.”
“Yes,” said the Master Trader.
“Then how can I make decisions based on the order book?”
“You cannot,” said the Master Trader. “But neither can anyone else. And that is the information.”
Sussman Attains Enlightenment (Hedge Fund Remix)
A Quant was training a reinforcement learning agent to trade equities. The Master Trader came to him as he sat before six GPUs.
“What are you doing?” asked the Master Trader.
“I am training an agent to maximize Sharpe ratio. It has already learned to avoid drawdowns, size positions, and hedge tail risk.”
“Interesting. What happens if you reset the weights to random?”
“It would have to learn everything again from scratch.”
“And how would you know when to reset the weights?”
“I wouldn’t reset them. The agent has already converged.”
“But the market has not,” said the Master Trader.
The Quant was enlightened.
The Meta-Koan
Here’s what all ten koans have in common: they’re about the gap between the map and the territory.
The backtest is a map. The live market is the territory. The VaR number is a map. The tail event is the territory. The risk tolerance questionnaire is a map. The decades of scar tissue are the territory. The code is a map. The strategy-plus-competitors is the territory. The order book is a map. The intentions behind the orders are the territory. The converged model is a map. The non-stationary market is the territory. Averaging down is a map. Whether the thesis survived the drawdown is the territory.
And this is exactly what computational thinking teaches you.
When you learn to program — really learn, not just memorize syntax — you develop an instinct for the ways that formal systems diverge from the reality they’re supposed to model. You learn that code compiles but doesn’t work. That tests pass but the software fails. That the specification is not the system. You learn, at a visceral level, that running it is different from reasoning about it.
That instinct is worth more in trading than any amount of mathematics. Not because mathematics is wrong — the math is essential. But because mathematics tells you what should happen in the model, and computational thinking tells you all the ways the model is not the market.
The market is a computation that cannot be shortcut. You cannot derive what it will do from first principles. You cannot backtest your way to certainty. You cannot model your way past regime changes. The only way to know what happens is to run it — with real money, in real time, against real adversaries.
If that sounds like computational irreducibility, it’s because it is.
The Master Trader is not smarter than the Quant. The Master Trader has merely been enlightened more expensively.
For more on computational irreducibility and why even deterministic systems resist prediction, see my book The Science of Free Will. For the original AI Koans that inspired this post, see the Jargon File. For the rigorous version of why your drawdown rules are probably hurting you, see my paper “The Stop-Loss That Stops Gains.” And for a longer conversation about these ideas, see my two-part interview with Simon M. of Algo Advantage: Part 1 and Part 2.
On the rationale-before-or-after question — there are two distinct overfit modes worth separating.
Parameter overfit: fix the strategy class (say, mean reversion with VIX gating), tune the parameters until the backtest looks good. Standard concern, standard fix.
Architectural overfit: you tried mean reversion AND momentum AND breakout on this data, picked mean reversion because it worked, then tuned its parameters. The class itself was the overfit, and retuning parameters won't surface it.
The post-hoc rationale your Master Trader interrogates is the move that smuggles the second kind in under cover of the first. The strategy "has a reason" — but the reason was found on the same data that selected the strategy. Naming this as a separate axis surfaces it; leaving it implicit lets it pass.