Highly correlated pairs and standard deviation | Forex Factory

Options

Permalinks
Print Thread

Similar Threads

How to calculate standard deviation? 7 replies

Standard Deviation Histogram 1 reply

ATR and Standard Deviation 10 replies

statistics and standard deviation 12 replies

Standard deviation indicator 3 replies

Attachments: Highly correlated pairs and standard deviation
Exit Attachments

Highly correlated pairs and standard deviation

  • #1
  • First Post: Aug 6, 2016 1:11pm Aug 6, 2016 1:11pm
Hello everybody,

Today I want to waste your time with my idea of codding EA for mt4 based on my strategy. It is kind of statistical arbitrage but based on my previous research, study, trading and so on. I am using some kind of this strategy for trading stock pairs (so fair profitable) and now I want to code an EA for automatic trading on forex (MT4 or MT). So here is the idea:

Standard deviation on high correlated pairs

To start, I want to trade two high correlated pairs against each other. I am thinking about AUD/USD vs. NZD/USD (also wti crude vs. Brent crude is possible). The whole idea is based on selling one pair and buying another one.

Untitled2.png

Maybe someone of you remember sfdisparity are somewhere here on the forum. I dont know how it works, but my EA should be little different. It will be not trading gaps we see on the screen when we put two graphs together. It should sell one and buy another one when the ratio of AUD/USD price and NZD/USD price is higher or lower than 1.5 x their standard deviation.

For better understanding look at this example.

Untitled.png

This example is based only on one month daily prices of AUD/USD and NZD/USD. As we can see, I want to enter when ratio reached 1.5 x standard deviation above the median ration. Then I want to sell Aud/Usd and buy Nzd/Usd. Exit is based, when ratio reach median value. Also another stiuation is, when ratio reached 1.5 x standard deviation under median ratio value. Then I would like to buy Aud/Usd and sell Nzd/Usd. Also I will not buy same value of each trade. In first trade AudUsd price is 0.74700 and NzdUsd price is 0.69970, so I will sell for example 10 lots of AudUsd and buy 10.68 lots of NzdUsd.

It is just a simple example, to show the idea. And now I want to ask you all of the experienced traders, some important points:
- What do you think? Is it possible to trade this way?
- Isnt it just trading Aud/Nzd?
- Is the calculation of ratio based on AudUsd /NzdUsd right?
- Is the idea of different value of trades right?
Fell free to ask anything you want. My native language is not English, so if you find some misunderstanding or mistakes excuse me.

Thanks Ondrej
Attached Image (click to enlarge)
Click to Enlarge Name: Untitled.png Size: 179 KB
Attached Image
  • #2
  • Edited 8:37pm Aug 6, 2016 8:06pm | Edited 8:37pm
Hello,
I'm actually thinking about a similar strategy, so your topic sounds very interresting. I'll go directly to your last point. I came recently to the conclusion that, because you expect to exit with a profit when the pairs re-correlate, it is necessary to buy/sell a little bit more/less of the pairs, depending on the ratio.

For the first situation, when correlated, AUDUSD is approximately equal to 1.068 NZDUSD.

When de-correlated above median, we have AUDUSD / NZDUSD = 1.068 * 1.5 (AUDUSD & NZDUSD in current decorrelated prices)

We expect the 1.5 ratio to revert back to 1. Am I correct?

Let me write this differently:

When decorrelated, we have (1+delta1)*AUDUSD / (1+delta2)*NZDUSD = 1.068 * 1.5 (AUDUSD & NZDUSD prices when correlated)

We don't care about the exact values of delta1 & delta2, but we can say that obviously delta1>0 and delta2<delta1 and when recorrelating, delta1 will diminish to 0 and delta2 will grow or diminish to 0. As it is smaller, it will diminish less than delta1.

For me this means we need to sell a bit more AUDUSD and buy a bit less NZDUSD in order to get a profit. I'd try to sell 10.03 AUDUSD and buy10.65 NZDUSD for example.

Let me know if anyone thinks I'm wrong and better let me know why
  • #3
  • Aug 11, 2016 6:06am Aug 11, 2016 6:06am
Yes fredix, your idea is right.
Today I am going to do manual backtest on 1 minute closing historical data form January 2016 till today. After I finish I will post here the file with the results and then we can move forward to details how should this strategy works.
  • #4
  • Aug 11, 2016 7:08am Aug 11, 2016 7:08am
Could be worth exploring further.

Just a quick note on your excel template: You are using in-sample fit by taking the average of the entire column. Only use data available up to that day.

Best,
  • #5
  • Aug 12, 2016 12:25am Aug 12, 2016 12:25am
Quoting stvorka
Hello everybody, Today I want to waste your time with my idea of codding EA for mt4 based on my strategy. It is kind of statistical arbitrage but based on my previous research, study, trading and so on. I am using some kind of this strategy for trading stock pairs (so fair profitable) and now I want to code an EA for automatic trading on forex (MT4 or MT). So here is the idea: Standard deviation on high correlated pairs To start, I want to trade two high correlated pairs against each other. I am thinking about AUD/USD vs. NZD/USD (also wti crude...
Because AUDUSD & NZDUSD have different tick value, the calculation for adjusted lot size to normalizes tick value makes sense.
But in my opinion selling one and buying the other is just a synthetic AUDNZD position, as the positive and negative USD potions cancel each other out.
For that reason, could you not simply apply Bollinger Bands to the AUDNZD and trade tags of the bands in direction of mean revision?

I'm aware this concept has been explored before in regard to Brent & crude oil decorrelations.
-Instead of trading whether or not price value will go "higher" or "lower"- it's trading whether or not decorrelation will "continue" or "reverse". Perhaps that is easier to do as statistics indicate so I believe the idea has promise.
  • #6
  • Apr 26, 2019 7:31pm Apr 26, 2019 7:31pm
Quoting stvorka
Hello everybody, Today I want to waste your time with my idea of codding EA for mt4 based on my strategy. It is kind of statistical arbitrage but based on my previous research, study, trading and so on. I am using some kind of this strategy for trading stock pairs (so fair profitable) and now I want to code an EA for automatic trading on forex (MT4 or MT). So here is the idea: Standard deviation on high correlated pairs To start, I want to trade two high correlated pairs against each other. I am thinking about AUD/USD vs. NZD/USD (also wti crude...
Do you want to encode EA for this system? Are you a coder?
Please excuse the bad English via Google Translate.
  • #7
  • Apr 28, 2019 8:32pm Apr 28, 2019 8:32pm
been waiting for someone to do that... not likely doable manually since it is very small bites so needs to be an EA in order to work one would think.......interested...subbed.
  • #8
  • Apr 29, 2019 12:51am Apr 29, 2019 12:51am
Quoting dagoods
been waiting for someone to do that... not likely doable manually since it is very small bites so needs to be an EA in order to work one would think.......interested...subbed.
I just wanted to point the coder to a better idea
https://www.forexfactory.com/showthread.php?t=911103
Please excuse the bad English via Google Translate.
  • #9
  • Edited Apr 30, 2019 8:51am Apr 29, 2019 11:07am | Edited Apr 30, 2019 8:51am
Quoting dokopy
{quote} I just wanted to point the coder to a better idea https://www.forexfactory.com/showthread.php?t=911103

I do not know anything about coding. You misunderstood.

I'm curious, what makes you think the other thread is a better idea? Bollinger bands work off of a standard deviation. it is essentially the same idea, no?

What i was referring to was if tow pairs are truly very highly correlated, then there is few pips between any divergences that will occur, and it could happen very fast. That would require an EA. for example in the past EU and Ucf were almost 100% correlated in their polar opposite movements.

so doing something with E/U and U/Chf... since they barely separate historically would be difficult to do something of this sort manually.


The other option is to try to use pairs that are not as correlated BUT the correlation indicators out there suck. it poses all kinds of issues if the paris you're doing it on aren't actually very correlated in a stable manner..

I'm interested in learning from the thread creator how he addresses the inevitable issues that come up with this kind of strategy.

The BB gives a nice visual though. If you are going to do what they are doing int he other thread, I'd think a dashboard would be helpful. That way you can see when multiple pairs are in the Bollinger band outer limits at the same time. Then once the multiple pairs that are in the outer bands, start turning over to start returning to their means, you could try to predict the next one to revert to its mean.

The theory of that strategy makes more sense because the entirety of the forex market is zero sum and is correlated somehow someway at all times. I'm not sure it can be identified for profit though with so many pairs. How you identify actual correlation, and the timing of taking advantage of the high probability of returning to the mean, is key one would think. But i'm not any kind of expert by any means at the stuff, just an interested student here.

cheers
  • #10
  • Apr 29, 2019 4:51pm Apr 29, 2019 4:51pm
Quoting dagoods
{quote} I do not know anything about coding. You misunderstood. I'm curious, what makes you think the other thread is a better idea? Bollinger bands work off of a standard deviation. it is essentially the same idea, no? What i was referring to was if tow pairs are truly very highly correlated, then there is few pips between any divergences that will occur, and it could happen very fast. That would require an EA. for example in the past EU and Ucf were almost 100% correlated in their polar opposite movements. so doing something with E/U and U/Chf......
Too bad you're not a coder. Manual trading by https://www.forexfactory.com/showthread.php?t=911103 is profitable for H1 and H4.
Please excuse the bad English via Google Translate.
  • #11
  • Jun 4, 2019 3:47pm Jun 4, 2019 3:47pm
Any progress so far on this front?
  • #12
  • Last Post: Jun 4, 2019 6:45pm Jun 4, 2019 6:45pm
Quoting dagoods
{quote} I do not know anything about coding. You misunderstood. I'm curious, what makes you think the other thread is a better idea? Bollinger bands work off of a standard deviation. it is essentially the same idea, no? What i was referring to was if tow pairs are truly very highly correlated, then there is few pips between any divergences that will occur, and it could happen very fast. That would require an EA. for example in the past EU and Ucf were almost 100% correlated in their polar opposite movements. so doing something with E/U and U/Chf......
EURCHF chart, put on some Bollinger bands, when price tags a band trade back towards the 20 sma. It is the same thing.
0 traders viewing now