| View previous topic :: View next topic |
| Author |
Message |
bearwatch
Guest
Joined: 22 May 2006
Posts: 30
|
 Posted: Tue May 23, 2006 4:33 pm Post subject: |
 |
|
| I am currently optimizing GO using SystemQuality function as the fitness criteria on 5min on 1 yr of data. It's still in 2nd test of 100 and it has been 2 hours. Is this normal? I was expecting it to be a bit faster than conventional optimization process. |
|
| Back to top |
|
 |
VladGor
Support Team
Joined: 26 Jan 2004
Posts: 528
|
| Posted: Wed May 24, 2006 4:42 pm Post subject: |
|
|
In this case run that is performed takes too much time.
Possible causes:
- big history
- big number of trades
- complicated calculation of fitness
What to do
- shorten the history (less bars)
- use biger time frame (less bars)
- try simplier fitness |
|
| Back to top |
|
|
bearwatch
Guest
Joined: 22 May 2006
Posts: 30
|
| Posted: Mon May 29, 2006 9:52 am Post subject: |
|
|
I'm trying to set a criteria for the fitness function. Is there a way to set a comparative equation for example putting this:
fitness = profitfactor>2;
Would this make the GO go faster directly to the solutions?
The other question is can we set multiple criterion in fitness, say high profit factor and low maxdrawdown? If yes, how can we write it? Thanks. |
|
| Back to top |
|
|
VladGor
Support Team
Joined: 26 Jan 2004
Posts: 528
|
| Posted: Mon May 29, 2006 1:37 pm Post subject: |
|
|
It's possible but it is poor choice.
Correctly in EasyLanguage it would be: fitness = IFF(profitfactor>2,1,0); But that is not the best way to calculate fitness because it can get values only 1 or 0. For example fitness = profitfactor is much better.
| Quote: |
| Would this make the GO go faster directly to the solutions? |
No, see the explanation above.
| Quote: |
The other question is can we set multiple criterion in fitness, say high
profit factor and low maxdrawdown? If yes, how can we write it? |
In current version only one fitness is possible. Multiple criterion can be combained into one for example multiplying or combining criteria with some wights.
For example
Fitness = ProfitFactor / MaxList(-MaxIDDrawDown,0.001)
In the future versions we will offer optimization on multiple criteria. |
|
| Back to top |
|
|
bearwatch
Guest
Joined: 22 May 2006
Posts: 30
|
| Posted: Mon May 29, 2006 2:23 pm Post subject: |
|
|
When do you expect the next version with multiple criterion to be released? And I purchase the current version, is the upgrade free or is there a fee? Thanks.
Back to the weights, I assume from the previous calculations is .0001 is the percentage in a scale between profitfactor and maxdd correct? If I want make 70% MaxDD to 30% PF, how do I calculate that? |
|
| Back to top |
|
|
VladGor
Support Team
Joined: 26 Jan 2004
Posts: 528
|
| Posted: Mon May 29, 2006 3:18 pm Post subject: |
|
|
| Quote: |
When do you expect the next version with multiple criterion to be
released? And I purchase the current version, is the upgrade free or is
there a fee? |
We can't exactly say when it will be next updates,
but our clients will get them in accordance with our policy:
http://www.tsresearch.com/support/upgrade_policy/
| Quote: |
Back to the weights, I assume from the previous calculations is .0001 is
the percentage in a scale between profitfactor and maxdd correct? If I
want make 70% MaxDD to 30% PF, how do I calculate that? |
No, constant 0.0001 only for elimination of division by zero.
It isn't easy to enter weights.
If standardize each factor to range (0-1),
than it is possible to just sum up with assigned weights.
But it isn't very easy to standardize. |
|
| Back to top |
|
|
bearwatch
Guest
Joined: 22 May 2006
Posts: 30
|
| Posted: Mon May 29, 2006 9:26 pm Post subject: |
|
|
| Does anyone have a fitness criteria that smoothens the equity curve using rate of return or regression? I know someone in TS uses them but not sure how to make a formula of it. Thanks. |
|
| Back to top |
|
|
Mak
Developers Team
Joined: 26 Jan 2004
Posts: 465
|
| Posted: Tue May 30, 2006 6:45 am Post subject: |
|
|
This topic began with smoothen the equity curve.
Some more example:
Function TS.GO.TestEquity(Len)
| EasyLanguage: |
Inputs: Len(NumericSimple);
Vars: R(0), X(0), X2(0), Eq(0), DEq(0), MaxD(0), MinD(0);
Eq = NetProfit + OpenPositionProfit;
DEq = Eq - Eq[Len];
X = X + DEq;
X2 = X2 + DEq*DEq;
R = Iff(X2 > 0, (X/CurrentBar)/SquareRoot(X2/CurrentBar - Square(X/CurrentBar)), 0);
TS.GO.TestEquity = R;
|
This fitness depends of all equity curve, not last equity value. |
|
| Back to top |
|
|
bearwatch
Guest
Joined: 22 May 2006
Posts: 30
|
| Posted: Tue May 30, 2006 9:01 am Post subject: |
|
|
| This "Len" parameter that is passed to the function, is that number of bars? |
|
| Back to top |
|
|
Mak
Developers Team
Joined: 26 Jan 2004
Posts: 465
|
| Posted: Tue May 30, 2006 9:08 am Post subject: |
|
|
| Yes. |
|
| Back to top |
|
|
bearwatch
Guest
Joined: 22 May 2006
Posts: 30
|
| Posted: Wed May 31, 2006 8:08 pm Post subject: |
|
|
| I've run a few tests on these and the results look good but I can't help thinking it's overfitted for the data being optimized. What does this function does other than being selective about when and where to exit? Is it possible to continue walking forward with such a nice EC? |
|
| Back to top |
|
|
VladGor
Support Team
Joined: 26 Jan 2004
Posts: 528
|
| Posted: Thu Jun 01, 2006 9:31 am Post subject: |
|
|
Sorry, we didn't understand your question.
Can describe in more detail what you need ? |
|
| Back to top |
|
|
bearwatch
Guest
Joined: 22 May 2006
Posts: 30
|
| Posted: Thu Jun 01, 2006 10:05 am Post subject: |
|
|
Sorry about that. I ran a few fitness criterion used above that placed smoothing out EC as important, and the tests when walking foward, the results were terrible. I believe this might not be an appropriate solution since it is curve fitting the in-sample data.
I wanted to ask about the TestEquity function. Do we insert this function as a fitness? Or is it just for testing after running the other EC smoothing criterion?
I was about to make a purchase of this tool as I see it convincingly it's a necessity to fast foward in building my strategies. I hoping that the Freshblood and TSGO.Mode will make live walking foward and live trading a lot better! |
|
| Back to top |
|
|
bearwatch
Guest
Joined: 22 May 2006
Posts: 30
|
| Posted: Thu Jun 01, 2006 10:12 am Post subject: |
|
|
I would like to calculate the Avg Monthly Return and the Sharpe Ratio as part of fitness criteria. One of the experienced system traders told me to calculate that the Avg Monthly Return * 4 > Max Drawdown.
Any idea how we formulate this? |
|
| Back to top |
|
|
VladGor
Support Team
Joined: 26 Jan 2004
Posts: 528
|
| Posted: Fri Jun 02, 2006 9:18 am Post subject: |
|
|
| Quote: |
Sorry about that. I ran a few fitness criterion used above that placed
smoothing out EC as important, and the tests when walking foward, the
results were terrible. I believe this might not be an appropriate
solution since it is curve fitting the in-sample data. |
It's possible that your system has tendency of curvefitting and such result can be with any fitness.
| Quote: |
I wanted to ask about the TestEquity function. Do we insert this
function as a fitness? Or is it just for testing after running the other
EC smoothing criterion? |
Question is not clear...
TestEquity valuates growth of equity on Len bars interval. TestEquity = Mean(DE) / Stdev(DE), where DE - growth of equity on Len bars. TestEquity can be used as fitness.
If you set Len as one month then TestEquity will evaluate similary to Avg Monthy Return * 4 > Max Drawdown.
Fitnes = Avg Monthly Return / Max Drawdown is similar to
Fitnes = Avg Monthly Return / Stdev(DE), the difference is only with divisor.
Max Drawdown è Stdev(DE) are different variations of evaluating of equity offset.
If you make testing on minute chart, try to use Len coresponding to longer time frame, for example weekly or monthly.
Len should be corresponding to time frame where several trades are made or average length of trade so the system can be evaluated. For example for hourly chart I would set Len to week but for daily Len probably should be month or quarter.
If you se Len < 100 on minute chart then equity will contain only noice.
If you use free version then you won't be able to do walk forward.
It's made for introducing purposes with this product. Number of runs of strategy is limited to 100 but for real optimization you need several times more. Besides that it uses only Mode=0. |
|
| Back to top |
|
|
|
