[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Metastockusers] The number of parameters in a trading system: Looking for a Balance?



PureBytes Links

Trading Reference Links

The number of parameters in a trading system: Looking for a Balance?

  Questions to the board:

  1.  Each parameter introduced into a system represents a measure of
  control, which limits the reliability of the outcome. The more control
  that is introduced into a trading system, the less reliable the will be.
  In a statistical term, you are loosing degrees of freedom. So the more
  variables added like stops, logic filters and indicators (RSI, MACD,etc)
  the more control you have asked and the less reliable will be the result.
  This is true in general because it depends in the cross/joint correlation
  between independent variables (ie:indicators). A well designed experiment
  will check for possible joint correlation among independent variables.
     In conclusion, the best system approaches are the ones which use a
     very low number of variables (2 to 5), if possible. The more control
     parameters the more likely a pair may be highly correlated.


  a.  Is the above statement correct?
  b.  Does logic filters (buy  if ..., exit if...) limits the reliability
  of the outcome?
  c.   If yes, how can I calculate the joint correlation with a RSI and a
  logic filter?
   d.  what it might mean if you can get
good results with a lot of variables, which is contrary to the popular
wisdom?


  2.  Designing a neural network  to have a minimal degrees of freedom
  (DOF) is critical when attempting to forecast financial markets. If a
  neural network has too many, the model may produce excellent training
  results but could generate unreliable forecasts when presented with
  unseen data.

  a. Is the above statement correct?

  3.  How much data to consider for Optimizing parameters? I think it
  should be a balance between long ad short period:
      i) If you take a small period you might leave a side certain market
      characteristics and the optimization might not hold in the future.
      ii) if you take a long period you might overfit and little data left
      to walk forward testing.
  With that said I would only reoptimize the system in real trading if
  results are not good letīs say in the last 4 weeks or the DrawDown
  reaches a certain amount.
  This is a reactive trading system not a continuously optimized system.

  a. What do you think of my reasoning?
  b. Any comments?




 

Yahoo! Groups Links

To visit your group on the web, go to:
 http://groups.yahoo.com/group/Metastockusers/

To unsubscribe from this group, send an email to:
 Metastockusers-unsubscribe@xxxxxxxxxxxxxxx

Your use of Yahoo! Groups is subject to:
 http://docs.yahoo.com/info/terms/