ES method - How to duplicate backtest?

Discussion in 'Strategy Development' started by mrpace, Apr 16, 2004.

  1. mrpace


    Here's the method's numbers, backtested over 7813 trades dating back to January 1999 on the e-mini S&P (this period of time excludes the ESZ3 contract due to the fact that I don't have this data available, but it does include the first quarter of this year, ESH4)

    In any event, this method I backtested came up with 7813 trades, and a winning % of 64% based on one contract traded per trade. The average win was $51.44 and average loss was $78.73 after comissions. The W/L ratio is .65

    Total P/L over the 5 years was +$36,509.80 for one contract.

    The method assumes you are entering at the CLOSE of the entry signal bar, and exiting at the CLOSE of the exit signal bar.

    We all know that in reality, you can not ALWAYS enter and exit at the exact close of the signal bar because of slippage.

    Since the profit per trade (as well as the loss per trade) is rather small, slippage could easily destroy much of this system's profits over the long haul.

    My theory is to always enter with a limit order, and if you get filled before an exit signal, then all is well and good. If you don't get filled before your exit shows up, you skip the trade.

    The exit is a bit trickier. I figure if the trade is showing a profit, when I get an exit signal, I will use a limit order to try and get out. If the trade starts going against me, I will change to a market order to ensure I exit with at least one tick of profit.

    If, on the other hand, I get an exit signal when the trade is already showing a LOSS, I will immediately get out using a market order.

    Is it just impossible to trade an ultra short term system like this which relies on almost zero slippage to remain profitable?

    It's quite frustrating to come up with system like this, backtest it over a good amount of time and a big number of trades, only to realize that in reality, this thing probably isn't going to work.....

    Any thoughts?
  2. mrpace


    I must add that I have the ability to fully automate this system via Ensign and a visual basic program, so the moment a signal is generated, the trade goes out to IB TWS. This will most likely help in securing a fill....

    I will also add that this system buys when the market is headed down (but trailing off in intensity) and sells when the market is headed up (but again, apparently trailing off in intensity). I think this will also help in securing limit order fills at specific prices, which is necessary.

    However, in the very limited time I have been trading this system with real $$, I am wondering if there are any other strategies or devices I can use to further enhance my ability to secure my fills.
  3. In your backtesting, what assumptions did you make for getting a fill with a limit order? This is crucial and could easily swing your already precarious results into loss territory. I could think of worse things than trading a system like yours, but I would need to believe in it 100% beforehand, as by its very nature it will inevitably put you through many crises of confidence.
  4. mrpace


    Signals are generated at the end of 1 minute bars. I assumed that all orders were filled at the closing price of the 1 minute bar that generated the signal.

    The same for exits. I assumed all orders were filled at the closing price of the 1 minute bar that generated the exit signal.
  5. mrpace


    I guess the ultimate question is how does anyone successfully trade an ultra short term method like mine that appears to work on paper?

    I can see being successful holding for bigger gains, as this would naturally incur less overall trades, less commissions, and a lesser chance that slippage would have a big impact on real world results. Does this mean that trading a system that only holds from between 2 and 15 minutes is impossible in real world situations?
  6. A_B


    This system will not work.

    Average Trade = Total P/L / # Trades = $4.67.
    Profit Factor = (Avg Win * % Win ) / (Avg Loss * % Loss) = 1.16.
  7. I work with tick data, but I think I understood what you meant. The FIFO nature of Globex orders invalidates your assumptions which, even after using best possible case fills, yield a marginal result. Sorry, but I would look elsewhere. However, get a second opinion.
  8. It doesn't work on paper.
  9. mrpace


    Not to be confrontational, but how does it not work on paper? It has a positive mathematical expectation. It shows a decent profit over a five year test period over thousands of trades. How does it not work on paper?

    I understand why it probably won't work in the real world, but what about what I see on my backtest charts?
  10. Your response is far from confrontational and one that is both welcome and expected. When I work with tick data, I use simulations which assume a "fill probability". For systems similar to yours, if I find that my system will be profitable for a fill probability of 20%, but not for a fill probability of 10% (and this occurs frequently), then I will reject the system. What fill probabilities have you used? Unless you only assume a fill if the price trades through, which you don't, you will have exaggerated and unrealistic results. On paper.
    #10     Apr 16, 2004