Suggestions to Improve Automated Trading...

Discussion in 'Automated Trading' started by greaterreturn, May 4, 2008.

  1. Folks,

    Do you trade automated? If not, please avoid replying. I already trade automated and want a little advice.

    I have a simple strategy that back tested well to make money without any indicators, just price action. And now it has been trading automated over a month on USD/JPY with a micro lot size without showing profit.

    So I already worked through all the technical issues so that now it runs smoothly without errors on a VPS server.

    Still, it turns out that the underlying strategy hasn't been profitable this month.

    So I'm considering whether to let this one run another month to see if it turns profitable again, consider modifying it, or taking on a new strategy.

    Most strategy ideas I have tinkered with were original discretionary strategies but it seems the auto strategies need different ideas.

    Hopefully, if you provide some useful tip or point me to a great thready I can return the favor somehow.

  2. lrm


    There are really only two options:

    (1) Continue on or pull the plug. Note I consider this the same option as it is based on emotion.

    (2) Compare the last moths of data to your only other set of data, which is your back test, and see how the live returns look statistically as compared to your back test.

    #1 should not be considered, IMO, when auto trading. You are auto trading for numerous reasons and one reason is likely to remove emotion from your decision making process. At the very least, this is a benefit from trading using automated systems.

    So, my suggestion would be #2. Analyze the last months set of data and compare the various statistics of measure against your back test. Indeed, you may want to take a look at your back test and pull from it pseudo-random time periods that look like last months price action and see how things look.

    If your live run doesn't compare statistically to your back test pull the plug. If you fall within normal operating parameters as your back test indicates, then carry on another month and repeat until you're comfortable.

    That would be my approach, at least.

    Take care!

  3. Louis,

    That was very kind of you. I did as you suggested and analyzed this month compared to last month.

    Especially, I looked at the biggest losing days and analyzed why they lost.

    They were related to news releases. So I found a simple new rule to add that makes it significantly more profitable.

    Also, I found a BUG in the strategy logic. I still need to troubleshoot that but it will improve the results even more.

    I normally always do this type of analysis. Guess emotions got to me like you said.

    Anyway, thanks for the encouragement!!

  4. Baywolf


    Are your entries and exits firing on a tick by tick basis or bar by bar?

    What other (external) variables are different from backtesting and real-time?

    Was the backtest even profitable? Try some more time periods and see if they are profitable without optimization.
  5. This strategy operates tick by tick. I use multi-time frame data in NinjaTrader. That way it shows me 5 minute data on the charts but the strategy actually operates off of 1 tick data.

    Changing time frames only impacts the chart display since the strategy trades off single ticks.

    There's no averages or other complexity. The strategy itself is brain-dead simple. It's implementing all the "plumbing" or infrastructure to manage the trades.

    Performance? In March it makes 459 pips with only 36 trades. That was the limit of my backtesting since that's all the tick data I had.

    In April it makes 240 pips with 64 trades. That's AFTER correcting the issue with news releases I found today. That's less but nothing to sneeze at.

    Note: I didn't actually make that money live because I had almost 2 weeks down time waiting on tech support to open that was blocking my live trading account. Since then it has been trading each day live.

    This month has mostly been dedicated to working out the kinks so that it behaves and follows the rules just like the historical data.

    I was hoping to let it run with a full mini lot next month but it seems I need another month of live testing with a micro lot.

    How about you Louis, what and how to do you trade?

  6. bespoke


    Dude, that's your problem right there. Only 1 month of backtested data. Spend the time and download at least a couple years of tick data. I know it takes a lot time but you NEED to do it. It took me months to download years of tick data for 100+ symbols but it was well worth it. And then use 1 year as your test data, and the other as your out of sample walk forward data. If it tests nearly as well on the out of sample data as it does on the test data you might have something there (but then I would download a few more years worth of data to confirm)

    I remember when I first started backtesting strategies I only looked back 3 months and I'd optimize the hell out of it. Then I would sit there wondering why it wasn't working in real-time. Don't make the same mistakes.
  7. Bespoke,

    Great. Thanks for the advice. I deluded myself by the number of trades that I had a statistically significant sample. But you're right. A year of data would be far better.

    I shudder to think how long that will take to process in NinjaTrader on my PC.

    It currently takes about 2 minutes and almost a gig of memory to historically test only one month.

    For a year? Whew. I lack the time right now to write a better performance back tester--although I certainly have the skill.

    What do you use as a tool for back testing, Bespoke? I'm guessing you wrote it yourself, right?

    Seems all the true ATS traders write their own stuff.

    - Wayne
  8. bespoke


    Ya, I wrote my own.

    I remember when I first started out it would take me about 12 hours to backtest my strategies on a year of data (for many stocks). But as I got better at coding, used a faster language, streamlined my data to be read faster, used simpler strategies, upgraded my computer, etc - It went from 12 hours, to 6 hours, 2 hours, 1 hour, 45 minutes, 30 minutes, 10 minutes, and now takes me only about 5 minutes to tests years of data for many symbols.

    All I can say is, just keep testing and trying. Something that really helped me out was reading all the posts by a member named "Acrary" and I'm sure many others would agree.
  9. Baywolf


    Is the NT strategy analyzer not working out for you?
  10. Wow Bespoke. That sounds like practical information. I'll read up on Acrary.

    Also, what tool do you use to chart your data to figure out what's going on? NinjaTrader probably can't handle that quantity of data you use.

    Also, I've been trolling around the net looking were to get free tick data for currencies going back several years. I'm only trading USD/JPY right now but I might as well get 17 pairs for future expansion.

    So far, the link to isn't working. That seems to be the favorite place.

    Anybody know where to get tick data to forex going back a few years?


    #10     May 4, 2008