Re:Ask HN: Trading Algorithms
by Kjeldahl on Thursday June 17, 2010 @05:04AM
(#1438331)
For API connectivity and instruments, few beat Interactive Brokers, at least for the price offered. As for strategies, avoid walking into the complex indicator traps (combining indicators x,y,z to make a new super indicator). Stick with the simple stuff for trends etc (moving averages), and try to combine with other stuff, possibly other data (from other markets, or completely different sources). And forget about having a model that runs unattended on its own while you're at the beach. Markets change and you should stay on top of it, adapting the models for whatever the markets demand (watching out for sell-offs the last couple of trading hours is one recent theme). And don't forget about money management. Good luck!
[ Parent
| on HN ]
Seconding interactive brokers for this type of experiment.
For a reasonably accessible, yet complete introduction to basic trade design, 'Pairs Trading' by Vidyamurthy is quite solid. The strategy may well be picked clean, but the book does a commendable job in exposing the general process of finding edge and minimizing slippage.
Question: Why combining indicators doesn't work?
The curse of dimensionality, I'd guess.
Overfitting, curse of dimensionality, etc. As in many machine learning tasks, extracting good starting-point features is very very important, and that often requires domain-specific knowledge.