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

Re: Fuzzy Logic NN etc. WAS: Rocket Science



PureBytes Links

Trading Reference Links


----- Message d'origine ----- 
De : Lawrence Chan <stnahc@xxxxxxxxx>
À : <omega-list@xxxxxxxxxx>
Envoyé : dimanche 6 février 2000 16:36
Objet : Re: Fuzzy Logic NN etc. WAS: Rocket Science


> 
> on pete's comments, my understanding is that he thinks :-
> 
> 1. big trades are the meat
> 2. small trading losses does not matter.
> 3. 50/50 win/loss is ok, as trends are not coin flips
> 4. please check dax 30min future - and i know nothing about it :)
> 5. all trading systems suffers from generalization
>    (or is it - no standard to measure against?)
> 
> and he attached a chart too on the trading signals his program generate
> with a comment saying the same system is used for cac and dax futures.
> 
> and now i am curious, the chart attached with his message is in fact
> very easy to trade, and in fact #4 is wrong, as I do trade dax futures :)
> 
> and I think using a breakout system on the chart he shows will yield
> better performance then his fuzzy logic system because a
> break out system will be in the market less time on the same chart,
> and most important of all, win/loss at a way better level :)

You are rigt, the standard channel breakout works better  on the latest portion of the dax chart
that I previously sent ( less trades and a slight better profit).

Now, if you want to discuss this on a larger scale, here are the results of the same channel breakout ( 10 bars , as provided in TS)
on the dax futures from 01/1993 to 08/1999, 30 min and the same fuzzy system on the same period .
Resuts expressed in  dax  points, no cost and slippage in both cases:

CHANNNEL BREAKOUT  10 bars
 Performance Summary:  All Trades       

 Total Net Profit  $3 564.98  Open position P/L  $25.00 
 Gross Profit  $23 921.23  Gross Loss  ($20 356.25) 
        
 Total # of trades  1 312    Percent profitable  39.71% 
 Number winning trades  521    Number losing trades  791   
        
 Largest winning trade  $523.80  Largest losing trade  ($447.60) 
 Average winning trade  $45.91  Average losing trade  ($25.73) 
 Ratio avg win/avg loss  1.78  Avg trade (win & loss)  $2.72 
       
 Max consec. Winners  7    Max consec. losers  13  
 Avg # bars in winners  25    Avg # bars in losers  9  
       
 Max intraday drawdown  ($1 746.90)    
 Profit Factor  1.18  Max # contracts held  1  
 Account size required  $1 746.90 

 Return on account  204.07%<===========Look!==========

NEUROFUZZY CAC DAX SYSTEM
Performance Summary:  All Trades       

Total Net Profit  $12 133.50  Open position P/L  $68.00 
Gross Profit  $31 591.00  Gross Loss  ($19 457.50) 
       
Total # of trades  1 874    Percent profitable  46.10% 
Number winning trades  864    Number losing trades  1 010   
       
Largest winning trade  $479.50  Largest losing trade  ($259.50) 
Average winning trade  $36.56  Average losing trade  ($19.26) 
Ratio avg win/avg loss  1.90  Avg trade (win & loss)  $6.47 
      
Max consec. Winners  10    Max consec. losers  10  
Avg # bars in winners  17    Avg # bars in losers  6  
      
Max intraday drawdown  ($772.00)    
Profit Factor  1.62  Max # contracts held  1  
Account size required  $772.00  

Return on account  1571.70% <=========Look!===========

I have no other comment to add...


 > -Lawrence Chan
> p.s.
> the reason why US has lower interest in fuzzy logic and NN,
> is that quite sometime ago some computer scientists in North America
> has proven that if fuzzy logic, or any other equivalent mathematical
> system, can produce exact accurate solutions - then P=NP. Thus,
> when a fuzzy logic model or NN model get closer and closer to
> solve a NP complete problem, it will theoretically take them forever
> to compute the answer. Thus, the academic interest in america
> immediately dropped to almost none - although practical applications
> and interest has been kept increasing as they are indeed better
> methodology to describe everyday real life conditions.
> Are securities time series functions NP complete, I don't know!  :)
> 

 Of course, if US computer scientists have proven this, I also have nothing to add!.....

Sincerely,

-Pierre Orphelin
Neurofuzzy Logic tools for TradeStation
Free evaluation versions and competitive upgrades available
web: http://www.sirtrade.com