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

RE: [amibroker] Re: How to spot over optimizing - Summary



PureBytes Links

Trading Reference Links



<FONT face=Tahoma 
size=2>-----Original Message-----From: Rick Parsons 
[mailto:RickParsons@xxxx]Sent: Tuesday, July 16, 2002 
9:49 AMTo: amibroker@xxxxxxxxxxxxxxxSubject: RE: 
[amibroker] Re: How to spot over optimizing - Summary
<FONT color=#000080 
size=2>Herman,
<FONT color=#000080 
size=2> 
These 
rules are very good.  However I wonder if one should be aware that trying 
to satisfy every rule listed will result in constant testing/tweaking trying 
to get everything perfect.  Some of us, I suspect, are constant tweakers, 
never satisfied, and never trading.<SPAN 
class=470582314-16072002>[Herman van den 
Bergen] 
<SPAN 
class=470582314-16072002>Absolutely right Rick :-) I am borderline myself. I 
love the development more than the trading. However, over the last few 
years I have developed many systems and i am beginning to see a definite 
correlation between equities - even if they are based on different types of 
formulas.. It may be possible to create a "System Effectiveness" 
indicator that is based on a collection of equities derived from the same 
formula by letting the system operate with slightly off parameters; the 
equities will often still synchronize at drawdowns but with different lead 
times. For a trader who is dedicated to a particular type of trading 
systems (I tend to concentrate on short term oscillator systems) there may 
be a time to trade and not to trade. The problem is that for most tradersit 
is hard to sit on the side lines for up to 7-8 months - even when this would 
be to their advantage in terms of risk and reward. But isn't this what a lot 
of people are doing at this very moment?
<SPAN 
class=470582314-16072002> 
I dolove the 
idea of combining various equity curves and getting out when they all head 
south.
<FONT color=#000080 
size=2> 
<FONT color=#000080 
size=2>Thanks,  
 
<FONT color=#000080 face="Vladimir Script" 
size=5>Rick

<FONT face=Tahoma 
size=2>-----Original Message-----From: Herman van den Bergen 
[mailto:psytek@xxxx]Sent: Monday, July 15, 2002 5:27 
PMTo: amibroker@xxxxxxxxxxxxxxxSubject: RE: 
[amibroker] Re: How to spot over optimizing - 
SummaryHere's the uncooked and unseasoned stewof 
things to consider (andunderstand) when optimizing. May or may not apply 
to your type of system.- Optimizing rules vary for different types 
of trading systems- The system works on only one stock- The gains 
are erratic with a final big gain- Optimization values are scattered 
(random peaks on a 3D chart)- The gains are too good to be true 
(1000,000,000%)- If you think you found the HG think again- The 
equity line is straight, almost no drawdowns- The system only worksin 
recent years- The system doesn't work out-of-sample- Optimizing 
parameters vary widely for different time periods- Greater than 75% 
winning trades- DrawDowns greater than 50% for the bare-bone system- 
The system was tested on only one period- Less than 30 trades- No 
out of sample test (can be done backwards)- Remember that the market has 
natural Friendly and UnFriendly periods thatyou can detect by overlaying 
the equities for as many systems as you canmuster, if all equities start 
to dip, the market is becoming unfriendly.- Consider using the C-Test by 
William Eckhardt [ See Stocks & CommoditiesV. 12:5 (218-221): The 
c-Test by William Eckhardt.]- Consider the type of parameter you are 
optimizing.- Do check out opt values outside of the expected range, 
there might besurprises out there.- Consider that some parameters 
interact and others do not.- Consider one-by-one optimization and 
on-shot deals, perhaps try both?- When incrementally optimizing 
parameters note that the requirement for itsfrequency is the same as for 
trades, if you add a new rule and it kicks-inonly once in ten yearsthat 
doesn't mean much, even if your gain triples.- To compare the effect of 
incremental system optimizations overlay theequity curve for each stage 
of the system (log scale). If equities track inparallel your enhancement 
may be insignificant. If the equity curvesseparate at an upward angle, 
this means the improvement takes placegradually, this is more 
significant.- To analyze the effect of optimizations you can also plot 
the Buy/Sellsignals for the various stages to see where and how 
differences occur.- Chart each and every parameter.- ...best 
regards,Herman.Your use of Yahoo! 
Groups is subject to the Yahoo! 
Terms of Service. Your 
use of Yahoo! Groups is subject to the <A 
href="">Yahoo! Terms of Service.