Hello to all fellows. I am a newbie in automated trading and it's my first thread here. I am from Hong Kong. Just curious.. any Hong Kongers here ? I found that most automated trading related books were lent from libraries but nobody discuss it in Hong Kong forums. I have been doing discretionary part-time FX/futures trading for nearly 1 year but I found that I am really tired of it since I have to do a lot of manual scanning of opportunties and wait for position opening and closing. In this mode, it really waste lots of chances esp for Gold futures, where many important TA points reached at electronic trading session... I am probably sleeping as it is 2am to 8am on my side... However, I dare not to just place a limit order since there are various graphical interpretions. When I discover this forum, I am really surprised as I noticed that there are tons of people already implemented an automated solution that I thought it ought to be very rare. However, I do believe that the automated trading size must not be small. The evidence is that in my discretionary exp, I notice, for exp, all currency pairs move in nearly perfectly synchronized mode, like GBP, EUR increase the same pips over USD .. generating the same graph in the short run esp right after the data release time. It seems most investment banks trade FX in an automated manner. Sorry to say so much irrelevent stuff with respect to the topic. Here comes my idea: In my concept for having an automated solution, there are basically these steps: 1. find a strategy 2. implement the strategy 3. back-test the strategy 4. forward-test the strategy with the demo account 5. go live ! I am a computer science graduate who do not know too much quantitive finance but it seems that I dont even need such knowledge ?! I can permutate the solutions until I find a way proved to be profitable. In the simplest form, I can just "steal" the idea from the winning guys, for exp, analyzing the trading records from the winners of FXCM's mini account competition to see what signals they take in for opening/exit like 10 day SMA cross over 50 day SMA for opening a position. Is it too simple ????? How long does it take for you fellows to get a profitable algorithm ? Also, how long does it take to test a strategy, will the running speed very slow when applying on several years of historical data ? Thanks in advance !!
i'm too tired to read all that long story short, yes, the simpler the better - less chance of curve fitting and more application to a broader spectrum. All the strategies I use at my private fund and when i worked at a managed futures firm were very basic in nature - moving averages, channel breakouts, volatilities, that's about it. Nothing fancy or any hidden secrets. take something that's standard, tweak it a little bit to your purposes, and run with it
Agree. Most profitable strategies are: (1) Only moderately complex. (2) Exploit a market inefficiency. (3) The market inefficiency can be rationally explained. (4) And 90% of the ballgame... the strategy is professionally executed in a very discplined manner.
You will find alot more info here: http://www.wealth-lab.com/cgi-bin/WealthLab.DLL/getpage?page=Top25.htm Good luck
Backtesting: Interpreting the Past http://www.investopedia.com/articles/trading/05/030205.asp Trading Systems Coding http://www.investopedia.com/university/systemcoding/
Thanks for your experience! You all are so generous! 99atlantic: As you have ever worked with futures, have you ever noticed "The Encyclopedia of Commodity and Financial Spreads" ? It has lots of quantitively proved profitable trades (with entry and exit dates) HoundDogOne: Your point 3 really enlighten me. When I read through the book mentioned above I quickly think that might be curve fitting but I notice that is seasonal behaviour of the agricultural futures .. seems rationally justfied right. I will follow it but I dont need to program it since the period for such futures last for months. potemkinvillage: Your link is absolutely great !! Have you ever tried any of it ? Is there any algorithm trade on minutes timeframe and still backtested to be very profitable ? How about scalping system ? but for scalping it seems that we need to programme on FIX/sockets levels and use a large trading volume (for an advanced account) to minimize the spread as well kiwi777: Your links are useful..especically the point over optimization leads to curve fitting. Besides, anybody tried Tradecision with its neural network abilities ? I had ever scanned a book called "The Ultimate Technical Trading Software" that somebody has the experience of blowned up by the neural network .. since it is too complicated and uncontrollable. As you all fellows all say the simpler the better.
6. Put a large sum of money into your trading account. 7. Watch it all or mostly disappear. 8. Report back to ET on one of the 'How I jacked away all my money' threads.
during 2005 i was working on a 4 parameter trading model that was optimized on a set of data and then walked forward. the performance on the walkforwards was bad, so we realized that we were overfitting on the optimization set. part of this was that we had too many parameters. i recommend you be careful to not overfit. i've recently been working on a much simpler strategy and the results are much better . the amount of programming you'll have to do is considerable, but if your schooling is good you can manage. i could not find any places that would let you setup a test account and do automated trading through an API without putting up ~1,000 USD. if anyone knows of such a place that would allow this testing without as large an upfront cost, do let me know. if your model takes a TON of processing power, you may be working in the wrong direction...