| View previous topic :: View next topic |
| Author |
Message |
RTS Trading
Customer
Joined: 19 Feb 2004
Posts: 17
|
 Posted: Thu Feb 19, 2004 6:37 pm Post subject: Fitness |
 |
|
It appears I get to start things off on this topic!
I would like to determine the best values for the few variables I have based upon the most linear equity curve generated for the entire data set being optimized.
I believe that the following will suit my needs but am not sure how to write this correctly(newbie at EL):
Fitness = (NetProfit)/stderror(c,for entire data set)
The highest value will give me the most linear equity curve.
Is there a better wayand how to specify the "entire data set"?
_________________
Thanks,
Bob |
|
| Back to top |
|
 |
Mak
Developers Team
Joined: 26 Jan 2004
Posts: 465
|
| Posted: Fri Feb 20, 2004 3:40 pm Post subject: |
|
|
| Quote: |
| I would like to determine the best values for the few variables I have based upon the most linear equity curve generated for the entire data set being optimized. |
It's not enough, the most linear equity curve is zero line.
Genetic optimazer very quickly finds this solution.
| Quote: |
I believe that the following will suit my needs but am not sure how to write this correctly(newbie at EL):
Fitness = (NetProfit)/stderror(c,for entire data set)
The highest value will give me the most linear equity curve. |
It's more correct ...
See example below
| EasyLanguage: |
{main code of signal }
Array: Equity[10000](0);
Equity[CurrentBar] = NetProfit + OpenPositionProfit;
if LastBarOnChart then Begin
Fitness = Equity[CurrentBar] / MaxList(0.001, StdErrorArray(Equity,CurrentBar-1));
GO_Fitness (Fitness);
End;
|
|
|
| Back to top |
|
|
DT
Developers Team
Joined: 03 Feb 2004
Posts: 188
|
| Posted: Fri Feb 20, 2004 5:27 pm Post subject: |
|
|
Try something like this:
| EasyLanguage: |
Vars: Equity(0), EqMax(0.0001), DD(0), IntegralDD(0), Fitness(0);
Equity = InitialCapital + NetProfit + OpenPositionProfit;
EqMax = maxlist(Equity, EqMax);
DD = Equity/EqMax;
IntegralDD = IntegralDD + 1;
Fitness = Equity/InitialCapital + 2/barnumber*IntegralDD;
|
It is minimize downside (underwater) equity area and maximize equity slope. Upside equity steps holds.
Didn't check it. |
|
| Back to top |
|
|
kshultz
Guest
Joined: 12 Apr 2004
Posts: 12
|
| Posted: Tue Jan 11, 2005 7:27 pm Post subject: |
|
|
In your publication you say:
"Building of non-trivial optimization criterion (Fitness)
In our opinion, building of non-trivial optimization criterion is one of the most essential moments of trading systems building. As it we mentioned above in order to lower the risk of overfitting it is necessary to use another characteristics than NetProfit etc., but those which are associated with desirable behavior of Equity as a whole, not only with its absolute value at the end of the testing period. We can maximize the average of the following value for example:
[log W(T)/W(0) + 2/T * sum { log (W(t)/Wmax(t) }], t = 0,..., T,
where Wmax(t) = max{Wmax(t),W(t)}
i.e. we tend to condition the monotonous growth of Equity W(t)). "
How can I code this in my Fitness( ) statement in EasyLanguage? Ken |
|
| Back to top |
|
|
DT
Developers Team
Joined: 03 Feb 2004
Posts: 188
|
| Posted: Wed Jan 12, 2005 11:14 am Post subject: |
|
|
| kshultz wrote: |
In your publication you say:
"Building of non-trivial optimization criterion (Fitness)
We can maximize the average of the following value for example:
[log W(T)/W(0) + 2/T * sum { log (W(t)/Wmax(t) }], t = 0,..., T,
where Wmax(t) = max{Wmax(t),W(t)}
i.e. we tend to condition the monotonous growth of Equity W(t)). "
How can I code this in my Fitness( ) statement in EasyLanguage? Ken |
Here you are:
| EasyLanguage: |
Inputs: InitialCapital(NumericSimple);
Vars: Equity(0), EqMax(0.0001), DD(0), IntegralDD(0), Fitness(0);
Equity = InitialCapital + NetProfit + OpenPositionProfit;
EqMax = maxlist(Equity, EqMax);
DD = Equity/EqMax - 1;
IntegralDD = IntegralDD + DD;
Fitness = Equity/InitialCapital + 2/barnumber*IntegralDD;
|
I changed logarithms to percentage. It is the same up to 2nd digit.
This fitness maximizes equity growth and minimizes drowdouns as size as duration in the same time |
|
| Back to top |
|
|
krc
Guest
Joined: 02 Nov 2005
Posts: 8
|
| Posted: Wed Nov 02, 2005 12:15 am Post subject: Expectancy / Quality |
|
|
I found code for expectancy and quality at http://unicorn.us.com/trading/src/_SystemQuality.txt
But it doesn't let me optimize for Expectancy or Quality, instead I have to examine in Excel after optimization. Is there an example of Fitness for expectancy and/or quality? |
|
| Back to top |
|
|
Mak
Developers Team
Joined: 26 Jan 2004
Posts: 465
|
| Posted: Wed Nov 02, 2005 4:34 pm Post subject: |
|
|
Function SystemQuality
| EasyLanguage: |
vars: nclosed(0), wins(0), losses(0), scratches(0), scratchloss(0),
scratchtrade(0), studystart(0), studydays(0),
p(0), j(0), k(0), AW(0), PW(0), AL(0), PL(0), expectancy(0),
expectancyscore(0);
{Initializations to occur on first bar}
if currentbar <= 1 then begin
nclosed = 0;
wins = 0;
losses = 0;
scratches = 0;
scratchloss = 0;
studystart = DateToJulian(Date);
{The maximum loss amount for a "scratch" trade will be a
round-turn commission and slippage.
Set Commission and Slippage in Format -> Strategy -> Costs.}
scratchtrade = -1.5 * (commission + slippage);
end;
{Do the following whenever a position is closed}
if totaltrades > nclosed then begin
k = totaltrades - nclosed;
for j = 1 to k begin {loop over multiple simultaneous closed trades}
p = PositionProfit(j);
if p <= 0 then begin
if p >= scratchtrade then begin
scratches = scratches + 1;
scratchloss = scratchloss + p;
end else
losses = losses + 1;
end else
wins = wins + 1;
end;
nclosed = totaltrades;
end;
if losses>0 then begin
studydays = DateToJulian(Date) - studystart;
if studydays < 1 then studydays = 1;
{Calculate results to be optimized, and output them}
if wins > 1 and losses > 0 then begin
j = wins - 1 + losses;{total trades excl scratches & max win}
AW = (GrossProfit-LargestWinTrade)/(wins-1); {avg win}
PW = (wins-1) / j; {% wins}
AL = (GrossLoss - scratchloss) / losses; {avg loss}
PL = losses / j; {% losses}
end else begin {handle division by zero cases}
j = wins + losses;
if wins <= 1 then AW = 0 else AW = GrossProfit / (wins-1);
if j = 0 then begin
PW = 0; PL = 0;
end else begin
PW = wins / j; PL = losses / j;
end;
if losses = 0 then AL = 0
else AL = (GrossLoss - scratchloss) / losses;
end;
if AL = 0 then expectancy = AW*PW
else expectancy = (AW*PW + AL*PL) / (-AL);
expectancyscore = expectancy * j * 365/studydays;
end;
SystemQuality = expectancyscore; {dummy return value}
|
Example of use
| EasyLanguage: |
Inputs: Gen(1), ShowInd(1), ModeTSGO(0), Population(50), FreshBlood(0), MyReportName("MBOX");
Vars: Len1(20),Len2(10),Len(0),Q11(0),Q12(0),Q21(0),Q22(0),KATR(10);
Vars: R(0),K(0),LastRun(0),Ind(0);
If CurrentBar = 1 Then Begin
R = TS.GO.Start(MyReportName);
If Gen=1 Then Begin
R=TS.GO.Mode(ModeTSGO);
R=TS.GO.Popul(Population);
R=TS.GO.FreshBlood(FreshBlood);
K=TS.GO.Chrom("O");
R=TS.GO.Gen("O.Len",K,1,100,1);
R=TS.GO.Gen("O.Q11",K,-1,1,0.1);
R=TS.GO.Gen("O.Q12",K,-1,1,0.1);
K=TS.GO.Chrom("E");
R=TS.GO.Gen("E.Len",K,1,100,1);
R=TS.GO.Gen("E.Q21",K,-1,1,0.1);
R=TS.GO.Gen("E.Q22",K,-1,1,0.1);
end;
LastRun = TS.GO.Next(Gen);
Ind = Iff(LastRun = 1,ShowInd,0);
Len1 = TS.GO.Get("O.Len",Ind);
Len2 = TS.GO.Get("E.Len",Ind);
Q11 = TS.GO.Get("O.Q11",Ind);
Q12 = TS.GO.Get("O.Q12",Ind);
Q21 = TS.GO.Get("E.Q21",Ind);
Q22 = TS.GO.Get("E.Q22",Ind);
R = TS.GO.ShowViewer;
End;
{Code of Signal here ...}
{.................................}
Vars: Fitness(0);
Fitness = SystemQuality;
IF LastBarOnChart Then Begin
R=TS.GO.Fitness(Fitness);
end ;
|
|
|
| Back to top |
|
|
krc
Guest
Joined: 02 Nov 2005
Posts: 8
|
| Posted: Fri Nov 04, 2005 6:21 pm Post subject: Fitness for Walk Forward Optimization |
|
|
I was reading the articles by Dennis Meyers at:
http://www.meyersanalytics.com/articles.php
and I wonder if I can build a fitness function that helps optimize strategies as he outlines in his articles.
i.e.,
limit the profit factor to a range (1 - 2.5)
restrict the number of consecutive losing trades
make sure that it trades daily (number of trades > N)
(varies by backtest period)
But not too many trades: (number of trades < M)
- Kevin |
|
| Back to top |
|
|
Mak
Developers Team
Joined: 26 Jan 2004
Posts: 465
|
| Posted: Sun Nov 06, 2005 12:48 pm Post subject: |
|
|
You can use any limitations in fitness that you can write with EasyLanguage.
Limitations can be presented as "penalty functions".
If set limitations are exeded, we penalize system by reducing fitness.
It's possible to use several penalty fuctions at a time.
For example:
| EasyLanguage: |
Fitness = SystemQuality {It's main criteria, may be any ...}
- Iff (ProfitFactor > 1 and ProfitFactor < 2.5, 0, 100000) {Limits to profite factor}
- Iff (TotalTrades > N and TotalTrades < M, 0, 100000) {Limits to number of trades}
- Iff ( ............... ) { .............. }
|
|
|
| Back to top |
|
|
bearwatch
Guest
Joined: 22 May 2006
Posts: 30
|
| Posted: Tue May 23, 2006 5:14 am Post subject: great post |
|
|
This is a great post! I've been busting my head trying to figure out how to use fitness to smooth out the EC.
I have one question, using these criterion, has anyone successfully walked forward their system with same curve line? Thanks. |
|
| Back to top |
|
|
VladGor
Support Team
Joined: 26 Jan 2004
Posts: 528
|
| Posted: Tue May 23, 2006 9:53 am Post subject: |
|
|
We didn’t understand your question.
Can describe in more detail what is happening? |
|
| Back to top |
|
|
bearwatch
Guest
Joined: 22 May 2006
Posts: 30
|
| Posted: Tue May 23, 2006 10:46 am Post subject: |
|
|
I just wanted to know if anyone has used these criterion (equity curve... high equity growth, small drawdown) and have it trade live or walk forward with same results as during backtest or optimization periods.
One other question is I'm experimenting this SystemQuality function with multiple criterion from the above post, how do implement this in EL code? The ProfitFactor word gave me an error (would not verify).
I have to say so far, this is better than GrailSystem on this very flexible feature to choose your own criteria. |
|
| Back to top |
|
|
bearwatch
Guest
Joined: 22 May 2006
Posts: 30
|
| Posted: Tue May 23, 2006 10:47 am Post subject: |
|
|
I cannot get this code to verify in EL. How would I do this? Thanks
----------------------------------------
Fitness = SystemQuality {It's main criteria, may be any ...}
- Iff (ProfitFactor > 1 and ProfitFactor < 2.5, 0, 100000) {Limits to profite factor}
- Iff (TotalTrades > N and TotalTrades < M, 0, 100000) {Limits to number of trades} |
|
| Back to top |
|
|
Mak
Developers Team
Joined: 26 Jan 2004
Posts: 465
|
| Posted: Tue May 23, 2006 11:06 am Post subject: |
|
|
ProfitFactor must be calculated befor
| EasyLanguage: |
Var: ProfitFactor (0);
ProfitFactor = Iff(GrossLoss < 0,-GrossProfit/GrossLoss,0);
Fitness = SystemQuality {It's main criteria, may be any ...}
- Iff (ProfitFactor > 1 and ProfitFactor < 2.5, 0, 100000) {Limits to profite factor}
- Iff (TotalTrades > N and TotalTrades < M, 0, 100000) {Limits to number of trades}
|
|
|
| Back to top |
|
|
bearwatch
Guest
Joined: 22 May 2006
Posts: 30
|
| Posted: Tue May 23, 2006 12:08 pm Post subject: |
|
|
| Oh yes, I forgot, I'll give it a try. Thanks for a quick response. Good product good support, I'm very close to buying the license! |
|
| Back to top |
|
|
|
|
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
|
|
