System with few trades, will you consider?

Forums › ProRealTime English forum › General trading discussions › System with few trades, will you consider?

Viewing 14 posts - 1 through 14 (of 14 total)
  • 04/23/2018 at 12:07 PM #68978
    CKW

    hi all,

    Strategy with few trades, will you consider ? E.g 30 mins , 2 years only 20-30 trades.

    if you will consider, how do you validate it ?(as WFA might not be accurate for system with less trades )

    1. If it’s work on other instruments, go ahead to live.

    2. avoid it ?

    3. Or simply remove one of the filter to increase the trade frequency with lower winning rate of course & return the same total gain. In this case, more accurate data for WFA.

    Which approach you will choose ?

    Br,

     

    04/23/2018 at 1:11 PM #68984
    GraHal

    Test on Demo for a few months and see if (the few) results show same performance as on Backtest?

    You could post the code on here and ask others to comment on their opinion of curve fitting and / or test on 200k bars for you??

    04/23/2018 at 1:15 PM #68986
    Vonasi

    Generally if I create a strategy and it has very few trades I will bin it.

    Either it has very few trades because it is over filtered so over optimized. Removing filters will make it open more trades but for most likely worse results of average gain, gain/loss ratio etc

    It is also not worth bothering with as just a few trades a year will never make you rich unless you can create hundreds of these strategies that all work.

    Also as you have said it is virtually impossible to back test and get any meaningful results so you can never be confident in it. Who wants to wait four months for a trade just to find out that the strategy was fitted to historical data and you lose money proving that.

    1 user thanked author for this post.
    avatar Leo
    04/23/2018 at 1:27 PM #68989
    CKW

    Hi Grahal,

    Test on Demo for a few months and see if (the few) results show same performance as on Backtest? -> This is already done for this example
    in fact not any new system, but combining mix & match the ideas from System A and System B by somebody else.

    Yes I guess request others help to test 200k bar is the easier way. otherwise, it seems difficult to validate for low frequency trade system.

    br,

    04/23/2018 at 2:09 PM #68994
    CKW

    Thanks Vonasi

    I am unsure if every filters are “over fitted” or some can really improve the “trade quality”. I am trying to understand…
    Let me provide 2 examples (entry point) and assuming core condition is fulfilled(step 1)

    Step 2
    1. with variables, e.g. adding RSI and optimize variables may lead to over fitting
    2. without variables, e.g High[0] > High[1] and Close[0] > Close[1] and Low[0] > Low[1], or “insidebar” breakout

    In my opinion, #2 is more reliable filter to improve quality of trade but reduce number of trades.
    Welcome to correct me if my understanding is wrong, maybe they are the same 🙂

    br,

    04/23/2018 at 2:37 PM #68995
    Vonasi

    IMHO they are all the same.

    Every time you apply a condition to a strategy it is a filter. Some conditions can be curve fitted in more than one way so for example with your RSI you can curve fit it with the level it is at or crosses or is below and also curve fit it with the period that you use for the RSI.

    High > High[1] can also be fitted several ways. One is by changing the look back period to High > High[2] for example and the other is by changing the thing that you are comparing high to so High > Close[1] etc.

    They all have multiple variations that will change the outcome of your strategy. The lower quantity of them that you have the less likely your strategy is to be curve fitted and the more likely it is to be robust and work going forward. There is also the benefit that you have more trades to test on to prove this the lower the number of variables/filters there are.

    1 user thanked author for this post.
    avatar GraHal
    04/23/2018 at 5:43 PM #69002
    Inertia

    Depends….

    It has to look like the attached picture…with a very low losses amount compare to the possible average winings…

    Other strategies with different ratios…No.

    04/24/2018 at 4:13 AM #69036
    CKW
    Vonasi wrote:

    High > High[1] can also be fitted several ways. One is by changing the look back period to High > High[2] for example and the other is by changing the thing that you are comparing high to so High > Close[1] etc.

    Agree. It shall not be changed to avoid curve fitting

    Inertia wrote:

    Depends…. It has to look like the attached picture…with a very low losses amount compare to the possible average winings… Other strategies with different ratios…No.

    Hi Inertia,

    Yes. It’s important criteria but how do you validate it? or just ensure acceptable Stop loss in place and GO live

    br,

    04/24/2018 at 10:25 AM #69059
    Despair

    I haven’t read the whole thread, so with the risk somebody wrote the same before. With so few trades I’d go for one straight 70/30 OOS-Test since as you correctly point out a WFA is not working.

    04/25/2018 at 6:19 AM #69113
    Inertia

    Does not make much sense. Not enought trades I suppose.

    Thx.

    04/25/2018 at 9:08 AM #69122
    GraHal

    So is the period on the right (showing the trades) truly out of sample?

    Or had you already optimised over this period anyway prior to doing Walk Forward … in which case the period on the right is In Sample / Not Out of Sample?   

    04/26/2018 at 10:53 AM #69204
    Inertia

    Hi Grahal,

    Kindly forget it. It is not really relevant due to the poor number of trades….so does not make much sense.

    Thank you.

    04/26/2018 at 11:29 AM #69205
    CKW

    Thanks all for your comments.

    I refresh nicholas WFE Blog & found the post from Jebus89. (https://www.prorealcode.com/topic/thoughts-on-profitable-trading-systems/

    Now i am more convinced having enough trades is very important for long run.. 🙂

    In my strategy database, majority are low number of trades mainly on indexes. Each system capped at limited losses and acceptable RR due to min contracts requirement from IG for indexes. Even through i am confident but i still not convinced to increase the contracts so i don’t see this approach will bring me anywhere ($?). It’s much safer to create multiple strategies on small trades, instead of “increase contracts” on big trades.

    br,

     

     

    04/26/2018 at 12:26 PM #69218
    Vonasi

    Yes making decisions to go live based on backtesting is a major problem for the small bank balance trader. In theory I could have made a lot of money running one of my strategies on the FTSE100 if I believe the backtests and if I started with a lot of money and could stomach some big drawdowns – but you always have to think that that is the optimal result so best to double the drawdowns and half the profit and suddenly the risk to reward doesn’t seem so attractive. So in the meantime that one remains on demo having annoyingly not had a losing trade yet and only a tiny amount of drawdown! It is the big market shifts that can destroy a seemingly very profitable strategy and unfortunately they are the hardest thing to predict. Just because doing x had you getting out of the market before one in the past does not mean that doing x is a profitable way to get out before one moving forward and this can be the worst form of curve fitting. A strategy that makes money whether the market is going up or going down is preferred but then you have two possible areas to curve fit it in!

  • Author
    Posts
Viewing 14 posts - 1 through 14 (of 14 total)

Find exclusive trading pro-tools on ProRealCode Marketplace

Similar topics:
you don’t know what you don’t know!
2 trading strategies in one system
auto tradeing system
autotrading – latent gain will not show on charts
backtesting will not begin
can you add compounding to a strategy?
cann’t you use a ttrailingstop and a stoploss together?
cascade ordering system
could you pleases covert this code from “volumatic variable index dynamic aver
dax dayly: system that works.

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