Jan

9

Hello, Specs

A while ago I was asking directions and opinions about Artificial Intelligence and Machine Learning applied to the Markets, and the answers were quite discouraging.

Nevertheless, inspired by works such as of Marcos Lopez de Prado, (who is now Head of Artificial Intelligence at AQR, if I recall correctly), I started my Masters Degree at Computer Science, with focus on Machine Learning for trading (UFRGS - Universidade Federal do Rio Grande do Sul).

The reason I'm telling you this is because what I'm learning differs from the overall view I got from the answers to my previous query.

I now see that the answers were in line with the idea of webscrapping for "market humor", and news analysis (all int he field of Sentiment Analysis), etc.

There's no dispute that this approach is not available to retail investors such as I, nor for minor asset managers. In fact, I would argue that Bloomberg and Reuters (or any other news agency) would have dominance on that.

(The same is the case with the so-called "high frequency trading"… We simply can't compete with the major players).

But, I'm also learning that Machine Learning (and Artificial Intelligence) is much broader, and I'm starting to see its relevance for trading… Specially, automated trading.

My first model developed at the Masters used an algorithm called Support Vector Machine, which is a classification algorithm.

There are three main "approaches" do Machine Learning: Unsupervised Learning, Supervised Learning and Reinforcement Learning.

SVM (short for Support Vector Machines) is an algorithm in the category of "Supervised Learning".

That's because you give it some features to work with (and learn from). These features are our own market analysis filters, whichever they are.

The thing is, the main value in this model comes not from the SVM algorithm itself, but the "feature engineering", i.e., from "which data filters should the algorithm learn".

That's useful to any (of us) who has his own trading style developed from the experience of many years: we intuitively (or through massive research) know what would be useful or not to develop a trading strategy.

Anyway, I'm still in the process of getting this Masters Degree (in Computer Science), and it's my goal to develop a new trading model from each algorithm I learn.

If you, Specs, have interest in it, I can have you posted.

Kind Regards from Brazil,

Newton Linchen Certified Analyst (APIMEC, CNPI-T 2167).


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search