Machine learning the market with the SVM winning

(Last Updated On: October 28, 2017)
I find quite surprising on how the Financial Times has actually done some analysis. This report was on how Machine learning does in the markets. So there was a test by SocGen which is a French bank. There all different types of Analysis types processed including Nonlinear SVM.

Click here for the FT article 

These benchmark models simply use an equal weighted average of the factors long of the top quintile vs shorting the bottom part of the stocks based on their average return. This is quite normal when it comes to classic Arbitrage in a Paper trading account. So obviously, there is non-linear support vector. Do you agree that this is best algo type machine learning. How do you find these algorithms do?

So it’s difficult to explain what is going on inside of machine learning algorithm. From this article, there were some tips,

One trick is to mathematically add an extra dimension to data until you see a clear linear separation can be created. So more and more dimensions need be added before the linear fit can be found. As a result is not possible to visualize many of the steps involved in.

Calculating the interpretation of the results can be very challenging. So overall what they’re saying is. this machine learning algorithms can do pretty well in general. However these machine learning stumble in 2016 and 2017.

So when you’re using SVM might do better than linear because certain factors might be the cause that high scores are good for higher performance but low scores are not bad. Linear patterns in your model cannot count the non-linear SVM Algorithm.

About four or five years ago there was a test put on by Dr Jonathan Kinlay with a friend of mine. They tested these algorithms but what was summarized was that support vector machines only 50 percent accuracy in its forecasting. Now unless something’s changed over the years, I am not interpreting this overall correctly when you drill it all down.

As this FT article states, 2016 was kind of tricky. Even 2017 was a market you could argue that the market is not driven by fundamentals. So that would tell you that the markets are driven literally by other algorithms out there. Because there’s 80 percent of the market is driven by algos,. This maybe not the best number for a model to train to detect historical patterns and fundamentals. So as they say, past performance is not indicative of future.


We are coming down to the last 24 hours of this Portfolio Optimizer source code walkthrough. This most likely I will be talking about it as this video does vanish within 24 hours.

Only my Quant Analytics member will have access to this.

Join here for that.

NOTE I now post my TRADING ALERTS into my personal FACEBOOK ACCOUNT and TWITTER. Don't worry as I don't post stupid cat videos or what I eat!

Subscribe For Latest Updates

Sign up to best of business news, informed analysis and opinions on what matters to you.
Invalid email address
We promise not to spam you. You can unsubscribe at any time.

Join our Private Chat Server for FREE Trading signals for ETH DOGE ADA

Scroll to Top