Quant Trading & Research

You can easily spot the shitwork from the above blog post that a professional researcher would love to dump on some eager trainee. Even if there is a small army of teams set up for the QR department there is always something tedious they would rather someone else to do for them like maintenance of the research infrastructure, writing documentation, analysis of the production system so the parameters can be tweaked, or reconciliation tasks or regulatory stuff end of day if you can be trusted with that work.

Besides internships there is no paid role for QR assistant I just invented it myself by walking over to that department and asking for work to do selling myself as an eager unpaid apprentice. You will have to do all your other duties then spend your nights, early mornings and weekends completing whatever work they assign you.

Why would you offer to do this? Because then they have time to show you the magic.

This is possible only if you sell yourself as a reliable portal where increasingly complex work can be dropped and that portal returns it completed with high attention to detail. If you can pull this off they will keep you around and eventually they won't want to do their job without you as you've become so useful to them you're almost essential so they move you in to the research department with a new salary and bonus. Depending on NDAs some may even pay you themselves to directly work for them.

If you are at a fancy school and you can land an internship during the summer then if it goes well they likely will offer you a large amount of money to drop out and learn on the job. Where I am does this all the time now trying to entice kids to forget about the degree we will teach you everything you need like having your own personal professor except now you are getting paid. I am always surprised when the kids say no because they think the piece of paper means something. It's like being interested in Cosmology and the world's best Cosmologist offers to personally train you and pay you an enormous amount of money but you decide nah I'm going screw around another 2-3 years taking watered down courses that I will never need. I understand not wanting to quit something you started but in this case I would parachute out of there immediately or suspend somehow a year with a leave of absence if they allow it.

Everything you learn in undergrad has to be re-taught anyway I can list a dozen things that are promptly thrown out that they drill you with in undergrad such as Rolle's theorem (special case of the MVT and worthless), the Riemann integral (garbage), almost all the linear algebra you learned (junk) as today matrices are so massive they have to be randomized (jack polynomials). All your work is now in high-dimensional black boxes and adaptive importance sampling is used to integrate to re-weigh everything and other tricks not taught in undergrad but any freshman could figure out.

I assume you can program so all you need is the basics of data analysis on noisy but semi-structured data meaning S-expressions, JSON, Apache Parquet column-oriented files, etc.

• Probability
  • this is a workbook really not a textbook
  • look who sponsors the course for this book
• Modern Regression and Data Analysis
  • uses R but Julia/Matlab/Python exists too use whatever you want

This is truly all you need to start. No I'm not joking. You can learn statistical learning when you get there.

• Numerical Methods
  • Preprocessing everything for efficient large scale linear algebra, throwing out all of undergrad basically
  • No more gaussian elimination, no more matrix inverses
  • You don't even store a matrix anymore ie: a 2d array you just compute linear operators
  • You don't even find eigenvectors or eigenvalues anymore you approximate them with some completely different algorithm
  • Every problem is a (minimum) 10⁶ or more matrix
    • A tiny 3D system model is 100 x 100 x 100 with 10⁶ unknowns
  • You end up preconditioning everything and gradient descent is too slow because it 'zig-zags'

This I would like to go through in fact if I were to rewrite 'Math for Computer Science' that famous MIT paper floating around it would just be numerical linear algebra because that's all we're doing now.

Transformers (LLMs) are used for things you wouldn't expect like predicting 40 septillion floating point operations.

I'm personally interested in Game-Theoretic Probability where essentially all concepts of randomness are thrown out. Probabilities are now forecasts and statistical models are tested with betting and non-stochastic. If you've ever been highly skeptical of stock markets being supposedly stochastic instead of a game played by multiple participants then you'll probably want to take this too. One of the authors Glenn Shafer designed a simple betting test for statistics tossing out p-values.

• Intro to Game Theory
  • helps to understand QR lingo and strategies
    • or try The Mathematics of Poker by Chen/Ankenman

I will go through game theoretic probability here if it's possible or in the Tripos.

The craziest example of casino martingaling can be seen in the videos floating around of Dana White and Adin Ross filling up a blackjack table with markers going deep into debt and then paying them all off. Strategies where you bet more when you are losing and less (or quitting) when you are winning obscures the risk of large losses and are called martingales.

This is not limited to sports betting or casinos. Corporate executives and professional financial traders deceive themselves with martingales all the time. A good paper to read is The Martingale Index by Shafer/Dimitrov to ensure that's never your strategy as it's an easy trap to fall into and also easy to spot when someone else in the market is martingaling and you can bet against them.

Never, ever, martingale. I will include everything I can about it to prevent you from doing it because I did it myself not even knowing at first. Thankfully I was lucky otherwise disaster could have happened.