Synthetic hedges, cointegration, mean reversion and similar stuff | Page 21 | Forex Factory

Options

Go to
Permalinks
Print Thread

Similar Threads

Simple Mean Reversion 621 replies

Pairs Trading: Reversion to the Mean 22 replies

Forward test of my new mean reversion strategy 11 replies

Question for synthetic and non-synthetic currency pairs 0 replies

Cointegration, Synthetic hedges, mean reversion in R, Tech Thread 0 replies

Attachments: Synthetic hedges, cointegration, mean reversion and similar stuff
Exit Attachments

Synthetic hedges, cointegration, mean reversion and similar stuff

  • #401
  • Jun 15, 2012 3:28am Jun 15, 2012 3:28am
This one also is interesting:
http://www.google.com/url?sa=t&rct=j...AzHq1VWHR5Zv_g
  • #402
  • Jun 17, 2012 9:41am Jun 17, 2012 9:41am
Hi guys,


Basically, i use Johansen Method to estimate the co-integrating relationship aka the slope coefficient aka hedge ratio.

the test does not give evidence of much co-integrating relationship between EUR-> GBP, as compared to the model using lm and then the ADF test.

Anyone using the similar approach that we can discuss?

BTW, i bump into some materials saying "when there are multiple beta (aka relationship > 2), we need to add economic restriction to identify beta. Hence i only stick with 2 beta


Till now, my back test result still has not promised consistent return for both (lm and Johansen) method use. Any one kind enuff to share the trading strategy?


ShinjiOno
  • #403
  • Jun 20, 2012 9:46am Jun 20, 2012 9:46am
Hey Guys,

Anyone still trading this ideas?

Something puzzling me for the past few weeks. I did a Johansen test on most of the currencies pairs, including EURUSD/USDCHF, I can't find any cointegration relationship at all and the cointegrating relationship estimated by the ca.jo(R function) does not work at all. The residuals not gaussian noise.

I thought the normal procedure is:
1. identify the co-integrated pairs
2. fit the model using lm/ pca / johansen method and etc
3. trade the spread

The pairs got to be co integrated to make statistical sense? else, we are just over fitting some datas?

However, 7 bits, thanks for the effort. I really do learn alot from your code. Keep up the good work and thanks for your willingness to share ur code with us

Really appreaciate it


ShinjiOno
  • #404
  • Jul 1, 2012 8:17am Jul 1, 2012 8:17am
I try to use arbomat but it don't work, there is nothing desplayed. It use RDEBUG 2 an the path is good. Here is the DebugView

http://img11.hostingpics.net/pics/521392test.jpg

Thanks
  • #405
  • Apr 18, 2013 11:03am Apr 18, 2013 11:03am
I've read through this thread and activated the arbomat on my computer and am creating an automatic system that tests for correlation and cointegration on HTFs to try to find where the optimal entries are across all currencies. Is there anyone out there still subscribing to this thread?
  • #406
  • Apr 19, 2013 1:54pm Apr 19, 2013 1:54pm
Quoting ambtn723
I've read through this thread and activated the arbomat on my computer and am creating an automatic system that tests for correlation and cointegration on HTFs to try to find where the optimal entries are across all currencies. Is there anyone out there still subscribing to this thread?

all former subscribers woke up one day and found out they were wiped out.
wo-yoy! wo-yoy! wo-yoy! wo-yoi! wo-yoy-yoy-yoy!
  • #407
  • Apr 20, 2013 12:21pm Apr 20, 2013 12:21pm
Quoting ambtn723
I've read through this thread and activated the arbomat on my computer and am creating an automatic system that tests for correlation and cointegration on HTFs to try to find where the optimal entries are across all currencies. Is there anyone out there still subscribing to this thread?
I believe cointegration simply does not work (anymore) on currencies. The work done here by 7bit and others is a great proof of concept and of great acedemic value and I'm sure can be applied elsewhere.
  • #408
  • Apr 20, 2013 2:11pm Apr 20, 2013 2:11pm
I guess this is where I'm having a tough time understanding what went wrong. No matter -- I'll come to my own conclusions and share them with the group if anyone is still interested. Thanks for chiming in Ultimate6 and pemully.
  • #409
  • Apr 20, 2013 10:15pm Apr 20, 2013 10:15pm
Quoting Ultimate6
I believe cointegration simply does not work (anymore) on currencies. The work done here by 7bit and others is a great proof of concept and of great acedemic value and I'm sure can be applied elsewhere.
Hey Ultimate,

I came to similar conclusions a while back but also feel (as you appear to) that further investigation of the concept could yield something worthwhile. If you or any other experienced developers with R experience want to form a work group to compare scripts and test ideas, please send me an email through this forum.

Cointegration is built on the concept of regression of two or more time series in order to find a long-term linear combination that can yield stationary residuals that may be suitable for pairs trading. At the heart of the algorithm is regression or an optimization mechanism for minimizing the sum of the squared differences between the data points and a straight line. The first question that should be asked of this method is whether it is reasonable to assume that a straight line will fairly capture the essence of a nonlinear time series like FX rates. That's pretty much a whopper IMNSHO. However, financial time series do exhibit a property of being difference stationary which may be of interest.

It is very easy through optimization to find solutions that appear stable but don't continue out of sample stable. The second question that comes to mind is whether it is reasonable to assume that by multiplying a pair or pairs by a single coefficient, that the resulting series will be stationary into the future. I think that's quite a lot to ask of a coefficient or coefficients. The SVECM was created as an effort to allow a model to adapt to changing conditions going forward.

The ADF test is a way to determine if the residuals of a regression are stationary. However, it is a weak test as it says nothing about future stationarity. Out of sample testing is a stronger test because it answers the right question regarding future stationarity of a spread rather than historic stationarity.
  • #410
  • Apr 21, 2013 3:27pm Apr 21, 2013 3:27pm
Quoting Ultimate6
I believe cointegration simply does not work (anymore) on currencies. The work done here by 7bit and others is a great proof of concept and of great acedemic value and I'm sure can be applied elsewhere.
I'm not sure how cointegration work well for one set of markets and not another. I don't think that's the issue and I mean no disrespect. What's of issue is that cointegration works *some* of the time and not others. My experience is that the formulas we have are great for range bound activity but not trending markets. After all, all LM methods are based on the foundation that there is a central mean. This cannot hold true over longer periods of time.

I read a fantastic explanation of unit roots as if they were describing a the flow of a river.
Just like the flows of a river, the directional bias of the individual flows of a basket of currencies change. At certain points in the river, the flows are all centering around a central location. AT certain other points they are accelerating away from this point. This is the entire basis of the unit root test.

I haven't studied ADF in depth to understand what its forecasting but based on my basic understanding I do not think this accurately helps us finds unit roots where the ebbs and flows of the river are many. I think we have to go to Johanneson or the other models to get accurate unit roots.

Next Im a big believer in HTF analysis. If we find that a basket is cointegrated on larger (h1+) timeframes, it opens up the possibility of trading on the shorter timeframes when data from these timeframes falls out of synch.

None of this was accounted for in anything I've read so far. Neither was rebalancing, range bound checks, S&R, etc.

Cointegration is a tool. It's not the holy grail. Managing the tool effectively should get us to where we want to be get which is to get an automated bot that eliminates the need to watch the markets constantly and still operate profitably.

FX-- I sent you an email. I'm game.
  • #411
  • Apr 21, 2013 4:41pm Apr 21, 2013 4:41pm
Quoting ambtn723
{quote} I'm not sure how cointegration work well for one set of markets and not another. I don't think that's the issue and I mean no disrespect. What's of issue is that cointegration works *some* of the time and not others. My experience is that the formulas we have are great for range bound activity but not trending markets. After all, all LM methods are based on the foundation that there is a central mean. This cannot hold true over longer periods of time. I read a fantastic explanation of unit roots as if they were describing a the flow of a...
I hear you.. perhaps I just threw in the towel with co-int and fx, but the fx market has become so efficient over the past few years that such relatively "simple" techniques don't work anymore along with FPI /triangular arb etc.
  • #412
  • Apr 22, 2013 9:29am Apr 22, 2013 9:29am
If anyone is interested, ArbMaker pointed me to a recent paper called Cointegration Based Optimisation of Currency Portfolios which contains on pages 16 to 23 some very interesting test results and conclusions. The abstract:

"Cointegration is a technique that has been used for some time to optimise equity portfolios, but there is limited evidence of its application in managing currency portfolios. This research examines whether there is any value to be gained by using cointegration based strategies to optimise currency portfolios that are U.S. dollar, Euro and Sterling based respectively. We build ‘major currency pair (MCP) tracking’ portfolios to replicate the classic index tracking strategy commonly applied to equity portfolios, overcoming the lack of an ‘index’ for currencies by using the most frequently traded currency pair for each portfolio, namely the EUR/USD for the U.S. dollar and Euro portfolios, whilst the GBP/USD is used for the sterling portfolios. We compare the out-of-sample performances of these portfolios to simple benchmark techniques of optimisation. The results are encouraging, with the detection of long-run relationships adding value, particularly for the sterling portfolios. Because of the generally low volatile nature of the cointegration portfolios, they could be leveraged to match the returns being offered to investors by the benchmarks, as they generally offered better risk-adjusted returns. They also enhance portfolio stability.

There are several side issues that are also discussed, such as whether profits can still be derived from applying simple trading rules to currency portfolios, and if so what are the best currencies to invest in and how far should an investor internationally diversify their portfolios. Some of the results found are encouraging, and point towards the added value given by ‘lesser’ currencies particularly the Scandinavian currencies, with the Swedish Krona appearing to boost the risk-adjusted returns of portfolios. Technical analysis is shown to still have a role to play in currency trading."
  • #413
  • May 7, 2013 2:01am May 7, 2013 2:01am
Is this system still profitable? no more update about this strategy at all?
  • #414
  • May 8, 2013 2:44pm May 8, 2013 2:44pm
Exotics are very well co-integrated and stationary i.e USDSEK & USDNOK, USDCZK & USDHUF etc.. However the spreads of these in the retail market are so large that it makes it impossible to make decent profits when trying to arbitrage.

All other pairs are not co-integrated like GBPUSD & EURUSD etc.. these sort of pairs are not co-integrated at all.

I've run ADF tests on almost all tradeable pairs and there's 8-10 pairs exotics that are co-integrated, however again, spread is so large on both that it's impossible to profit well.
  • #415
  • May 10, 2013 2:24am May 10, 2013 2:24am
Quoting drgoodvibe
Exotics are very well co-integrated and stationary i.e USDSEK & USDNOK, USDCZK & USDHUF etc.. However the spreads of these in the retail market are so large that it makes it impossible to make decent profits when trying to arbitrage. All other pairs are not co-integrated like GBPUSD & EURUSD etc.. these sort of pairs are not co-integrated at all. I've run ADF tests on almost all tradeable pairs and there's 8-10 pairs exotics that are co-integrated, however again, spread is so large on both that it's impossible to profit well.
Is there no broker offer low spread for exotic pairs? so do you use this strategy?
  • #416
  • Edited 5:06am Oct 30, 2013 4:08am | Edited 5:06am
Quoting 7bit
The versions on my website are the currrent versions and I use then myself currently. They contain no errors (other than maybe the fact that the whole approach of using a linear regression and hoping for useful results might be flawed). I am still hoping that someone else (who is more experienced in R and in statistics than me) will take the existing code and replace the simple regression with something else or otherwise enhance the script and publish the modified version here. I am doing some experiments myself but have nothing ready for publishing...
Truly wonderful and sincerely much appreciated. Love your work and will post my progress if it adds value to this thread.

One thing I plan to look at is using TLS instead of OLS.

You've earned my vote and voucher, although I can't issue a voucher.

Thanks again, Snow
  • #417
  • Oct 30, 2013 6:02am Oct 30, 2013 6:02am
Quoting Snowski
One thing I plan to look at is using TLS instead of OLS.
TLS=total least square?
Which norm are you going to use to mesure the distance between two points?
No greed. No fear. Just maths.
  • #418
  • Oct 30, 2013 7:29am Oct 30, 2013 7:29am
Quoting Dirtybrown
{image} here's another experiment simply throwing all of the majors together. I run a custom VSA template on the spot dollar index using sigtrader. Seeing that it was in the middle of its range this morning before the NY stock market open, I entered the recommended arbomat trades while the price line was within 1 deviation of the blue center line. So far now, profitable.
From what I gather, I would have traded that the exact other way around.
  • #419
  • Jan 30, 2015 3:04pm Jan 30, 2015 3:04pm
Quoting Ultimate6
library(zoo)
library(tseries)
back<-1000
now<-1
cthis <- 1
x <- scan("c:/regdata.txt")
regressors <- matrix(x,ncol=6)
regressand <- regressors[, cthis]
regressors[, cthis] <- rep(0, back)
y <- regressand
x <- regressors
model <- lm(y ~ x + 0)
summary(model)
stddev <- sd(resid(model)) coef(model)[-1]
//Plotting part
pred <- as.vector(predict(model, newdata=data.frame(x=I(regressors))))
curve <- rev(regressand-pred)
linea <- 0
lineb <- 0
plot(curve, type='l', ylab='', xlab='', main='', sub='', col='cornflowerblue') ......
Hi, could you explain the results of this script?
I see the plot, and I get this line

x2 x3 x4 x5 x6
0.580635548 -0.003192527 -0.001234799 0.995316248 -0.574710354

Does it means that I should Buy 1 lot of first pair, Buy 0.58 lots of second pair, Sell 0.003 lots of third, Sell 0.001 of fourth, Buy 0.99 of fifth and Sell 0.57 of the last pair?

Thanks
  • #420
  • Last Post: Dec 6, 2015 5:27pm Dec 6, 2015 5:27pm
Looks like a lot of people have been thrown off by the new MT4 versions - only seemed to work when compiled in pre v600 versions.

7bit's DLL and MQL files were updated here: https://www.mql5.com/en/code/11112

Helpful instructions for installing by FXEZ: http://www.forexfactory.com/showthre...08#post6545008
2 traders viewing now