Quantitative Finance

Discussion in 'Strategy Building' started by Spectre2007, Jan 9, 2007.

  1. Your entire backtest is based on 2 days' worth of cable 1m bars and you are averaging a trade about every 20 minutes, is that correct? If so, you may be disappointed in a longer forward test, let alone live trading.

    Also, why is your MQ "n/a" rather than the maximum 90%, essential for all MT4 testing, especially with such high-frequency method?
    #11     Jan 9, 2007
  2. the first trade was entered 4 months ago. I'm using the system as a indicator, and directionally trading, but adherein to systems SL and TP's. With custom entries within the TP-SL range.

    I'm going to post the 1 minute report in a few minutes, its still running.
    #12     Jan 9, 2007
  3. I hear you. I wasn't really trying to poke the details of your idea out of you as much as get a feel for what's going on. Here are some generic suggestions.

    A good site for pure quants is Wilmott.com. Of course, they tend to trade and model much more exotic instruments than what you will find in the retail crowd but if you are really into quantitative modeling that's a good site to start with. You can jump to the blogs of some of the more notable site members as well and if you search you will find links to other interesting quant sites.

    FWIW, I would suggest you actively search for a period where this particular system doesn't do well and see if you could withstand the draw downs. Frankly, a 10% stop loss/trade would give me an ulcer. I couldn't trade that way at all.

    Another way to approach it is to look at the cost of buying puts or calls to hedge your position. Or you could simply move over to puts and calls anyway and ignore the underlying. That would limit your risk but reduce your profitability as well. You'd have to test it out to see what the trade off was.

    As to the other ways to fine tune the system dynamics you might look at the Hurst exponent and see if that provides a better fit than LR. You could also add a larger time frame analysis to give you a bigger picture of what's going on. The basic idea would be that pullbacks against the trend -- which is what you are trading in part if I read the description properly -- are less likely to succeed than one's with the dominant cycle.

    Hope it adds a few ideas.




    I noticed LateApex's comment on your testing period. I missed that. Obviously that's a nonstarter -- as is only 4 months. Basically, I try to use enough data to run through at least one up and one down cycle to see system robustness. In that period there is almost always a chop phase or two as well.
    #13     Jan 9, 2007
  4. Bars in test 42284 Ticks modelled 84468 Modelling quality n/a

    Initial deposit 100000.00
    Total net profit 18940.00 Gross profit 98940.00 Gross loss -80000.00
    Profit factor 1.24 Expected payoff 461.95
    Absolute drawdown 0.00 Maximal drawdown 30000.00 (21.75%) Relative drawdown 21.75% (30000.00)

    Total trades 41 Short positions (won %) 24 (83.33%) Long positions (won %) 17 (76.47%)
    Profit trades (% of total) 33 (80.49%) Loss trades (% of total) 8 (19.51%)
    Largest profit trade 3000.00 loss trade -10000.00
    Average profit trade 2998.18 loss trade -10000.00
    Maximum consecutive wins (profit in money) 11 (32940.00) consecutive losses (loss in money) 2 (-20000.00)
    Maximal consecutive profit (count of wins) 32940.00 (11) consecutive loss (count of losses) -20000.00 (2)
    Average consecutive wins 5 consecutive losses 1

    I said I would post this, heres the one minute data backtest. I know about the quality of the backtest issue. I'm mainly using this model to see what the system tells me, about price behavior. The market seems to behave based on the gross sum of its participants. Trying to find the 'average behavior pattern' of its participants.
    #14     Jan 9, 2007
  5. I've got to agree with LateApex here. I don't think your analysis is meaningful due to the duration issue. With regards to finding the "average behavior pattern" I think the same criticism applies.

    Best of luck with it.


    #15     Jan 9, 2007
  6. thanks for the link, cool sites.

    The system posted is just a blunt knife. There is a repository of 1M data. And ways of improving the modeling quality if others are concerned.

    That blunt knife has been sharpened.

    #16     Jan 9, 2007
  7. Yes, you'll definitely want to get your MQ up to 90% before you do anything else in MT4. Until you address that issue, none of your development work, no matter how painstaking, is going to be meaningful.

    Fortunately, it's not hard to do and only needs to be done once, for each currency or instrument of interest. A step-by-step guide from Hendrick is attached.
    #17     Jan 9, 2007
  8. Look...this is all "good stuff", but for each strategy and approach, it will produce drastically different results. There are two sides to trading: entry and exit. In general, I agree with you....but each system/strat stands on it's own.
    #18     Jan 9, 2007
  9. If you are talking about multiple independent variables to derive a price proxy or expectation, then you are right. If you are talking about just using Linear Regression on the actual price as a smoothing tool, you are totally wrong. LR is terrible at smoothing because it has a huge overshoot....everyone is aware of this. There are tons of much better smoothing tools out there.
    #19     Jan 9, 2007