300% Annual Return -- Fully Automated

Discussion in 'Journals' started by frostengine, Jul 2, 2011.

  1. The goal of this journal is simple. To prove that it's possible to have huge returns from a 100% automated system. After years of frustration and "almost" moments, I nearly gave up. A few months ago a light bulb went off. I had an idea, one that I had many times before, but this time was different.

    I have set aside roughly $25,000 in an Interactive Brokers account that will be used for this journal. The strategy is flat at the close of business everyday. I will aim to post screenshots of my account daily or at least weekly. I could fall flat on my face, if so I will post screenshots of that as well.

    The strategy goes live on Tuesday. In subsequent posts I will describe what markets are traded, a brief overview of the strategy, as well as detailed statistics based on expected return.
  2. DT-waw


    300% annual return?
    do you plan to increase your size during the year and how many times?

    you can do it. just don't use counter trend systems and don't do any manual tweaking.
  3. DT-waw


    btw, i'm up 90% during the last 12 months on an account triple your IB acct. without re-investing. so you can definitely make 300%.

    i wish you patience in posting here.
    i've never felt any benefits of running a journal - nobody is interested or curious.
  4. Looking forward.
  5. byteme


    Went off?
  6. The strategy consists of around 20 neural networks. Each NN was trained against ES data from 2000-2009. I have used neural networks in the past with sub-par results. However, typically as soon as I test out of sample data the strategy falls apart. This time was different. I decided to throw out everything I thought I knew about NN and try something "unique". The result of that experiment became this strategy.

    The strategy will trade ES, CL, and NQ. Keep in mind the NN was only trained against a subset of ES data, never trained against CL or NQ. However, once I post the equity curve for each instrument you will see the CL and NQ results are actually better than ES.

    The number of contracts traded will grow as the account size grows. 300% is actually a conservative number which is attainable even without growing the number of contracts.
  7. The strategy is designed to protect as much capital as possible. Every trade has a "worse" case stop of around 20 ticks. This very rarely gets hit.

    When trading ES, NQ, and CL all as a portfolio, the strategy has less than 1% chance of any given week to be a loser. With that said, I am sure this upcoming week will be one of those 1%.

    There is around a 11% chance that any given day will be a loser.

    Max draw down as a portfolio should be less than 10% of account value.
  8. DT-waw


    99 out of 100 weeks profitable? very impressive.
    let's see how it will work in out of sample, live trading.
  9. In many respects it appears "too good". I have done two sanity checks over the past month, each lasting only a few days.

    #1 I let it run on SIM account for a few days then compared those results/fills with backtesting that same period. Matched up very closely.

    #2 I let it run for 1 day with real money and compared it to both SIM and backtest and it matched up.

    The sanity checks were done against ES not CL or NQ however. In my backtesting I am including 1 tick of slippage per trade on ES and 2 ticks on both CL and NQ. I know 1 tick for ES is a fair assessment, however for CL and NQ I am unsure.

    Will be a very interesting week.
  10. Look forward to checking out your results.

    Was wondering if you could you share some stats:

    1) Ave no of trades/day/symbol

    2) Ave hold time

    3) Calmar Ratio (Annualized Return / Max Drawdown)

    4) In vs out sample returns

    I'm sure you're already aware of this, but 20 neural nets allows one quite a large number of degrees of freedom, especially on one symbol. Even an HFT system might not generate enough trades to prohibit curve fitting. How many weights do you have per NN on average? Have you walk-forward tested your system and/or performed Monte Carlo runs to validate your results? (I not criticizing here, just making an observation).

    I'm an automated trader myself and developed an intraday RTM bot that's been trading a basket of stocks successfully for about 2 years now. I'd like to adapt it to ES or NQ, so I'm totally routing for you. Best of luck!
    #10     Jul 2, 2011