new (current) symbols vs backtested symbols

Discussion in 'Strategy Building' started by DTorSwing, Jul 10, 2015.

  1. never done so long. who provides 10 years of hist data though
     
    #11     Jul 14, 2015
  2. spacewiz

    spacewiz

    If you backtest any trading algo (RSI2 or anything else) on a specific symbol and you were able to get good results (PnL, Sharpe, whatever you use) - you should get similar results when backtesting same algo on other symbols. That would actually mean you have found a decent algo with good parameters and not just overfitted it to your first symbol's price history.

    You should look at correlations between symbols you are testing - if correlation is too high (say over 85-90% ) - then an overfitted algo will show similar performance on both symbols, so I would pick instruments with correlation somewhere between 50% and 85%. These levels are just guidelines..
     
    #12     Jul 15, 2015
  3. Good point, but in that case, I doubt my algo based on (Dynamic SR and ATR) does good point on all symbols I tested. But my algo , I restricted to change only 1 parameter which is ATR multiplier and others static, to minimize curve fitting

    But I did chose symbols all from only 2 industry “”Biotech and Drug manufacturers”” and I know they are very well correlated, so I need to work on this. I can pick up
    5 from Bio, 5 from Gold sector, 5 from energy, 5 from financial and 5 from technology to see how my algo works
     
    #13     Jul 15, 2015
  4. But I did chose symbols all from only 2 industry “”Biotech and Drug manufacturers”” and I know they are very well correlated, so I need to work on this. I can pick up
    5 from Bio, 5 from Gold sector, 5 from energy, 5 from financial and 5 from technology to see how my algo works

    but I also need to make sure then the correlation somewhere between 50% and 85%
     
    #14     Jul 15, 2015
  5. from my other thread about correlation, someone did great job of posting CC ranks of all symbols, attached here

    as you see the ones in RED are non-correlated per this ranking

    Rank Sym CC Sector Industry/Category
    ---------------------------------------------------------------------------------------------------------------------
    1 SPY 0.9959 ExchangeTradedFund LargeBlend SPDR S&P 500 ETF
    2 IVV 0.9916 ExchangeTradedFund LargeBlend iShares Core S&P 500
    3 VV 0.9783 ExchangeTradedFund LargeBlend Vanguard Large-Cap ETF
    4 VOO 0.9776 ExchangeTradedFund LargeBlend Vanguard S&P 500 ETF
    5 SCHX 0.9744 ExchangeTradedFund LargeBlend Schwab US Large-Cap ETF

    4697 JXI 0.00016 ExchangeTradedFund Utilities iShares Trust - iShares Global Utilities ETF
    4698 WWSG 0.00000 Technology DiversifiedCommunicationServices Worldwide Strategies
    4699 WCHS 0.00000 Financials PrivateEquity WINCHESTER HOLDING
    4701 SSIT 0.00000 BasicMaterials BaseMetals SSI International Ltd
    4702 RDA 0.00000 Technology SemiconductorDevices RDA Microelectronics Inc


    7647 DOG -0.9239 ExchangeTradedFund Trading-InverseEquity ProShares Short Dow30
    7648 SPXS -0.9277 ExchangeTradedFund Trading-InverseEquity Direxion Daily S&P500
    7649 SPXU -0.9364 ExchangeTradedFund Trading-InverseEquity ProShares Trust -
    7650 SDS -0.9467 ExchangeTradedFund Trading-InverseEquity ProShares UltraShort
    7651 SH -0.9515 ExchangeTradedFund Trading-InverseEquity ProShares Short S&P500
     
    #15     Jul 15, 2015
  6. aqtrader

    aqtrader

    The CC has one caveat. It does not apply to penny stocks or the price is flat. So, the above syms in red, only JXI is good. Other syms WWSG, WCHS, SSIT and RDA should be deleted from CC analysis. You can just check the daily year charts, you will see the problem. CC does not count pairs when their prices are equal in historic range.
     
    #16     Jul 16, 2015
  7. aqtrader

    aqtrader

    A few things you should be aware about CC and the private similar correlation measure corel between a pair of two syms.
    If a few syms have the same (or close) CC(or corel) value from a base sym, it does not mean these syms between them are in the same degree of correlation. Like this: If City B and C are in the same distance from City A, it does not mean City B and C are close. City B and C may be even farther away from each other.
    Give an example, here using syms: SPY EWT IYR. EWT and IYR seems are in the same distance of correlation from SPY.
    Sym CC Corel (SPY as base)
    EWT 0.4098 0.2143 ExchangeTradedFund ChinaRegion iShares_MSCI_Taiwan
    IYR 0.4012 0.1968 ExchangeTradedFund RealEstate iShares_US_Real_Estate
    SPY 0.9959 0.00000 ExchangeTradedFund LargeBlend SPDR_S&P_500_ETF
    But the distance of CC between IYR and EWT is not close
    Sym CC Corel (EWT as base)
    EWT 0.9924 0.00000 ExchangeTradedFund ChinaRegion iShares_MSCI_Taiwan
    IYR 0.1479 0.2628 ExchangeTradedFund RealEstate iShares_US_Real_Estate
    SPY 0.4098 0.2143 ExchangeTradedFund LargeBlend SPDR_S&P_500_ETF

    Also CC (and corel) is caculated using a range of certain length of historic data, the value could change while the range is moving on time.
     
    #17     Jul 16, 2015
  8. i did notice that when I put those symbols in TWS, but other 00, anything close to 0, like 0.01 or so, but their price did go with S&P, but I would check for few weeks and see how non-correlated ones go against S&P
     
    #18     Jul 16, 2015
  9. appreciate you detailing out
    and great example of correlation bet EWT and IYR. I see what you mean but I need to watch these symbols for some times and see what the corr means
     
    #19     Jul 16, 2015