Artificial Intelligence Applied to Stock Picking
The Herald Tribune has an interesting article written by Charles Duhigg about using artificial intelligence in the stock market. Among others, the author has interviewed Ray Kurzweil, a hedge fund manager involved in artificial intelligence.
In this article, as it is often the case when applying AI in finance, the two buzz words are "neural networks" and "genetic algorithms". Nothing about decision trees and support vector machines, for example. Maybe SVM is too recent to be yet applied in finance. But what about decision tree? Not trendy enough?
If you read the article, you will notice that, the author highlight one important drawback of such techniques: "black box-ness". This is true, but another drawback is certainly more important: overfitting. Technical analysts may manually overfit their data when predicting future trends. However, it is much more "easy" to overfit the data when using data mining techniques.
Link to the Herald Tribune article
5 comments:
SVM too recent for stock picking? I'd be shocked...
It is certainly used by a bunch of people, but it is clearly not as spread as GA or ANN for example.
Usually, one can see 10, 20 or even more years delay between academic research and large industry applications.
Hi Sandro,
Very good post Sandro. I have found that Neural Networks severely overfit data and perform very poorly on unseen cases.One of the best performing models i have seen for stock prediction is (although not a decision tree)the 'humble' Logistic Regression...Tree models are very good for showing 'what is going on' and for extracting 'algo trading' rules
Simple is beautiful :-)
However, I also think that techniques such as decision tree are limited, especially in finance, due to their "linear" relationship discovery.
Thoughtful blog thanks for sharing
Post a Comment