3 registered members (MadJack, AndrewAMD, Quad),
540
guests, and 1
spider. |
Key:
Admin,
Global Mod,
Mod
|
|
|
Training Z strategy with different AssetsFix.csv
#469326
11/10/17 23:44
11/10/17 23:44
|
Joined: Jun 2013
Posts: 1,609
DdlV
OP
Serious User
|
OP
Serious User
Joined: Jun 2013
Posts: 1,609
|
Hi all, I'm probably not remembering something right, so would appreciate help/clarification! Using Z3 since it Tests faster. I install 1.66.5, and Test Z3. Results are per the manual. I replace AssetsFix.csv with my own Broker's parameters and reTest. Results are different, as expected - unacceptably worse, in fact. I Train Z3 and it downloads data and creates a new Z3.par. I reTest expecting better results, as presumably Z3 has now been Trained to my AssetsFix.csv. However, the Test results are exactly the same as before Training. What am I missing? Why did the Train not improve the Test results? Is there no way to Train to my AssetsFix.csv? Thanks.
|
|
|
Re: Training Z strategy with different AssetsFix.csv
[Re: MatPed]
#469343
11/11/17 19:22
11/11/17 19:22
|
Joined: Jun 2013
Posts: 1,609
DdlV
OP
Serious User
|
OP
Serious User
Joined: Jun 2013
Posts: 1,609
|
Thanks all! @jcl: Thanks for the clarification. Is this in the manual somewhere? Sorry if I missed it. Followups: Do I understand correctly that there is therefore no way to verify the results produced by the Train? I.e., what Trade results would be expected using the new Z3.par and my Broker's values? My AssetsFix.csv came from the Download script without any changes - are Roll values still suspect in that case? @MatPed: Agreed. The Z Systems have yet to completely satisfy my Proof-Of_Concept perspective. But (aside from Z5 ) they haven't completely crashed and burned yet either... How long have you been on Oanda? Has it been an OK/good experience? Thanks again all.
|
|
|
Re: Training Z strategy with different AssetsFix.csv
[Re: DdlV]
#469361
11/12/17 14:24
11/12/17 14:24
|
Joined: Feb 2015
Posts: 652 Milano, Italy
MatPed
User
|
User
Joined: Feb 2015
Posts: 652
Milano, Italy
|
Thanks all!
How long have you been on Oanda? Has it been an OK/good experience?
Thanks again all. One year, mor or less. So far so good, but tts a MM broker not the best... No ETFs, Stocks, Futures and Options are the real downside...
|
|
|
Re: Training Z strategy with different AssetsFix.csv
[Re: jcl]
#469437
11/15/17 22:01
11/15/17 22:01
|
Joined: Jun 2013
Posts: 1,609
DdlV
OP
Serious User
|
OP
Serious User
Joined: Jun 2013
Posts: 1,609
|
Then *sigh* I'm confused again. Why when I Train or reTrain Z3 do I only get the Z3.par file and not the Z3_n.par files? I suppose it's because Z3 has the PARAMETERS flag OFF? Thanks.
|
|
|
Re: Training Z strategy with different AssetsFix.csv
[Re: jcl]
#469439
11/16/17 06:08
11/16/17 06:08
|
Joined: Jun 2013
Posts: 1,609
DdlV
OP
Serious User
|
OP
Serious User
Joined: Jun 2013
Posts: 1,609
|
Thanks for confirming, jcl! Moving on... The published ~85% AR is produced using the included AssetsFix.csv. For Z3, the pertinent lines are:
Name,Price,Spread,RollLong,RollShort,PIP,PIPCost,MarginCost,Leverage,LotAmount,Commission,Symbol
NAS100,4185,2,-0.1,0,1,0.0871,2,0,0.1,0,
SPX500,2032.4,0.5,-0.5,0,0.1,0.0871,8,0,1,0,
US30,17520,2,-0.5,0,1,0.0871,6,0,0.1,0,
XAG/USD,17.235,0.046,-0.0009,0,0.01,0.43547,5,0,50,0,
XAU/USD,1260.82,0.47,-0.0124,0,0.01,0.00871,11,0,1,0,
My AssetsFix.csv unchanged from the Download script Tests to ~30% AR (& other metrics reduced as well):
Name,Price,Spread,RollLong,RollShort,PIP,PIPCost,MarginCost,Leverage,LotAmount,Commission,Symbol
NAS100,6295.60,1.00000,0.1700,-.3400,1.0000,0.100000,3.5000,179.874,1.0,0.000,NAS100
SPX500,2578.17,0.50000,0.1200,-.6300,0.1000,0.100000,14.0000,184.155,1.0,0.000,SPX500
US30,23388.50,2.0000,-.0700,-.3300,1.0000,0.100000,13.0000,179.912,1.0,0.000,US30
XAG/USD,17.0470,0.04300,-.0100,-.0020,0.0100,0.500000,9.0000,94.706,50.0,0.000,XAG/USD
XAU/USD,1279.15,0.39000,-.0100,0.0000,0.0100,0.010000,7.0000,182.736,1.0,0.000,XAU/USD
I don't see values that differ by an order of magnitude. From the Test log, the trades start out the same, but the trade volumes are different. Thereafter, the trades taken differ somewhat, but the different trade volumes continue... Does anything leap out at you explaining the drastic reduction in AR (& other metrics)? If not, what would the procedure be to determine what's going on? Thanks.
|
|
|
Re: Training Z strategy with different AssetsFix.csv
[Re: jcl]
#469450
11/16/17 18:33
11/16/17 18:33
|
Joined: Jun 2013
Posts: 1,609
DdlV
OP
Serious User
|
OP
Serious User
Joined: Jun 2013
Posts: 1,609
|
Yes, 30% AR is not bad. But I think it's important to understand what's going on and why simply changing the Assets parameters seemingly for the better gives a worse backtest... Comparing the base Transaction costs -1007$ spr, -507$ slp, -655$ rol to my AssetsFix.csv's Transaction costs -1569$ spr, -935$ slp, -2028$ rol shows Spread ~50% worse, Slippage almost doubled, & Roll more than tripled. I don't see that the differences in Assets parameters directly accounts for this... The base Portfolio analysis
Portfolio analysis OptF ProF Win/Loss Wgt% Cycles
NAS100 avg .068 2.32 22/19 37.9 \///////
SPX500 avg .000 0.60 7/11 -13.8 ///....
US30 avg .056 2.23 22/16 37.2 //////./
XAG/USD avg .020 1.74 17/17 51.1 ////X/X..
XAU/USD avg .000 0.61 12/24 -12.5 .X/X..//
becomes
Portfolio analysis OptF ProF Win/Loss Wgt% Cycles
NAS100 avg .149 4.32 16/10 73.7 ///.//////
SPX500 avg .025 1.28 10/14 11.5 ///../..
US30 avg .105 2.22 22/15 41.8 //////./
XAG/USD avg .004 1.10 18/24 15.5 /X/XXX..
XAU/USD avg .000 0.67 9/23 -42.6 .XX/..//
and again I don't see a direct cause. There is an indirect cause in that my AssetsFix.csv causes larger trades, which causes different trades to be taken. Could the optimization for one trading pattern not work for another? Since I cannot, could you Train Z3 using my AssetsFix.csv and see what results? Thanks.
|
|
|
Re: Training Z strategy with different AssetsFix.csv
[Re: jcl]
#469469
11/17/17 10:05
11/17/17 10:05
|
Joined: Jun 2013
Posts: 1,609
DdlV
OP
Serious User
|
OP
Serious User
Joined: Jun 2013
Posts: 1,609
|
Thanks both! @MatPed: There's so much about this I don't fully understand yet... I hadn't even really gotten to the OptF's on the belief that they're driven by the backtest results... @jcl: a) Isn't it more complicated than that? I.e., the trades taken aren't exactly the same, so while more leverage is involved that doesn't a priori mean bad (-12.5) must become worse (-42.6). F.i., SPX500 changed from losing (-13.8) to winning (11.5). Wouldn't optimizing using the different AssetsFix.csv therefore produce different results than using the leverage 100 values on a leverage 200 account? b) Finally getting to OptF's , would the 0 OptF for XAU prevent it being Traded and therefore Trade would give better results than the Tested 30% AR (which includes the XAU loss)? Thanks.
|
|
|
Re: Training Z strategy with different AssetsFix.csv
[Re: jcl]
#469547
11/20/17 17:56
11/20/17 17:56
|
Joined: Jun 2013
Posts: 1,609
DdlV
OP
Serious User
|
OP
Serious User
Joined: Jun 2013
Posts: 1,609
|
Thanks jcl. The veil has lifted a little, but of course that leads to more thoughts/questions... Re. Optimizing & Leverage: Leverage determines trade size, and hence affects trade mix, and hence system results. Optimization is working at the individual trade level to maximize individual trade results. But is this an inherent restriction? Couldn't optimization also be used at a "higher" level and affect the trade mix? Re. Excluding Asset(s): I will of course try this, but for the benefit of other readers, will Test then run w/o that Asset(s)? Similarly, for the Cold Blood Index feature to work, do I need to (re)Test with the Asset(s) excluded to get a corrected DBL file? Thanks!
|
|
|
Re: Training Z strategy with different AssetsFix.csv
[Re: jcl]
#469666
11/29/17 17:48
11/29/17 17:48
|
Joined: Jun 2013
Posts: 1,609
DdlV
OP
Serious User
|
OP
Serious User
Joined: Jun 2013
Posts: 1,609
|
Thanks jcl. I was hoping to save some of that effort if the answer is already known. Or at least better target the investigation... Summary of what I think I experienced, so far: A Strategy's results are directly influenced by the account Leverage (only - nothing else was changed). Different Leverage gives a different trade mix and the results will be different, sometimes dramatically. Nothing is "wrong" per se, and individual trades remain optimized per the Train. But the mix difference means less (or un-) profitable Assets may end up being traded more than the more profitable one(s). Excluding an unprofitable Asset(s) does not return the Strategy to the original Leverage's results. In that context, are either or both of these true? a) A difference in Leverage should not change the Test trade mix! There is some problem which needs to be identified! b) If an Asset(s) are Excluded, the Strategy must be reTrained since different trades will be taken for the non-Excluded Assets. Using the prior optimization values will lead to different Test results. Thanks.
|
|
|
Re: Training Z strategy with different AssetsFix.csv
[Re: DdlV]
#469682
12/01/17 07:56
12/01/17 07:56
|
Joined: Jul 2000
Posts: 27,978 Frankfurt
jcl
Chief Engineer
|
Chief Engineer
Joined: Jul 2000
Posts: 27,978
Frankfurt
|
|
|
|
|