Developing a Trading System Step by Step

[Murray Ruggiero]

Quote from horribilicus:


Among the spectacularly trending non-USA markets that you didn't include, Thursday's volume on March06 Japanese Govt Bonds was 24,000 contracts, March06 EuroBund was 1.2 MILLION contracts, March06 EuroSTOXX was 591,000 contracts, and March06 London Robusta Coffee was 6,800 contracts.

Yes I realize these market exist and they trend well, but using the non US markets add a layer to the analysis which I did not want to deal with. For example the Japanese Govt Bonds min move point value is in Yen, not dollars . This means that we have to adjust on a trade by trade basis these world markets trading profits based on the exchange rate for that trade. If we do not do this step we can not combine accurately these markets with the US futures markets which we want to trade. This is too much for a basic tutorial.

It is a good point , but most people don't realize you can't include these markets in your baskets without doing the proper conversion.

 

[Prevail]

Following.

 

[Murray Ruggiero]

Quote from dloomis514:

5 20 50, hi profit per trade short and long, high PF for short and long

Good answer but we need more information, What happens beyond 5,20,50. If the best parameters happen at the end of our optimization window in this case 50 for the long term MA, we need to expand the area. We can see that longest term average in the area of 20,25 do not produce good results so we will rerun our optimization as follows

3...9 steps 2,15..30 step 5 , 30...60 steps 5. This will allow us to look at both sides around 5,20,50. One side around these parameters for example 5,20,45 made 612K, so that acceptable for a neighboring set of parameters, but we have no information on the other side, so we need to get it.

 

[Murray Ruggiero]

New optimization grid, with new parameters around the set of parameters we liked from the first run.

 

[dloomis514]

3 20 60, 5 20 60, 7 20 60, 9 20 60 all look good to me from the avg gain per trade side (short or long)

 

[zuijlen]

Quote from dloomis514:

3 20 60, 5 20 60, 7 20 60, 9 20 60 all look good to me from the avg gain per trade side (short or long)

Which seems to indicate that further improvement may be achieved by looking at values beyond 60. In other words, we cannot be sure that we have reached the most optimal solution yet.

 

[Murray Ruggiero]

Quote from zuijlen:

Which seems to indicate that further improvement may be achieved by looking at values beyond 60. In other words, we cannot be sure that we have reached the most optimal solution yet.

Yes , you are exactly correct, and this was the next point I was going to make. In addition I want to make the point that sometimes it helps to develop a custom measure to rank parameter sets. This can be done in TradersStudio 2.0. We have a custom ranking field which can be filled using any formula you want to measure the performance of any given set of parameters. It is very simple to add this to a system and requires only a few lines of code. For example to create a measure using the ratio of NetProfit to Drawdown.

If ISoptimize()=1 And BarNumber=LastBar Then
SetOptimizeFactor(NetProfit/Drawdown)
End If

This create a extra-column we can sort by in TradersStudio. You can use all types of measures, for example you could subtract or add points based on the percentage of winning months a set parameters has had or how much money has the system made over the past year.


Let's think of some measures we would like to test on and the range of parameters for that testing. Please give me input. I will run them over the weekend and post our results. After that we will select a set of parameters and continue on our way.

 

[Murray Ruggiero]

Please give me some feedback on some measures you would like to use to score optimization. One example would be to weight based on Slope of the equity curve over past year in addition to Netprofit/Drawdown.

I will run my analysis and put it up Sunday night so you have time to give me some feedback. They take a while to run so the quicker I get feedback the more time I will have to run several different measures and discuss the pro's and con's of each of them.

 

[Murray Ruggiero]

Does anyone have any ideas of what measures they would like to see me use in my optimizations as a custom performance measure. I plan on combining the concept of NetProfit/Drawdown with momentum of the equity curve. Does anyone have any better ideas. I would like to hear from you.


I will run this analysis tomorrow night , not tonight. This will give people one more trading day to give me some input.

Otherwise, I will run my analysis using my ideas and will post it by sometime Tuesday.

 

[Murray Ruggiero]

I did not get any feedback, so I am going to run this analysis using the following measure

(NetProfit+2*(NetProfit-NetProfit[250]))/Drawdown

This means we count the netprofit over the past 250 day 3 times versus once on older data. This will give systems that have performed well recently better scores. We can use 3X weighting because we are running our analysis on over 24 years of data.