LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.
A useful quant shelf is a set of labs: trading, modeling, and a few books that ignore markets entirely. The lecture's list is short on purpose. Leave the page, try the idea, see what fails, and come back. Reading every chapter before you build anything is the habit the list is meant to stop.
Why it matters
Desks punish shallow reading. Natenberg's option book is the one described as living on a trader's desk, because a first option trade can make money, here roughly $700 about ten years earlier, while the trader still cannot name the exposures behind the gain and loss. In machine learning, cleaning data and choosing features are jobs of their own. A book is the classroom half.
If the research interest is equity signals from alternative data, you also need a door into language processing and a short account of transformers. Linear algebra sits under pricing and under learning models. Thin linear algebra kept showing up in student meetings.
The idea in plain words
Trading comes first. Sheldon Natenberg's Option Volatility and Pricing introduces pricing and strategies, tied to that early trade. Two further books shaped the view of quantitative trading. One is Marcos López de Prado's Advances in Financial Machine Learning. The other is Inside the Black Box. The first is heavy. It assumes math, probability, statistics, and some machine learning, and it is recommended only with that background. Two remarks stay attached. Another trader said people have gotten rich off an R-squared of 0.02. A former employee said an R-squared above 0.15, unless it is a mistake, is a reason to walk away, with a joke about regulators. The point to keep is suspicion of a glamorous in-sample fit.
Modeling is the second stack. Aurélien Géron's hands-on scikit-learn and TensorFlow book, and François Chollet's Deep Learning with Python, cover the path from raw data to a model you can run. Chollet is singled out as the strongest Python deep-learning book in this reading. A getting-started language-processing book is the on-ramp for text, images, articles, and text pulled from images. The signal comes from those sources, not from a model aimed at the price path. A hundred-page language-model book is the short door into transformers. ChatGPT, Gemini, and Claude make that architecture hard to skip, and generating text is a different task from scoring sentiment.
Linear Algebra Done Right is included from necessity. Pricing, simulation, and dimensionality reduction are matrix work. The described habit is to reopen it for problems when the background feels thin.
Pricing gets two heavier mentions, both with prerequisites. The first book read in that area was Pricing and Hedging Financial Derivatives. The favorite textbook is the one on mathematical modeling and computation in finance, with Python and MATLAB, used when you need a characteristic function or a discretization. Three Stoic books close the list and are explicitly not finance: Ego Is the Enemy, Meditations, and Epictetus's Discourses and Selected Writings. The nutshell is to attend to what you can control. Holiday's book followed an earlier Stoic read about seven or eight years before the lecture. Discourses is described as reread past counting.
How to use it
Match a category to the job in front of you. If you cannot explain an option's gain and loss, start with Natenberg and reconstruct one position. If you already have the math, use López de Prado while you build, and audit any very high R-squared. For a data pipeline, stay on one Géron or Chollet chapter until you can acquire data, make a feature, fit a model, and say where it failed.
For alternative data, score a document before you chase chatbot demos. If a proof is opaque, do linear-algebra problems before adding another library. For derivatives, look up the function, implement it, and compare it with a simulation. Reopen the Stoic books when a project becomes a story about being right.
What usually goes wrong
Finishing the shelf before building is the main miss. Opening the heavy machine-learning book with a thin background, then blaming the field, is the second. A chatbot is a poor substitute for a pipeline: it can discuss a sentence and still never tie a corpus to a name or test a signal. Skipping linear algebra shows up later as stuck pricing code. Treating the Stoic titles as trading systems is a category error. The R-squared remarks are other people's comments, not a target and not a legal finding.
Questions people ask
Which options book is first here?
Natenberg's Option Volatility and Pricing. It introduces theory and strategies, and it is meant to sit beside positions whose exposures you can name.
Is López de Prado a beginner text?
Only with a strong technical background. The lecture calls it heavy and keeps Inside the Black Box beside it on the trading shelf.
Why is linear algebra on a finance list?
Pricing and dimensionality reduction are matrix work. Student meetings kept revealing a thin background, so the Axler book is practice, not decoration.