Statistical arbitrage and pairs trading on Forex

How often will you watch the page and press the button?

Once per 20-30 minutes.

So why not make the page auto refreshing?

Hello everyone! I have made a small upgrade.

Now, you receive two signals in the telegram channel.

The first - READY.
This signal comes when the spread has reached 2.3 or -2.3. Upon receiving this signal, you can set up a trading panel to enter a trade, set the necessary pairs and parameters.

The second - GO.
This signal comes when the spread goes below 2 or above -2. Upon receiving this signal, you can open a prepared trade.
Important!

When you receive a second signal, check the situation in the dashboard.

It is important that the following parameters are met:

1. correlation > 0.8
2. cointegration < 0.05,
3. stationarity < 0.01.

If you received a second signal, checked it on the dashboard forexpairtrading.com. If there is a signal, but the conditions are not met, just ignore it.

Sometimes, the connection between couples can weaken for a while, and then strengthen again.

This is not as easy to do as it might seem. Otherwise, I would have done this a long time ago!
I'm working on this moment, it takes time.

Hello everyone! I got a little sick, so I took a break for a couple of days.

Sorry.
Hope you are alright now.

Hi everyone,

After a lot of time and work, I get to automate and run the first version of a bot based on pair trading, stationarity and cointegration.
Today I get to run the first "stable version" and Im sharing some of my tests and results.

-Im using 1440 periods of 1 minute, but previous to any calculation(ADF and others) I apply logarithm
-My Entry, StopLoss and TakeProfit is based on deviations of Spread:
For entry i use 2 deviations from the mean
For SL i use 1 deviations from the entry
And for TP 2 deviations from the entry or spread <> mean

Today Closed Trades:


Current Open Trades:


Currently im using the API for Python of MetaTrader5 for open and manage trades, but im trying to develop a MT5 version.
Mainly for do a better backtest of the strategy

Please share your EA

@ moment i'm following this

Great work brother . hope we can see one day in mt4 or mt5 format

Telegram account canceled.

And the Website removed bro

Would anyone be able to program an indicator?

Hello to everyone and a warm welcome to every trader using this trading style of cointegration (or statistical arbitrage).

First, as I know there is no longterm cointegration, so you need profit target, stoploss and time stop (at least stop the open trade before red news).

Second, I have some questions. How many trades do you get per day or per week ? Because consistency can only be built from trades. So a few trades per day would be good. In case you only get 1 trade per week in average then this is not for me. Tell me more here, what is your experience in trade frequencies here ?

Third, you could also add some commodities like oil and gold with narrow spreads, maybe 10year T-Note (some brokers offer CFDs on interest rate products here too), and Indices too. I mean you can also compare Nasdaq100 with Dow30 or SP500 or DAX together, to get more trading opportunities. You can test up to 20 indices for themselves looking for intraday cointegration here too, just as an idea. Most brokers have several stock indices to choose from with good spreads. Also USDNOK, USDSEK and some other USD and EUR based exotics do also have narrow spreads for intraday trading when you trade less than a few lots per trade. So for any start it can be considered as well. I might guess USDNOK, USDSEK and USDHUF and USDPLN or USDCZK are also most of the time cointegrated (spreads are just double of any minor pair, but of course spread must be checked before entering any trade). Also USDCAD and Oil, AUDUSD with Gold, SP500 with T-Note, or most major currencies with T-Note too because of the carry trades effect, when considering changes in interest rates. The goal is to have more trade opportunities. But as long you have it tested and traded only for majors and minors, tell me how many trades you got for every week in average so far. Thanks.

Fourth, on the mql5 market I watched the demo on test with IndCorrelationTable from Denis K., but only to see that they were 2 trades within the past 2 weeks, so 1 trade per week in average (and that also from the same pair combi). That is by far too low for me. Also no trade and risk management is included in this indicator. I also have experience with CoIntegration Pair Trading indicator from Lucas B., but there is no search for cointegrated pairs, so very time consuming, and usually the cointegration breaks to that point, when you find a cointegrated pair. So you only get stopped out, when they move out up to 3 to 6 times the z-score. I tried it with no success for 2 months some time ago. Also I could not backtest it. You would need to do this with profit target and stoploss otherwise you are lost here. There is no scanner table for pairs, so beware here.

Fifth, to goal is to have profit consistency every month, otherwise I would be disappointed with that approach. I think it is doable. If you look on intraday opportunities with several trades per day, then you should be profitable on most weeks of any month too. Maybe you have 1 bad or flat month in a full year. But that should it be. Because on stocks you get good sharpes, so in Forex it must be comparable. The approach is very similar. And I also think the currencies are fundamentally more linked together than individual stocks on their industries. So currencies should show their mean-reverting behavior in combination together. Just my two cents here.

Sixth, anyone have some trade results and statistics at all, could you tell me more here ? Because in the past Arbmaker(Plus Pro) existed as software, but it is no more available for purchase so difficult for me to perform portfolio backtests over an extended period. Anyone have a license and want to sell this piece of software ? Any experience with this sofware to share more here ? Does Arbmaker still work with latest MT4 for automated trading ?

Seventh, anyone who can provide an indicator or even an EA with some portfolio backtest capabiltities would be very welcome to do so. I would like to see some stats.

Anyone would like to answer please ? Any other useful comments about profitability and other software options are also welcomed.

Thank you in advance. It is an interesting topic on this forum.

Hi everyone

Trade arbitrage is so hard, and you cant do this manually; you will need automate this, because you need speed for execution and manage.

Currently Im still trying to dev a EA in python for that cointegration using this portfolio of symbols

Ill remove GBP pairs from my strategy soon.

EA looks good, but is so hard to automate, I have like +3 weeks working on this project, and today im running again a test version
(Im no expert in python, so is no a stable version and i cant share this code, but if you know python and wanna continue testing this strategy, let know and we can work hand in hand; find me on telegram @SJPazmin)

This strategy can open around +100 trades per day
I have running this strategy for less of one hour and check
Open Trades:


Close Trades:


And while I was writing this, more positions were opened.
Therefore, I am attaching a report so that you can draw your conclusions.


Regards,
Sebastian

Can somebody explain how exactly the entrys and exit for this trades are calculated?

Dear Pazmin,

as I see the spreads do count as trading cost, so no Exotics of course.
But how much is backtesting performance due to cointegration ? Did you test this ? I mean does it really make much difference having cointegration at 85%, 90%, 95% or even 99% ? Do you get with higher cointegration much higher profit factor and/or higher average net trade profits ?
I would just like to know this fact, because I know that cointegration only last for a very short time period. I also get good results with no cointegration measure at all. I tried it with Hurst exponent but Hurst does not matter for achieving any better performance.
On GBP crosses I like GBP vs. NZD, but spread is high about 0.012% of value, less than 0.01% is good for minors, and less than 0.0005% for majors (inklusive commissions).
What do you think ? Please tell me little bit more about your backtesting results. You do not need to make only forward realtime testing for my questions to be answered. Thank you.

Hello to everyone,

I mean if there is a relationship like KO and PEP in stocks then you do not need to measure cointegration, because you know they are mean-reverting (also short term here). But to measure cointegration short term in FX, is like to say they do mean-revert, because of that statistical measure so just a definition by itself. Of course out-of-sample tests, you can do walk forward tests also in the past on unseen data too, do tell us more here, how long that pair combi has been working. It is like a bit to say, ok they do work mean-revert, because they did so in the past for a certain time and that will continue for some time, but how long you look on that time period when you can say now it is working for longer ? So take you profit target and stoploss and go then. One really need to do some more testing, otherwise you can easy fool yourself. For my testing, everytime I found a relationship it works for very small profit (cost do really matter much, so it is limited in any case) for a long time period at least. In addition I am forced to do somefiltering for higher profitability. That is the point I am looking for cointegration now, but it can be done with other filters too, I found out. I also feel no need to measure cointegration and do the whole nasty calculation at this stage (it is tricky with my backtesting software as it is not built-in as function), because I already know it works. Some academics just use cointegration as a proxy instead of measuring by simple calculation to go in and out here and there, but I feel you need to a proper profitable backtest instead of just relying one statistical measure. The whole concept of cointegration looks a bit strange to me after making some thoughts. Because if there is a fundamental strong relationship it works and you do not need to calc cointegration in addition. But the other way to measure cointegration without having a sound relationship can be just nonsense because this cointegration is just spurious and non robust or stable over time. So it can be tricky to do the math and calculation. I would want to see more here to continue with that approach of cointegration. Any further thoughts from anyone here ? Does one really get outstanding out-of-sample results when relying on cointegration ? Could someone shed some more light here ? Thank you for any comments and some more trade statistics with different level of cointegration tests.

You need to read the whole thread. This is how the author of the thread did it. Over time, he stopped contributing and made all aids inaccessible. Only he has long-term results. The author probably wants to use the system commercially.