Pathfinder Trading System

Forums › ProRealTime English forum › ProOrder support › Pathfinder Trading System

Tagged: 

Viewing 15 posts - 136 through 150 (of 1,835 total)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 … 119 120 121 122 123
  • 10/26/2016 at 7:30 AM #15530
    silotests

    I saw that they raise Germany30 from 0.5% to 1% due to elections (starting October 28).
    Is it the same for you?
    Will you run as usual with Pathfinder during this period?

    10/26/2016 at 8:47 AM #15535
    miguel33

    all trading systems are vulnerable to market shocks as they go well when the market takes directionality. pathfinder and great for that kind of market, but with unpredictable events (elections, disasters, terrorism, etc.) becomes dangerous.
    This and my research.

    10/26/2016 at 11:58 AM #15550
    Reiner

    Hi guys,

    Wilko made an interesting comment in the library about Pathfinder and I really appreciate that kind of feedback. Wilko doubts that Pathfinder’s backtest result will be profitable in real trading.


    @Wilko    : It would be helpful for all of us if you could explain your arguments a little bit deeper. Do you made some tests or what is the basis of your opion? Do you have any ideas how we can improve or harden Pathfinder’s algorithm?

    Thanks, Reiner

    >>>>

    Dear Reiner, dear fellow forum members,

    After writing the previous post, I have myself decomposed the system into several independent systems and backtested them separately without the seasonailty boosting. What I found was that each of the parts/strategies independently generated not many trades over the full available database. In other words, allthough the outcome for each part/strategy may be positive, it is based on few observations, meaning the statistical expectation that historical results would repeat is weak. If you merely add a lot of weak parts/strategies together, this does not in itself guarntee the end-result is a strong (or rubust) system. Therefore, I would issue a warning: The parts/strategies incorporated in this system need to be tested independently on a much bigger dataset in order to be able to say with any statistical significance that the integrated system is likely to repeat past performance. In my book, one needs at least 100 observations for each of the different parts/strategies. To avoid misunderstanding, with parts I mean one part/strategy buys break of monthly high, the next part/strategy buys on break of weekly high, and so on. Pathfinder is comprised of 7 different parts/strategies. On top of that filters and seasonality boosts are applied, not uniformly, increasing the risk of curve-fitting. IMHO.

    There are some nice ideas in the code that have been well applied, and for these ideas I thank Reiner!  I’m not saying the integrated system won’t work, I’m saying the certainty that it will work live is not high enough for me. I hope my findings can be of some help to fellow forum members. Please be careful out there! (Quote h/t the tv-series Hill Street Blues from the 80’s)

     

    10/26/2016 at 12:09 PM #15551
    Reiner

    roxxuz, I’ll run Pathfinder but I’ll reduce the saisonal booster for November from 3 contracts to 1 or 2. In my opion it’s a valid option for everbody to stop the algo in potential fast market scenarios such as the comming election day.

    1 user thanked author for this post.
    avatar Wilko
    10/26/2016 at 2:25 PM #15563
    miguel33

    in fact right. I when there are elections, important news Fed or ECB etc. just reduce the seasonal multiplier.

    10/26/2016 at 3:18 PM #15572
    Wilko

    Dear Reiner,

    Thank you for your positive response, appreciated! First of all, please allow me to clarify that I have no indications that Pathfinder will not be successful, rather I have some doubts that the backtest is robust, in other words that the available backtest data is too limited to be able to say anything about probabilities for future results.

    What I did was this: I looked at the code and what I saw was 7 basic parts (definitions as in the code):

    1. Go long if signal crosses over monthly high
    2. Go long if signal crosses over weekly high
    3. Go long if signal crosses over daily high
    4. Go long if signal crosses over monthly low
    5. Go short if signal crosses under monthly high
    6. Go short if signal crosses under monthly low
    7. Go short if signal crosses under daily low

    On top of that some trades are filtered using MA50 or MA250, and depending on what time of year shorts or longs are boosted (seasonality)

    In order to test the core of the system, I removed all seasonality and then proceeded to make one system for each signal, meaning one system that trades signal no. 1 in the above list, then another system that trades signal no. 2, etc for all signals up to no. 7. Applicable filters (MA50, MA250) and exit parameters (max length, SL, TP) were inserted into each system. This allowed me to backtest and evaluate results for each signal in isolation. What I saw was this;

    Even though the result are positive for most subsystems, for several of the subsystems there were very few positions over the entire backtest period. To me that means that in isolation it is difficult to say if the result is by chance, it is the result of curve-fitting or if it is likely to repeat. When I evaluate my own systems that I write myself, I want to see the result of at least 100 trades, in order to make my mind up if it is a trustworthy system.

    The fact that all the 7 sub-systems above are based on the same smoothed signal does not in itself mean that it is likely that they will all contribute to similar results in the future. It could in fact be an advanced form of curve-fitting, and the profitability of the combination of 7 systems into one with seasonality boosting on top would then just be an illusion. My point is not that “This will not work!”, rather my point is “I do not have enough observations to draw any informed conclusions on if this combined system will generate profits going forward”. It is possible that it will work and generate positive results and it also possible that it will not. It is relatively easy to see situations where it will work less well, and even generate losses, for example in a market that is range-bound and oscillates around monthly low. How likely is that and how often will that happen? I don’t know. But it is one situation that will not suit the system, and there could be others. We just don’t have enough data to be (close to) sure.

    I hope my findings may help contribute to future profits for fellow forum members. Happy trading!

    Kind regards,

    Wilko

    10/26/2016 at 3:35 PM #15574
    Mikael Andersson

    Hi Roxxuz,

    IG raise security demands on DAX and some other indexes from 4th November.

    Thanks,
    Mikael

    10/26/2016 at 6:22 PM #15584
    miguel33

    Reiner,
    would you be so kind as to write in detail and individually all the rules of your pathfinder?
    I also want to bring it back on MT4 and compare. Then I share the results

    Thank you

    10/28/2016 at 4:19 PM #15671
    Reiner

    Hi Patrick,

    you requested an update for the AEX with versionV5B2. 24H quotes for AEX are available since Feb 2011. I didn’t find an AEX Pathfinder version with a drawdown that fulfill my personal requirement of not higher than 20%. I attached the best result so far but the drawdown is still high and around 37%.

    best, Reiner

     

    10/28/2016 at 4:25 PM #15676
    wp01

    Thank you very much Reiner. I really appreciate that you took the time to improve the code for the AEX. Thanks again.

    Best regards,

    Patrick

    10/28/2016 at 4:34 PM #15678
    wp01

    Reiner,

    Have you already thought about adapting Pathfinder to the PRT Complete 10.3 in combination with futures from Interactive? Or are you only interested in the 24hrs trading at IG?

    Regards,

    Patrick

    10/28/2016 at 5:22 PM #15679
    Reiner

    Hi Wilko,

    Thanks for your detailed feedback. I really open for any review and we all make only a step forward when we learn from others and respect another opinon. Pathfinder’s framework based on some very easy ideas that I’m trading since years with good results. I tested many single parts and bundled the best performing things together – exactly the reverse step you have gone in your above comment. It’s really interesting that you mentioned the weak condition “cross under of mothly low”. I also observed that behavior and changed it with V4.

    Pathfinder makes mainly money only on the long side thats maybe a weakness. Another topic is that the system delivers only these exceptional returns with the help of cumulation and that could be a risk.

    The aim of Pathfinder is to have more than a fancy backtest and we all want to avoid to lose money. I mentioned it before I can’t guarantee that Pathfinder will work in the future because I really don’t know it. I tried to avoid curve fitting but there is no doubt that all backtest are a best case apporoach.

    I want to make a step forward and that’s the reason why I have opened a 10k account only for Pathfinder trading the DAX. All the guys here can wait and watch what happens. When I lose the double of the calculated drawdown of 15% (3.000 Euro) then I will consider the project as failed.

    best, Reiner

    10/28/2016 at 6:11 PM #15684
    Reiner

    Miguel,

    with the source code a good MT4 developer should be in a position to migrate Pathfinder to that platform. Describing 250 lines of code in words is’n easy 🙂 but the basic rules are:

    • in the trading window (DAX 9-22) every H4 candle (for the DAX 9, 13, 17 and 2100 CET) the trading algo verify the 6 (variable l1-4 and s1-2) conditions
    • for trading some conditions are combined with 4 Filters (one is your idea!)
    • Key concept of Pathfinder is the smoothed average (if not available in MT4 Hull moving average delivers good results as well)
    • ProOrder manage the position in the following way
      • No pos: buy or sell
      • already long: cumulate in case of a long trade / sell the existing pos and go short in case of a short trade
      • already short: sell the existing long pos and go short in case of a short trade / do nothing because the system is already short
    • The position size is boosted with a monthly saisonal multiplier until a maximal watermark
    • money management will be managed sepratly for long and short trades in the following way:
      • overall take profit and stop loss (valid 24 hours)
      • trailling stop
      • maximal holding for profit and loss triggered by a maximal number of candles

    remarks:

    • a cross outside of the trading window will ignored
    • sunday values will be considered for weekly and monthly high/low if available (DAX, FTSE, DOW)

    I hoply consider all parts. Ask again, if something isn’t clear.

    best, Reiner

    1 user thanked author for this post.
    avatar reb
    10/28/2016 at 6:25 PM #15685
    Reiner

    Patrick,

    As you know my time is very limited and at the moment I focused only on cfd trading with IG.  I will really check if this idea work or not with PRT/IG. Maybe later but I don’t want to promise something.

    best, Reiner

    1 user thanked author for this post.
    avatar wp01
    10/28/2016 at 7:05 PM #15686
    miguel33

    ok i try reiner.

  • Author
    Posts
Viewing 15 posts - 136 through 150 (of 1,835 total)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 … 119 120 121 122 123

Find exclusive trading pro-tools on ProRealCode Marketplace

Similar topics:
2 trading strategies in one system
combining manual trading and auto trading
default prt vwap bands (daily) not selectable for trading system
time-based dax trading system
auto tradeing system
bb trading one long one short on dax 9-17,30
calculate trading day between two date
can’t find automated trading
cascade ordering system
collapse trading toolbar all charts

Create your free account now and post your request to benefit from the help of the community
Register or Login