Wednesday, November 1, 2017

Machine learning sux

Still catching up on the last week's news items....

FT Alphaville - machine learning sux. Quote:

But the failure of all of the algorithms to perform well in 2016 and 2017 highlights the age-old problem with models.

“2016 was a kind of a tricky year,” says Oikonomou, noting that the algorithms “didn’t handle sharp changes in the market”.

“Even this year you could argue that the market is not driven by fundamentals,” he adds, which is maybe not the best environment for a model trained to detect historical patterns in fundamentals.

Or, just maybe, once you put enough algos into the market, all the beta gets arbed.

First iron law of the markets 1, twentysomething quants with grad school training and a working knowledge of Matlab 0.

No comments:

Post a Comment