I want to clarify, the intention of this system is specifically to NOT be a HFT system because I'm aware of the large handicap associated with inferior data available at slower speeds. It's intended to primarily help with stock picks since it can look through more data than a human can. This would involve both fundamentals and technical analysis. The most frequent trades would be once a day and may end up being closer to once a month. It is intended to be a low-cost solution. I would like the trades to be done by the system but the approach will not require it.
Based on some of the comments it seems such a system will be the underdog and will be the prey for the large HFT shops. And that's one of the risks of such an experiment. Once I put in real money into this (not too much, but just enough to not be consumed by the fee) there will be a hard point after which I'm ready to throw in the towel. I think of this more like an investment management software that has more knowledge than myself (various trading techniques, tax laws, etc.).
Yes. That is a massive handicap that the system will have (it's not going to be co-located and probably won't have access to early information, etc). I'm wondering whether it's possible to develop a system that accommodates these deficiencies while still ending up on top.
Thanks for the advice. I will definitely err on the side of caution and not get caught up with the backtesting and over fitting. Allocating resources (funds) will probably play an important role to manage risk in the system.
At the end of the day, I'm trying to answer the question: can you make money in the stock market with a low-touch (eventually no-touch) approach with the help of advanced software and analytics. Hopefully the answer is yes ;-)
Yummyfajitas, that's exactly why I do not want human input. Also, I'm not trying to make money on every bet, but to make more money in my winning bets (less fees) than I lose in my losing bets.
That's very interesting. I'm trying to integrate the concept of "special events" like earnings or dividends or announcements into the mix. Although that isn't the driving force for what I'm trying to do. I'd be interested to learn how it works out for you guys.
I had actually looked into quantopian when I was getting started, but decided not to use that for my primary source of backtesting because they place limitations on users since that reduces pressure on their systems, allowing them to scale. This restricts backtesting ability and was a dealbreaker for me. The publicly shared algorithms, on the other hand, seem like they can provide a lot of insight, and I will look into those. Thanks!
I haven't based this off of any existing papers yet, I'm intending for it to start off as fresh research, and depending on how it goes, maybe include/test theories from existing papers. Do you have any papers that you would recommend?
I haven't read the book. From the description on Amazon it seems the investor/mathematician let his emotions come in the way of his investment decisions which led to him losing money. I think that's a risk we run by providing that human touch, which sometimes may even help us if the software is making stupid decisions.
My goal is to build a fully-automated system (no human interference) that makes "confident" decisions based on statistics and hard numbers.