Synthetic hedges, cointegration, mean reversion and similar stuff

this doesn't look like a buy to me. At least I would not buy it, but I have not enough experience with this kind of trading. I first created these types of charts just a few weeks ago, so I am new to this too. This is why I am not yet going in with the full money.


btw: you can draw a vertical line into the chart and give it the name back and then you can move this line around to change the time window.

And I have noticed the following: If you switch timeframes sometimes it takes a few bars until it has loaded the recent history in all used pairs, so the last few bars might be wrong until you either wait for the next open or switch timeframes forth and back one more time.

One week on the M5 was not enough. If this goes wrong I deserved the loss.

Could you have a problem with 5M prices where you "create" correlation?

I mean, when there is a downtime at markets close/open, price-action will get really slow, hence repeating the same value for a few candles in a row without much action.

This will shift the correlation factor if you repeat a value while there is nothing going on as they are not significant.

Going with the 1H prices seems more robust.

these are my 2 trades for today. the third is still open. the same trade like 7bit in a very short-term window.

first trade with AUDUSD Basket in range-mode gives me nearly 2R.
the second trade with the GBPUSD Basket in Trend-mode was a burner.

and the third is still open. yes i entered to early too. but now i'm in green pips, because i reentered a new basket with the difference lots to the first basket.

Is there a guide for the techincally challenged....cannot get EA to compile. Downloaded all the files and they are in the correct location.....problem with the Oanda label buttons and I have no idea about this. Help?
Cheers
CanuckCT

this Oanda buttons are for trading with oanda out of MT4. but you'll need an external programm. i think the Oanda API or with a scripted bridge. they do not work now.

but what is your problem? if you get these warnings you could ignore them.
but what you need is the common_functions library from 7bit's homepage. look at his signature. i think this is why the EA did not compile.

I think I just had the same problem as you had Canuck.
With Mediators hint I just managed to solve it.
Download the common_functions libary from here http://sites.google.com/site/prof7bit/common_functions
and put it in the right directory.
Then compile the EA once more and it should work.
Now I can finally start to play arround. Thanks 7bit!!

To get this working:

Install R from http://www.r-project.org/

Download mt4R.mqh and mt4R.dll from http://www.forexfactory.com/showthread.php?t=260422

Download common_functions.mqh from http://sites.google.com/site/prof7bit/common_functions

Copy common_functions.mqh and mt4R.mqh

To Crogram Files[WHATEVER TRADER]expertsinclude

Modify the EA to point path for RTERM:

#define RPATH "C:/Program Files/R/R-2.12.0/bin/i386/Rterm.exe --no-save"

Copy mt4R.dll

To Crogram files[WHATEVER TRADER]expertslibraries

Hope this helps.

Regards,
Vip

When you get in a trade, you buy/sell the basket accordingly to the chart, but then do you re-ajust the lots as new ticks come in?

Also, do you have it logged to Oanda, if so, how?

Finally, I guesss you just test some random pairs together and when you see a nice pattern with a spread between 1 and 2 stdev, you get in?

Looks like going 5 minutes bar with only 400 values is kinda risky statistically. Did you get good returns in terms of pips? How many lots were involved?

Also, did you try with both the range and trend mode some random baskets until one was looking good enough?

If you guys do look at some random baskets on random timeframes as I'm doing now as I get comfortable with the EA, maybe we could all help eachother and share our results in terms of trading hours, basket composition and timeframes.

OK....I have this completed....all seems good. Now what.
Cheers
CanuckCT

I readjust the lot sizes, this is my attempt to have something like a "soft stoploss", it will reduce the coefficients (and my exposure) for the pairs that go into the "wrong" direction too much. This seems to give me a chance to at least break even on this trade (and this idea seems to work). Overall the lot sizes that I use are so small that I can sleep at night without nightmares.

The oanda thing needs an external application, if you click (and move) one of the oanda buttons it will create a text file (a ticket) in the oanda_tickets folder (this folder acts like a message queue) and my external application will pick it up from there and trade what is written in this ticket and delete the ticket. The tickets contain relative changes to the exposure (MT4 keeps track of what position sizes in total there SHOULD be open at oanda and only creates a ticket about the difference. The oanda software will then buy or sell the difference. From time to time I make sure that they are both still in sync but I never see any differences. I have no plans to make this external oanda software available.

I test many combinations on many timeframes. M5 seems too dangerous, M15 and upwards seem to be more promising. On H4 and upwards it becomes very stable and nice regular cycles can be observed (but also slow to trade, you cannot force an entry at any time, you have to wait). I try to use not more than 4 pairs on the lower timeframes (smaller than H1) to save transaction costs.

I move the back line back and forward and also switch timeframes on the same pairs and try to judge from what I see (how much the coefficients change, how regular the spread curve looks and how it changes when coefficients change) until I find something that I find promising. This is not easy to quantify, it does not only have to be stationary, it also has to somehow look "attractive" to me.

When I trade the M15 or M5 with a smaller time window (because I tried to "enforce" an immediate entry because of my impatience) I should(!) also compare it with bigger time windows to see whether it might just be in the process of breaking a range or bouncing or doing strange things on a higher timeframe and whether there might be a trading direction that should be avoided, but I have no exact procedure for this and don't yet know whether this actually makes any sense at all.

I cannot give any detailed advice how to trade this or even a complete trading strategy, I might be good at scripting MT4 but I am still a miserable trader. These synthetic baskets are as new to me as they might be to you. I have created something of which I do not yet know exactly how to control it (Frankenstein Syndrome: Crazy mind, crazy idea -> monstrous creation, powerful and not easy to control ).

Berndt,

It may be "...Frankenstein Syndrome: Crazy mind, crazy idea -> monstrous creation, powerful and not easy to control http://cdn.forexfactory.com/images/s...m/thinking.gif..." at this stage but what you have created is something wonderful and many, many people are very interested in this. Your tools are extraordinarily nice. (My Matlab stuff is very hard for new people to use as it is very unfriendly so there is no point posting it.)

With your ingenious mind, this could go a very long way!

Very best regards,

Old Dog

Hello 7bit,

what i like to test in the next days is, if i change in trendmode the slopestrength and from your EA i'd like to find out how to calculate basket with other pairs. like JPY, GBP, CHF pairs etc. after that it could be possible to write a scanner with R which combines all posssible Baskets and gives signal if a basket is at 2nd Dev.Line. at first i would take baskets with only three pairs.

maybe it is also possible with R to scan if some baskets have an oscillating nature.

This won't have an effect, if you double the slope, it will simply half all coefficients proportionally and vice versa.

I have a new version that I will post soon that can fit the model in a different way and might lead to more stable coefficients. This might be useful for the trend mode and it has also a different behavior in range mode.

At the moment I do the regression with the prices directly but in the books is written the regression should be run on the first difference of the prices. This has a huge impact: Currently it is able to force almost *anything* into a horizontal range, no matter what it is and whether there is a genuine relationship or not. If I do the regression with the first difference then it will try its best to hedge the ups and downs of the individual candles but not care so much about the resulting overall trend, it will then try to make it as smooth as possible. The coefficients tend to remain quite stable over time. In the ranging mode this can either lead to something that is really stationary (cointegrated) (doesn't happen often) or something that is trending but in a much "smoother" or much more regular and predictable way than any individual pair. The resulting curves look quite interesting. I will post it soon.

this sounds very interesting. but for now i have to be patiently waiting.

what kind of books did you take your knowledge?

With "written in the books" I meant what seems to be the commonly accepted doctrine when I research this kind of things on the www. I didn't study this or have any real (paper) books about it, I am only an enthusiastic amateur who tries to learn more about these things through reading what other people write and making experiments like the ones in this thread.

i made some findings while trading in trend-mode. but i do not know if it's good or bad:

i trade GBPUSD vs. EURUSD & USDJPY. after a while the trades goes against me and i adjusted the positionssizes like it was displayed in the graph. i had have to partial close loosing and winning positions. after this nothing happens. the equity and the balance remains the same. only the used margin lowers to nearly the half. so i thought about this. two scenarios could happen now:

1. the spreadline goes directly to the meanline and everything is perfect
2. the spreadline moves farther away and i had to readjust again. but the more it moves away the lower the lotcoeffcient shrinks.

what i can do now? from this point i could double up my positions to scale into better price. or leave it and hope that the mean line does not fit too fast to the breakout of spreadline. if this happens the spreadline have to cross the meanline an have to move more to breakeven my basket.
or i double up and the spreadline only have to move to meanline even if the meanline fits to the breakout.
the worst case is that it move away more and more. but how great is this probability? i think it exists, but very seldom. what i can also do is to stack up my positions if it move in my favoured direction. this would gain profit exponentially.

i decided to double up and wait what happens.

another thought would be. what happens if i never close the upmove basket and reentry a downward basket? is this hedging the hedge or could this lead into more and more profits because if a new upmove basket triggers i would have then two upmove and one downmove basket. hmm many things to try out in the future.

Mediator

Hi Bernd,

This is interesting. Thanks for sharing.