System Developers Journal Thread: Revisited

I too am writing an ATS that will use End of Day Data on futures. I use CSI Data for EOD Futures and have found it to be very good. Supposedly it contains the fewest errors out of all of the data providers but the study they quote it fairly old.

One item I plan to write into the app is some error handling logic to ensure that the current days data has been received because I have found with both Genesis & CSI that from time to time there will be missing data.

The main difference for me was that when calling tech support, CSI's tech support was able to immediately tell me how to correct the issue where as Genesis tech support argued with me about the existence of missing data.

Regards,
Eric

 

Thanks Murray & Trader922 for the answers regarding the data. I shall consider them before making as decision.

Moreover, I have a question; I am thinking of building a system, my own “Trend following system” and wish to compare the same with the harmony one.

 

What happens with long term systems when a trade lasts for 6 months or more and you have multiple slippage and commission how do you test for that?

Also, when you optimize a system, how do you tell if a set of parameters is robust or not?

 

Can we know of your decision and what is your question exactly?

 

I have just got the results of Trend Harmony (http://www.elitetrader.com/vb/showthread.php?s=&threadid=116102&perpage=6&pagenumber=14).

I shall use the same as a benchmark for my own systems. I shall also start posting the different systems as I develop them.

 

I have received Murray Trend Harmony results.

We shall start our analysis by comparing these results to a Triple Moving Average crossover system. It means that we shall use Triple MA system as our basis. The reason of using the same is that it does not have the phase problem of a dual Moving Average (MA).

There is an interview of John Fritz in “Traders Secrets” book wherein he had mentioned that he had used a Simple Quad Moving Average (MA) type system and optimized it walk forward.

Thus, we shall test both the “static optimization” and “walk forward” and discuss the results.

 

Firstly, I would like to apologize to all of you for the long gap which was due to health reasons.

This is reference to the earlier post made:

I am elaborating on the above as follows:

We shall be trying out a small experiment wherein “We shall try to develop a system that shall perform as well as Trend Harmony”

System under study: Triple Moving average system (available in TradersStudio)

The system has the parameters as follows:
ThreeMACrossover(SLen As Integer,MLen As Integer,LLen As Integer)

This system uses 3 different markets as follows:
SLen = Short MACrossover
MLen = Medium MACrossover
LLen = Long MACrossover


Basket: (data set) : Pinnacle Data.

The data symbols (total 21 in numbers) that shall be used are as follows:

1. CT_REV.CSV
2. FX_REV.CSV
3. JY_REV.CSV
4. NG_REV.CSV
5. SB_REV.CSV
6. TA_REV.CSV
7. BP_REV.CSV
8. CL_REV.CSV
9. DX_REV.CSV
10. ED_REV.CSV
11. GC_REV.CSV
12. HO_REV.CSV
13. JO_REV.CSV
14. KC_REV.CSV
15. LB_REV.CSV
16. LH_REV.CSV
17. RL_REV.CSV
18. S__REV.CSV
19. SF_REV.CSV
20. SI_REV.CSV
21. UA_REV.CSV

Procedure:

• We shall optimize the parameters such that values of the parameters are in the following order:
  SLEN < MLEN < LLEN, for valid combinations
  
• We shall stagger optimizations on that basket, and the range of values used shall be as follows:
  SLen => 5 to 30
  MLen => 20 to 80
  LLen => 60 to 150
  
  In other words, we shall optimize the system with the values that fall within the above combinations and range and such that (SLEN < MLEN < LLEN) holds true.
  
  The values shall be increased in the following manner:
  i) Short (SLen) => In steps of 5
  Mid (MLen) & Long (LLen) => In steps of 10
  
  ii) Then all in steps of 2
  
• We shall use maxbars with value as 200. We shall also cover GA (Generic Optimizer) and Walk-Forward. We shall also compare the results with Dual MACrossover.

Summary
In short, we shall run systems based on the three MACrossover (Simple, Dual & Triple) and check out their results.

 

Can you not do better than Triple MA crossover?

 

Since, this is a classic systems and a starting point, in the next version I shall begin to apply filter and separate exits and entries.

 

This has reference to the earlier post wherein I had discussed the details of the experiment I shall do.

I ran the system as follows:

• Created a new session (MA_Crossover) using the system THREEMACROSSOVERS
• Used the Pinnacle Data (the very same symbols mentioned in earlier post)
• Maxbars as 200
• Values of the parameters used are the same as mentioned in earlier post.

The results of the above system has been posted (MA_Crossovers.xls).

Then for the next step, we needed to optimize the system. Hence, after the system ran, we ran the simple “Optimization”, not the “Walk forward one”.

The results for the optimization has also been posted (optim_MA_Crossovers.xls).

Rest of the experiment details shall be given in the next post. If you have any queries, please do not hesitate to ask, thanks.

Both the excel files have been put in a zip file and posted (Results_MACrossover.zip).