41–50 of 77 posts
I'd recommend you read Quantitative Trading by EP Chan for a great overview of the basic methods and some example implementations.
Disclaimer: I've done exactly this successfully for >2yrs now.
Start with a few questions. How often are you targeting trades? Do you anticipate trading thousands of times per day, a few times per day, a few times per week, a few times per month or a few times per year? Know that there are sophisticated algorithmic trading systems in all of those spaces.
Once you know how often you will be trading, you can make more appropriate decisions about execution platforms. What kind of data feeds do you want to use? How will you get the data into your systems, how will you get the outputs out of your system. How will you translate those outputs into orders? Who will take your orders? As an aside, most brokerages will want to know if a computer is making all the trades and will either outright ban fully automated trading or will want a kill switch (see the TSA for your broker of choice).
This leads us to the actual hardest part about automatic trading, operational risk. What are you going to do to prevent your system from making out trades? What is your shut down procedure? How is it triggered? How will you monitor the system? Know that managing automation risk is the last mile of human intervention in the system, and that lots of people have lost a lot of money by not intervening soon enough, or intervening too soon.
You also need to have a target benchmark to compare to. What are you going to benchmark against? How often are you going to evaluate against that benchmark? How long will you allow your system to underperform the benchmark? There are long game algorithmic trading systems that have built into their capital allocation models >10 year underperformance against benchmarks for instance.
Finally, ask yourself, what inefficiency are you taking advantage of, that will allow you to profit from trading? What is it that you bring to the table that isn't already there? Once you honestly assess that, ask why wouldn't that same efficiency be more valuable in some other field that is less crowded because automatic stock trading is a very low percentage game right now.
I used to do it with bitcoin, but not to get rich, just for curiosity. The most algorithms is just chart reading, playing around with periods, standard deviations etc. You can achieve something really quickly, but it only works if the market jumps up and down within a short time period and stays on the same level if watching a bigger time frame. What if the market drops by 90% suddenly because some bank decided to do…
What most algorithmic trading models that are looking to prey on other algos do is either wait until their model says the market is already in some bad decision state or move the market in some way to make the bad decision state more likely.
Someone can please give me some trustworthy, scientific reading material to give people asking me to proof the point 'systematically beating index performance on the stock market, using only public available data, is not possible'? Intuitively i understand this is true, so no need for holywar. Same also applies to commodities trading.
"A mathematician plays the stockmarket" is an interesting read of a smart person who made a bunch of mistakes and lost money. http://www.amazon.com/gp/aw/d/0465054811/ref=mp_s_a_1_sc_1?q...
Don't get discouraged by the negative comments. It's completely possible to make a profit (with low risk) without resorting to HFT. I'd recommend you read Quantitative Trading by EP Chan for a great overview of the basic methods and some example implementations. Disclaimer: I've done exactly this successfully for >2yrs now.
I have been building also my own platform (with nodejs as for high IO), but you easily loose focus when you have to build it ALL by you're self..
(opensource) platforms like MT4 / MT5 give you more time to spend on actual alghorithmes, then on data structures etc
I once set down this path but stopped. Your big problem is that the future market doesn't look like the past market. There is a statistical error inherent in excessively curve fitting past data. You could develop a system which says you'll return 25% per year and then lose 100% in he he first six months. So don't fall into that trap of excessively back testing. What you should be spending time is your money allocatio…
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.…
This type of approach is far better than trying to build your own, unless, 'your own' has an edge you have come up with. I stopped when I realised I had no edge, no unique method or insight.
I still believe you could still develop a neat system that deprioritised the entry signal - and focussed just on letting winners run and cutting losses. That requires a very low cost transaction model however.