Auto trading 1 contract 62K 5 years...

Discussion in 'Automated Trading' started by frostengine, May 19, 2006.

  1. I have been back testing different trading methods for a while and generally I just end up with some over curve fitted garbage. This time I developed the method while actually trading and then implemented it to just see what the back testing results would look like. I had to simplify the entry somewhat to make it easier to test. Because in the real method I use discretion on when to exit I had a big problem with exits. So right now as it stands I have it exit and reverse when the opposite condition is met. So its always in the market.

    So bottom line, I ran the implementation the first time and ended up with 62K profit over 5 years for 1 contract of ES. The number of winning days and loosing days is about 50%. The BIGGEST issue is the max 1 day loss. The backtesting results showed a max loss of $1200 in one day per contract which I would not have the stomach for. Especially considering winning days is about 50% meaning I could have a long string of loosing days which I coudlnt handle.

    So now my issue comes how do I get these max losses down? Currently the system uses a 3 pt stop. I attempted to put in a max loss variable that when hit trading woudl stop for the day. I was able to achieve a max loss of $500 in a given day and still keep profit over 42k for the 5 year period. Doing htis however made the winning day % significantly lower.

    A few things I did to make sure I didnt over curve fit. Is I did not optimize ANY parameters after the first run. Simply went with what I had. I then tried moving parameters around and profit stayed roughly the same area, 50-70k. In all my previous systems where i curve fitted too much I noticed that if i shift the time in the morning when the system woudl start for example from 9:30 to 10:00 the profits of the systems would drop very sharply negative.

    So for this system I tired offsetting the starting time in the morning from 0 to 120 minutes and in every case the profits did not change much at all... So I feel fairly confident the system is real and not curve fitted.

    The real issue now becomes finding a way to get those max losses down. What are some good exit techniques to use in automated systems?
  2. lrm



    Sounds like you're making progress, congratulations! What money management method are you using? Since you are around a 50% w/l ratio, you might want to look into using a variable method. As your string of losses increases, you risk less and less until you start winning again.

    A very seasoned strategy trader has outlined this topic here: Size

    Please keep us informed as to how you make out.


  3. In an effort today to continue to try to prove to myself this system is not overly curve fitted. I ran a little more than 3 years worth of ER2 data. I simply plugged the data in made NO changes to the system. Here are the results:

    In the 3 years the system ran it would have made a total of $40,300 PER contract.... had a max 1 day loss of $1500. Win % was around 56%.

    These are incredible results. In my mind at least it shows that the system is NOT curve fitted at all. The max loss is still WAY too high... I still have not attempted to play with ways of getting the max losses down any yet.

    If only I had the gut to run this thing... but with losses that high in one day I couldnt stomach it... woudl pull the plug the first day it had a loss like that... To feel confortable running the system I need max loss in a day to be less than 350 or so per contract.

    Next I will test this system on about 6 years of 6-7 Years of NQ data. Assuming it shows similar results I will begin trying to get the max losses down.
  4. Oh yea, one more thing.... to make sure slippage is deffinetly factored in. I just ran the er2 data again, this time making all buys and sells happen a whole 1 minute bar past the entry/sell signals. This resulted in a profit of $32,000 for the 3 year profit. I was actually expecting the profit to drop a lot more than that considering this gave a much worse entry/exit than what would be experienced in a live environment.
  5. sulli


    Out of curiosity, what was your max drawdown? (over many days)

    You mention that your daily loss is >$1000 and you don't have the stomach for that. What about a series of days with losses accumulating > $1000.

    your 50% win method probably needs a better risk/reward per trade.
  6. Here is a performance to check against.

    100% mechanical


    intraday only

    1 turn a day.
  7. Trade on a fractal several times slower (you are on 1 min now) and keep your slippage data on the 1 minute.

    Read profitlogic stuff as well.

    My suggestion is based on getting a different ratio than 50-50. it should slip to more skewed positive. Cut the long trades down in # and decrease the long trades duration for each. This should also make your drawdownes occur more on the loong trades and be less in value. Do not screw around with your advantage definition.
  8. what is profitlogic?
  9. I've been really enjoying his writings as well.

    There is some controversy surrounding them, but hey, so long as all they cost is "time" and I can sit and read them while catching Leno, it's all good.

    You can also do a search on his name (ProfLogic).

    Great posts by Grob109. So what happens when you backtest with all of those different criteria, both singular AND in as many combinations as there are criteria listed for the ES, YM and NQ?

    "I" don't it's going to be possible get that drawdown any lower while trading the ER series (ERU/ER2, etc., depending on your data feed/broker), because it has such a $value per point value, when it trends, you make a lot of dough, but when the trade goes against you, you can lose it just as fast, and when it chops (because of the high $value to point value) YIKES!

    Best Regards,

  10. Like so.

    For ES, YM and NQ, include these parameters in your program:

    a) 100% mechanical

    b) 1k/month/contract

    c) intraday only (no overnight holding)

    d) 1 turn a day (try this critera with 2 turns a day, 3 turns a day),

    e) "timed trading sessions", ie., 9:30 - 12:30pm for the morning session / 1:30pm - 3:30pm for the afternoon session for entering trades

    for the different "fractal times" try:

    a) 3 minute

    b) 5 minute

    c) 10 minute


    and last but no least, try different markets

    a) curde oil e-mini

    b) any number of the different currecy combinations

    Let us know how it works out for you.


    #10     May 20, 2006