About to go live and........ apprehensive

Discussion in 'Automated Trading' started by futurecurrents, Jan 12, 2012.

  1. Actually I'm pretty freakin scared. After years of trading and watching the markets I've now spent the last few months developing some strategies on Tradestation. I coded my ideas in and they look really good. They work long and short and the equity curve is beautiful. It's based on last year. It's an intraday system and it's about one trade per day usually a few hours long. It's trading the Q's simply because it's what I always watched. I figure if it works there I can always switch to the futures later but for right now it's where my comfort level is. It seems too good to be true but whichever way I look at it, it looks good. Worst case is slippage may reduce returns, 20% tops but the curve should still be nice. Of course past performance is no guarantee.....but.

    My plan is to initially watch it at home but eventually I want to be able to leave it so I can do my usual work. Even though I'm a lowly tradesman, I enjoy getting out and meeting people and doing my mechanical things. Besides, it actually pays the bills.

    I'm going to start slow with low BP and ramp up if it proves out.

    I just bought a new laptop today that will be dedicated to the trading. That was a huge purchase for me...psychologically.

    Part of the reason I'm so freakin nervous is that I've been down this road before....and it didn't pan out. I didn't lose and in fact made some money daytrading stocks, but it was stressful, and interfered with my real job so I gave it up. It was not automated and was totally different than what I'll be doing now. I went from dreams of big yachts, to buying a dinghy. Really. Nice dinghy though. I have this feeling that if this doesn't work out, then that's it, nothing is going to work for me. That may not be rational but...

    Anyhow, not sure why I'm writing this and boring y'all. I guess I'm being self-indulgent. I'm just so wound-up about this. I need to vent. Perhaps some of you have been here before. Can this really work? Maybe some issues I should look out for? I'm getting a UPS and intend to monitor while away via team-viewer on my DroidX.

    I guess worse comes to worst I still have my dinghy. Sanity? Who needs it?
     
  2. dom993

    dom993

    I know the feeling ... it has taken me several attempts with various automated systems of my own creation before one delivered a live performance comparable to the backtesting.

    I would suggest you start trading your strategy in a simulated account for a few months (I have no idea if this is possible w/ Tradestation). 1 year of backtesting is really not a lot (how many trades total in your backtesting ?), and if you have not tested your strategy "out of sample" (on data other than the one which you used to develop/test/optimize the strategy) you should suspect it to be over-optimized (which usually leads to poor or negative performance in forward testing / live execution).

    My other suggestion is for you to decide on the maximum drawdown you will let your system run into, before turning it off. Better do this before you commit any money to live trading, as you might realize you do not want to lose "that much" money to find out the system doesn't work.

    Good luck with it!
     
  3. Thanks. My plan is trade real small at first. Maybe I'm wrong but I feel I could test it every which till Sunday and still not be sure until I actually move forward with it.

    Your suggestion about deciding on max drawdown before turning off is a good one. If it does not continue to act like planned it's getting stopped or adjusted. One good thing is the over-all plan is composed of a number of single strats that can stand by themselves. So hopefully if one fails the others will take up the slack. Ie. the longs go south the shorts should do better.

    But damn, it feels like I'm staring into the abyss. Or I just walked into a whole new world and I barely know the language.

     
  4. heech

    heech

    Instead of buying a dedicated laptop + UPS... I'd suggest you look into a collocated solution. If not a dedicated server, then something like Amazon EC2 would be a great option. Much lower startup cost, and you only pay for what you use. You also have much greater confidence your feed/power isn't going to go down in the middle of the day.
     
  5. If you have a truckload, no, a boatload of statistics, it makes it easier.

    For an intraday system with only 1 trade per day, you'd probably need about 14 years to get statistical relevance. Then, your backtesting needs to be almost perfect with pessimistic assumptions (slippage, spread). Then you need to create a graph which will distribute the drawdowns in $ terms and in % terms during the entire period. That way you have something that should look like a normal or log normal distribution of drawdowns. Same thing for profit and loss, distribution graphs are important (and simplier to interpret than monte carlo sometimes). So when you go live you pay close attention to where your "barrage" of trades is hitting "the ground". If it's somewhere within the estimated area of the drawdown and pnl (again in $ terms and in % terms - it's important to have both because those distribution can be VERY different) then that means that your "guns" are calibrated well enough and you can feel safe. The other thing is that if you have a boatload of data you can establish statistics such as "what is the % chance that 3 loosing trades of X magnitude will happen in a row". If the % chance calculated statistically is getting too low for too long, you should be worried. If not, it might be that your first 3 trades are loosing and you panic, thinking that the system isn't working. But if you did those stats right, you might see that the chances are "pretty high" for that to happen, such as 5%. 3 trades are far less than the number of total trades in your backtest, so 5% is high. Then you can feel safe. So yeah, in short, it's all about statistics.
     
  6. I like your advice braincell, especially the pointers on how to know when your system is not performing as expected.

    14 years * 252 trading days/year would make 3528 trades for evaluation. Is that really needed for statistical significance?

    Further, it seems there may be significant differences in the way markets behave over the years. I remember a couple of years ago many traders who had been raking in money started remarking how it had become much harder to make profits. Could that be a case to test the last few years exclusively or with greater weight? Of course, markets change again, and so it goes.

    If the trading days are independent of one another (do not depend on sequential multi-day data), then dividing up 1000 days or so into 2-3 sets of tests might be a good idea. That way you can develop your system on one data set and test on the others to make sure you aren't over optimizing by accident.
     
  7. 377OHMS

    377OHMS

    Are you capable of the brutal honesty needed regarding fills, slippage, dropouts and other execution vagaries in your backtesting?

    You've backtested with one year of data? Do you think that is sufficient?

    You don't mention risk management. Are you risking a fixed percentage of capital per trade or is it dynamic? Are you accounting for serial correlation of your returns, that is, are you randomly sampling with sufficient lag to achieve reasonable independence?

    You mention "beautiful equity curve" but you do not mention drawdown or rigorous application of accepted statistical treatment. Are you calculating Sharpe and the other measures?

    Your apprehension might be justified.
     
  8. nkhoi

    nkhoi Moderator

    whatever you do in test do the same in live. If you didn't trade trade real small in test, do it now.
     
  9. rosy2

    rosy2

    nobody does that. Not even the best HFT firms. you have to have a human watching the bots and infrastructure to make sure nothing hits the fan :cool:
     
  10. 2steps

    2steps

    Here is my question:

    If an automated system can easily make money, wouldn't big banks and hedge funds and mutual funds spend money on developing automated systems?

    If they have developd automated systems that are making money, who are the ones who lose money?

    If 95% of the market participants lose money, that must mean 95% of the automated systems lose money.
     
    #10     Jan 13, 2012