magic shift | Page 3 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Magic numbers question 12 replies

What does GMT shift and GMT shift Sun do in this indicator? 0 replies

magic is back! 11 replies

Magic Numbers for Mutiple EA 9 replies

Attachments: magic shift
Exit Attachments

magic shift

  • #41
  • Nov 26, 2009 8:16am Nov 26, 2009 8:16am
I try to optimize one year and then use optimized paremeter for th next year (H4 chart) results are negative. So for me so far it's nothing more then curve fitting

Quoting 7bit
a shift found by fitting it to only 3 months H1 data worked also during 4 other months. This should also work forward.

I just started a new forward test on EUR/USD H1, magic_shift found by fitting it from March until now is 192, positive correlation (long when positive, short when negative).

I will post updates whenever it opens a new trade:
Attachment
  • #42
  • Nov 26, 2009 8:20am Nov 26, 2009 8:20am
Quoting Koyco
In your first post you recommended that if AC was neg. go long but in the 1h. chart you appear to do the oppersite is this the case?
Yes. 61 with negative correlation as mentioned in the first post works perfectly for EUR/USD, H4. This one is the same system on H1 with parameters adjusted so that they work for H1.

Since not only the shift is adjustable but also the sign of the correlation there may also be markets or timeframes where the best shift has a positive correlation.
  • #43
  • Nov 26, 2009 8:27am Nov 26, 2009 8:27am
Quoting Dezil
I try to optimize one year and then use optimized paremeter for th next year (H4 chart) results are negative. So for me so far it's nothing more then curve fitting
Try it 1 month at a time But yea I agree it's curve fitting unless u use previous optimizing for future test. You can't for example optimize for November then use the best setting and test November. I can get every ea to look good that way
  • #44
  • Edited 9:22am Nov 26, 2009 8:32am | Edited 9:22am
Quoting Dezil
I try to optimize one year and then use optimized paremeter for th next ear (H4 chart) results are negative. So for me so far it's nothing more then curve fitting
Of course it is some sort of "curve fitting"¹, but not in the negative sense of this word. Every System is created by fitting a model to the reality, some work for years, some stop working when the market changes and new models or new parameters must be found. If you had read the thread you knew that this system doesn't work for infinite time. Like any other Systems it's performance must be constantly monitored and it must be fitted to *current* Market conditions, not the ones 2 years ago.

A working parameter set (only one parameter in this case) may continue to work for some more time or it may not. There is reason to believe that certain lags stay intact longer than the period over wich they were fitted. The current shift parameter would have been found already in September or October and it would have worked perfectly in November and I will continue to ride this horse until it is dead.

---------
¹ Maybe we should first define what "curve fitting" really is. The usage of this term amongst traders is usually seen when they really want to say overfitting. Overfitting when fitting a model is a problem, but the fitting of a model itself is always a needed step. Every Model must be fitted to existing data before it can be used to apply it to new data for the purpose of prediction, and if it cannot be fitted then the model is wrong. If http://en.wikipedia.org/wiki/Overfitting is what you really mean when you are saying "curve fitting" then you are wrong, this model simply can't be overfitted. The only remainig concern is the possible "multiple hypothesis problem", but i hope i can rule this out by the big number of observations vs. the number of tested hypothesis and by better validation techniques.
  • #45
  • Nov 26, 2009 8:53am Nov 26, 2009 8:53am
Quoting 7bit

A working parameter set (only one parameter in this case) may continue to work for some more time or it may not. There is reason to believe that certain lags stay intact longer than the period over wich they were fitted. The current shift parameter would have been found already in September or October and it would have worked perfectly in November and I will continue to ride this horse until it is dead.
The problem is how to know tnat you are riding dead horse. And how mush you will loose while it will be dieing
  • #46
  • Edited 9:39am Nov 26, 2009 9:10am | Edited 9:39am
Quoting Dezil
The problem is how to know tnat you are riding dead horse. And how mush you will loose while it will be dieing
This very same problem appears with almost every mechanical System. It also happens when you are just riding a trend: When will it end? The easiest way to find out is to watch for signs of reversal, see how the equity curve looked at the end of such trends in the past (this is some sort of trend also, only in another domain), watch it closely or trail a stoploss on the equity (I'm already in profit) and then at the right moment simply pull the plug.

Another possibility would be to constantly try finding an even better shift by always searching the recent few months and seamlessly switch over to the new shift as soon as you find one. If you identify past "events" in the market where most or all shifts were suddenly invalidated and completely new ones appeared (such as last september) start your search only after such an event. Try using only shifts that produce smooth equity curves without big jumps, long ranges or deep drawdowns.
  • #47
  • Nov 26, 2009 10:46am Nov 26, 2009 10:46am
Great strategy. No one can't deny it. This is the first time I have ever seen with my own eyes a method that actually works in the long run.
Thank you 7-bit.

And if someone is having doubts about the results, I've attached the results from my modified program(MM included, otherwise same idea). wow
Attached Image (click to enlarge)
Click to Enlarge Name: MyTestResults.gif Size: 8 KB
  • #48
  • Nov 26, 2009 12:45pm Nov 26, 2009 12:45pm
Quoting 7bit
s lags?

I agree that out-of-sample testing is the only way to tell that the model is really worth anything. A few months ago...
Cool, looks like you've already got that issue covered then.
  • #49
  • Nov 26, 2009 12:52pm Nov 26, 2009 12:52pm
Quoting 7bit

Another possibility would be to constantly try finding an even better shift by always searching the recent few months and seamlessly switch over to the new shift as soon as you find one. If you identify past "events" in the market where most or all shifts were suddenly invalidated and completely new ones appeared (such as last september) start your search only after such an event. Try using only shifts that produce smooth equity curves without big jumps, long ranges or deep drawdowns.
To me, this seems like the way to go in a method like this. If your frequency of testing & replacement of the parameter value with the new "champion" value is less then the frequency at which the average value stops working then you should be constantly ahead of the game.
  • #50
  • Nov 26, 2009 12:56pm Nov 26, 2009 12:56pm
Quoting serpo
Great strategy. No one can't deny it. This is the first time I have ever seen with my own eyes a method that actually works in the long run.
Thank you 7-bit.

And if someone is having doubts about the results, I've attached the results from my modified program(MM included, otherwise same idea). wow
Nice results. What mm modifications did u make?

Mind sharing your version of the ea?
  • #51
  • Edited 1:28pm Nov 26, 2009 1:02pm | Edited 1:28pm
Quoting serpo
Great strategy. No one can't deny it. This is the first time I have ever seen with my own eyes a method that actually works in the long run.
Be careful with the words "in the long run".

I believe that there is nothing that will work forever. However, there will always be something that currently works and will continue to work at least until tomorrow or even until next week.

If a system stops working you need a new one.

This whole magic_shift thing is an attempt to have something that will create a new working system for you (one that will keep working for at least some time) whenever you press the start button in the optimizer.

Instead of having to read all the 1000s of systems threads here again and again everytime one of the systems stops working, you simply search a new valid shift number for this one-fits-all system and continue profiting a few weeks/months. This is my plan, time will tell whether it can work this way or not.
  • #52
  • Nov 26, 2009 1:26pm Nov 26, 2009 1:26pm
Attached is shifted AC (shift is a parameter) for visual backtesting. From the first glance there is no definite system, but I might be wrong.
Attached File(s)
File Type: mq4 Accelerator_Orest.mq4   3 KB | 444 downloads
  • #53
  • Nov 26, 2009 2:45pm Nov 26, 2009 2:45pm
I'm wondering how I can modify the EA to automatically optimize itself.

I don't want to have to go as far to build my own optimization engine in the EA, but it looks like I don't have a choice.
  • #54
  • Nov 26, 2009 3:10pm Nov 26, 2009 3:10pm
Attached a screenshot from the forward test on H1 (modified parameters for H1)
Attached Image (click to enlarge)
Click to Enlarge Name: magic_forward.gif Size: 18 KB
  • #55
  • Edited 3:36pm Nov 26, 2009 3:13pm | Edited 3:36pm
Quoting Ronald Raygun
I'm wondering how I can modify the EA to automatically optimize itself.

I don't want to have to go as far to build my own optimization engine in the EA, but it looks like I don't have a choice.
I have seen a library somewhere on mql4.com that does this by remote controlling a second MT4 instance.

I am already thinking about writing something similar for myself but using only one MT4 instance (maybe using AutoIT scripting or directly using Win32 API) and sending keyboard and mouse events to MT4.

Currently I do all my optimizations (i have some other EAs that are fitted to very short lived behaviour and need dayly optimization) by hand. Once a day at 12 o'clock (when i am awake) it is optimization time while the coffee machine is running.
  • #56
  • Nov 26, 2009 3:28pm Nov 26, 2009 3:28pm
the original version on H4 is still long:
Attached Image (click to enlarge)
Click to Enlarge Name: magic_h4.gif Size: 20 KB
  • #57
  • Nov 26, 2009 3:43pm Nov 26, 2009 3:43pm
I don't understand what's the problem with MT4 optimizator. The EA is super-simple and this is just one parameter. Besides you don't need to optimize it like every day even if you use it on H1. I would say weekly re-optimization is your best choice. More on that - I would watch the chart and notice if general trend
was changed or significant replacement was in action. This would be my re optimization start point. Also I think 12 month is a way too much data, I would take last 4 months, re-optimize till last date, then I would take last month and do the same. Notice if there is any significant difference in shift value between these two. If there is then currency's behavior has been significantly changed in between. My point is it's hardly fully automatic process. BTW, forget about MT4 down loadable data for optimization, only live broker's ticks you acquired.

Quoting Ronald Raygun
I'm wondering how I can modify the EA to automatically optimize itself.

I don't want to have to go as far to build my own optimization engine in the EA, but it looks like I don't have a choice.
  • #58
  • Nov 26, 2009 5:03pm Nov 26, 2009 5:03pm
H1 Test: A win and a loss, now short for a while.
Attached Image (click to enlarge)
Click to Enlarge Name: magic_forward.gif Size: 18 KB
  • #59
  • Nov 26, 2009 10:16pm Nov 26, 2009 10:16pm
Quoting 7bit
Be careful with the words "in the long run".

I believe that there is nothing that will work forever. However, there will always be something that currently works and will continue to work at least until tomorrow or even until next week.
Wrong words, I admit.. But still I have never before seen an equity curve performing so smoothly in 11 months test period with over than 100 trades (133 actually). Even though the parameter was optimized, yeah...

Any thoughts about the magic shift number 61?
I optimized the EA too from 1.1.09-now and let the program test parameter values of the shift ranging from 0-300 and the results were a bit strange (like they all are.) I'll attach the results later since the optimizing took about 40mins and I screwed up the saving part.

Someone asked for my modified EA. If the original contributor doesn't mind, I attached it too.
Attached File(s)
File Type: mq4 AC.mq4   < 1 KB | 396 downloads
  • #60
  • Nov 27, 2009 2:34am Nov 27, 2009 2:34am
Interesting idea 7bit, thanks for open this discussion... maybe it will be a great forex strategy system!

Quoting serpo
... Someone asked for my modified EA. If the original contributor doesn't mind, I attached it too.
I dont know why serpo, but i have tested your expert (with 61 and 62 shift), and it gives me the best results. Is there is any way to disable mm on expert?

Thanks!
0 traders viewing now