Phoenix optimization - page 5

 

To answer your question, you don't need to wait for 5.7.2 but the variables will be renamed in 5.7.2 so all of the variables are displayed when we copy/paste the results. Since results pages need to be manually input still, there isn't any loss of labor by waiting. If you make a bunch of .set files in 5.7.1, they won't be compatible without changes

I turned over code changes to PContour, maybe if we ask nicely he will include an Excel file converter up to 5.7.2 and then we can stop changing the names of things.

5.7.2 doesn't (at this time) include the Money management tuning that was submitted.

 

MaxTrades and ConsecutiveSignals

I wonder if someone could explain exactly what the function of MaxTrades and ConsecutiveSignals is. I am not sure how they are affecting the outcome of my optimization. Thanks.

 

In all modes, consecutive signals are checked to make sure they all point in the same direction. Several people noticed that signals would often oscillate back and forth between buy and sell, buy and sell. When a trade exited, it would just pick the first one and trade it. Default behavior is now waiting for 5 consecutive signals (by tick) before trading. A very conservative value. During high volume times, that could be 2-3 seconds. During low volume, spikes may interrupt and cause 'dry spells' where no trades occur. It is also possible to uncomment 2 lines in code and require a minute based period, e.g. 1 minute of consistent signals, 2 minutes, etc. If you only have 1 trade going at a time, the default tick based setting won't impact the overall number of trades.

Maxtrades is a setting specific to mode1 only. Mode1 was limited to 1 trade, Mode2 split that into 2 trades, and mode3 split into 3 trades. I looked through the code, and didn't see any reason why Mode1 needed to be limited to 1 trade only. So I included a parameter for experimenters like myself to see what multiple trades would do. If it gets a signal, trade it. this often leads to a buy and a sell on the same currency, but sometimes they both profit. Because of this, optimization with maxtrades increased favors tp>sl.

With some settings, maxtrades rarely gets several trades going at once. With other settings, it executes as many trades as possible. (Based on AvailFreeMargin this is safe but the lotsizes decrease with a lot of trades.)

With maxtrades turned on, it is a good idea to increase consecutive signals and choose settings with a low drawdown. I have more info about these (and all other settings for signal) in the optimization thread.

I don't like "fragile" settings, I disregard settings that break by incrementing a value one or two, and prefer settings that have solid performance within a range. It is philosophical, but I think I'm eliminating some of the backtest settings that curve fit. More details in the notes of post 38 in optimization thread.

 

Always "brilliant" Darak !

 

Hello, Are we ready for the next step?

daraknor:
To answer your question, you don't need to wait for 5.7.2 but the variables will be renamed in 5.7.2 so all of the variables are displayed when we copy/paste the results. Since results pages need to be manually input still, there isn't any loss of labor by waiting. If you make a bunch of .set files in 5.7.1, they won't be compatible without changes

I turned over code changes to PContour, maybe if we ask nicely he will include an Excel file converter up to 5.7.2 and then we can stop changing the names of things.

5.7.2 doesn't (at this time) include the Money management tuning that was submitted.

daraknor,

I sent you back the code on 1/16/2007 at 11:08 pm and I wanted to find out.

  1. Did you get my code changes by email?
  2. What happened with all the changed variable can you copy and paste everything now?
  3. I had a bug fix on the new code. Was I right about that?
  4. I was waiting for confirmation that everything was OK then I could:

  5. put together another converter for 5.7.1 to 5.7.2.
  6. Revise the Optimization Manual.
  7. Update the ultimate guide.

I had a really busy week so I couldn't have done anything until today.

 

Hi Darak

Hope you recover asap from your illness

Attach a "Trend Catcher" indicator to test.

Settings put to : Trend CCi Period 34, Automatic Time period : true, M15 CCI period 34

Interesting.

Will check another settings

See you

Files:
 

One note about Phoenix

Hi guys,

I am running Phoenix live for about 2 weeks now on fxdd, results not so impressive. I noticed one thing when I tried to compare actual trades versus tester trades - I ran tester for the live trading time period. Results were absolutely different, no tester trade matched live trade. This is because of one thing - multiple signals. When we launch Phoenix, it takes first available signal and trades it. You may launch it midday, in the morning or in the evening. But when testing, tester takes data from 00:00 server time, so it may (and probably will) catch different, more early signal. Then this trade will end at a different time that the real one, and tester will get another signal, and in the end you will have totally different testing report. So using Phoenix could be a lot of random walk, having in mind that some signals are good, some are false. Conclusion - if you want your Phoenix to give you results as close to tested as possible, start it at 00:00 server time.

 

Tester

mrv:
Hi guys, I am running Phoenix live for about 2 weeks now on fxdd, results not so impressive. I noticed one thing when I tried to compare actual trades versus tester trades - I ran tester for the live trading time period. Results were absolutely different, no tester trade matched live trade. This is because of one thing - multiple signals. When we launch Phoenix, it takes first available signal and trades it. You may launch it midday, in the morning or in the evening. But when testing, tester takes data from 00:00 server time, so it may (and probably will) catch different, more early signal. Then this trade will end at a different time that the real one, and tester will get another signal, and in the end you will have totally different testing report. So using Phoenix could be a lot of random walk, having in mind that some signals are good, some are false. Conclusion - if you want your Phoenix to give you results as close to tested as possible, start it at 00:00 server time.

Do you have a suggestion about how to proceed if we have to change versions or if we need to reboot the computer in the middle of the day ?

 
Pcontour:
daraknor,

I sent you back the code on 1/16/2007 at 11:08 pm and I wanted to find out.

  1. Did you get my code changes by email?
  2. What happened with all the changed variable can you copy and paste everything now?
  3. I had a bug fix on the new code. Was I right about that?
  4. I was waiting for confirmation that everything was OK then I could:

    1. put together another converter for 5.7.1 to 5.7.2.
    2. Revise the Optimization Manual.
    3. Update the ultimate guide.
    I had a really busy week so I couldn't have done anything until today.

  • Did you get my code changes by email? YES
  • What happened with all the changed variable can you copy and paste everything now? YES
  • I had a bug fix on the new code. Was I right about that?
  • YES

    So I answered my own question. 5.7.2a is the code I sent you.

     
    m6m6:
    Do you have a suggestion about how to proceed if we have to change versions or if we need to reboot the computer in the middle of the day ?

    Use a higher consecutive signal count and/or start testing on sunday + EA started on sunday. I often use saturday as a start day because nothing is happening.