Simple High Profit low Drawdown Triangular Arbitrage

@!iii!
I realise there is lag and slippages, but I do not have funds to go live so my only options are backtesting and demo.

Unluckily or luckily my demo account has got a 300 to 400ms delay, so at least that will test a bit of slippage.

Lot sizes are important and I found anything below $1000 is risky. Another reason I can not go live. I can write the EA, but I can not use it.
At least it is fun so far.

It will remain theoretical then, unfortunately.

Nevertheless, your thread brought me back to my own studies into identifying the currencies causing the imbalance and finding the pair(s) to trade it - ideally only one. Both the arb and accumulated spread calculations result in timeseries of stationary nature and it is just too tempting not to try do something with that...

Again, good luck with your project.

@!iii!
Thanks and thank you for your input

Very last one: you mention your lag of 300-400ms. Your aim must be to get that down to zero. With the exception of your client receiving ticks, all your procedures in MT4 are sequential. You can get around the tick based refreshes by using a milliseconds based timer event in your EA - with the risk of using stale ticks also. In the time when the ticks and your orders travel from and to the trade server, prices change. No matter how close you are to the trade server, you are always seeing a broker digested past.

Cheers.

@!iii!
I know, but as I said I see that 300ms - 400ms delay as a good thing for testing purposes.
Similar to slippages.
By the time the buy/sell order reach the broker the price has most definitely changed.
Therefore I see this as worst case scenario.
If I can make it work on a bad server like this then it will work even better on a good server later.

I will never run a live test on this lag server, just tests.

I always program for the worst, rather than the best and in this case I would sacrifice some of the profit if it make the program more stable and able to survive bad times.

In any case.
This is what I have to work with and even though it is slowing down development, it is better than nothing and I can make do. :|

Thanks for sharing. I like your idea very march. Do you try to use another pairs for trading? For example : EURJPY EURUSD USDJPY.

slippage and spread will kill us. i think worth to try

@GregOr
I will try others later.
I first want to get this program finished and stable by focusing on one triangle.

The program is already compatible with EURJPY EURUSD USDJPY and any others, but I am already struggling to do all the coding inbetween other work to fund this project, so I would rather focus on a few things at time and once everything is stable then I will Add the option to do other triangles.
My idea was to use the triangle that is part of the chart you attach it to, but that is something for another day.


@dragonexpert
Absolutely, mostly spread.
So the lower the better.
I am now busy building in some buffers to cater for higher spreads, but that does affect the profit a bit, but worth it.
Still months away from a final live EA at this rate.

I had an itch to scratch.
Some interesting findings.
I tested the leverage affect.

Keep in mind that this version of the program is a few versions newer than the one in post #1.
I will add it or a better version a bit later.

Unfortunately my account only go up to a 200:1 leverage.
Starting with $1000.

Time period tested between 2015.06.23 --> 2016-03-23

Leverage 50:1
End = $4013.61
Max Lots = 0.43
MaxoutDte =
Draw Down = 09.087307875138648%

Leverage 100:1
End = $18,546,35
Max Lots = 3.43
MaxoutDte =
Draw Down = 18.18960949917412%

Leverage 200:1
End = $107,431.65
Max Lots = 38.53
MaxoutDte =
Draw Down = 37.456292074711%

Interesting work man, really good.

Of course, higher leverage means higher profits, but bigger risks.

Nonetheless, you realize you have here very good results, that all traders could only dream of.

I think it worth a try, with spreads, etc ...

@crashout
Thanks.
I actual improved profit since previous post by 50% and draw down decreased by 5%.

I still insist you NOT use the EA on a live account yet.

I can monitor the EA and compensate for any unforeseen problems while the current EA is incapable of doing it herself.
And unfortunately I have no funds to test it live at all and what I have is focused on the development of this project.

Once ready I will tell all and you can use it freely, but that might take quite some time.

OP,

Well done on making a good EA...

...however, (I hate to be the bearer of bad news) this has been tried many times on metatrader going back to '06, and it doesn't work. You might--maybe have a chance with MT5 asychronous orders on the best ECN broker in the universe, but even then you're competing with hedge funds with real big-boy HFT platforms running on colocated servers and full LP book access with sub-millisecond transaction times.

A couple of things to note:

1. If there are any draw-downs (at all) then the lot sizes are miscalculated (all currency units aren't hedged) and it's not arbitrage. In this case you're better of trading those units straight up to free up margin and reduce swap.
2. Trades are not entered into arbitrarily. The fractional product of inefficiency must first be calculated for bid and ask and then run through an algo that can determine whether the spreads and commissions can be covered by a relative swing in the opposite direction.
3. The back-tester sucks at multi-currency (even MT5 which was designed for MC). In theory this is how arbitrage works, but with the back-tester it's garbage in garbage out. Backtester feeds in simulated ticks. You will never see these tick disparities IRL.

Back in the day we were using this system to arb swap by finding market neutral hedge rings with net positive swap, but those day are also long gone...

Here is the OG thread that started the rush to this type of system... https://www.kreslik.com/forums/neoti...ble-hedge-t307

@Nicholishen
Thanks for the information and I will definitely look into the material you supplied me.

I am still not at the stage where I can go live, but I am almost at the stage where I can start running on demo accounts with hopefully realistic price feeds.
The slippage and missed trades parts can only be tested on live accounts, but I have already build in buffers to cater for those.
If the demos can prove the potential for market inefficiency, similar or maybe even less than the back tests then I will be happy.

Still.
I love calculations and coding so I will be pushing this until I hit a brick wall and have conclusively proven it can not be done or alternatively it can.
In theory there must be inefficiencies in the live environment.
That to me is not a question, but my question is if the inefficiencies are big enough to make a profit on.
So far even bad spread demo back tests are making profits so I am hopeful the demo/live price feeds can reflect some of these profits.
I'm still far away from finishing all my coding and tests seeing that it is very difficult for me to take of work during the week without funds to support me, so this is an after hours/weekend project, which will eventually be done.

Let hope and see

Please everyone.

I have received a LOT of PMs about how people want to run the EA on this and that and I am grateful you find it interesting, but please note that I am one person doing a lot of work and coding like crazy as I get time.
For now the EA was developed on MYFXCHOICE's pro account on a 1:200 leverage and worst of all it ignores swap and some high spreads.
This new version I have not uploaded yet takes these and more into account.

Remember the theory is based on a perfect world without spread, swap or commission.
Now for the amount of profit it generates those losses are small and should not interfere with the profit, but the EA does not know that and that is were all the coding time goes into is for the EA to realise that this is not a perfect world and to take it into account in its calculations or risk getting stuck or making bad choices.

Note that the EA in post 1 is very unstable.
I have spent all of 1 day working on it during the weekend.

It might be good for back testing it on myfxchoice's pro demo account on MT5, but not much more.

I have been working on it to the disappointment of my wife so I do have a much better version almost ready.

I tested the EA on my fxpro account, which has got no commission, but bad swap and spreads.
It kills the poor EA, which does not take swap into account seeing that fxchoice's swap is negligible.

I just taught the EA to look at current swap loss and ignore it, and it runs much better and into profit, which the one in post 1 most certainly will not as it get stuck in a swap loss loop it knows nothing about.

Problem is that triangle arbitrage profit/loss is very little and stable.
Therefore if I do not take into account swap then it will go down bit by bit, but enough to prevent to open position to go back into profit.
Now in the grand scheme of things swap is but a minor part of profit.
So by ignoring it I lose a little profit on one ticket, but then the ticket can close as normal at a profit-swap(witch could be a small loss) and then allow for other tickets to be opened after it to pick up the slack and ultimately generate more profit.

This strategy need lots of trades to profit.
So being stuck because of swap is a NO NO, which I did not cater for.
I'd rather close a long lived trade in loss than have it hold up other more profitable trades after it.

Still a lot of stuff to do.
Will update the forum once this EA it ready to be released.

WARNING: DO NOT USE THE EA ON A LIVE ACCOUNT.
It is just there to test the theory, but it is unstable.

As I get time I will improve it, but I am still months away from a fully stable version so be patient or support me to speed up development.

But never stop posting ideas.
I need to get this theory bomb proof by putting it through the grinder to see if it can survive anything.

wondering if we are better served with instead of triangular arbitrage...correlation arbitrage>>>>> https://www.forexfactory.com/showthread.php?t=677100

Correlation in non-stationary timeseries is based on a faulty concept. Co-integration is a better approach for Statistical Arbitrage, but all the inherent challenges of choosing the right sample size, dealing with stochastic trends, etc remain and render it as useless.

Check out the relevant posts of Codebreaker : https://www.forexfactory.com/showthr...02#post1927002

codebreaker is very smart...but i respectfully disagree that you need causality in order to profit from inefficiency. you do not IMO BTW the only casual predictive model in nature is synchronization.. for example...when you put 5 females in a house together you can predict with 100% certainty that their menstrual cycles will align lol. they can predict certain things about locusts and cows and such kinds of groups it's very interesting... the market is the same... it is a herd ... if you want causality that is probably your best bet... not seen it applied to markets yet however.... would be fun to predict the herd! lol

How do you define inefficiency, when you don't know efficiency other than statistics occasionally lining up. I can create a non-forex "Synthetic" for USDCAD (btw given that any relative such as pair pricing becomes a spread why would adding more or comparing it with other spreads mean anything?) that lines up beautifully / correlates and then inverts. Does that mean it is now out-of-sync / inefficient and will mean-revert ? And at what weights measured from where ? I don't believe in the value-price concept being superior. Of course, exposure needs to be hedged in other markets but it is not consistently predictable. There is no objective inefficiency, there is only order flow, liquidity and spread.

Let's not deviate from the topic of this thread being deterministic arbitrage.

yes. arbitrage is the thread's purpose... arbitrage of correlated pairs may be doable...https://www.forexfactory.com/showthr...99#post8376599 despite your objections, i've not heard any reason that makes me say "oh that's why it isn't doable"......i have OTOH seen many traders post, based on actual experience, many reasons why triangular arbitrage won't work in a live market... i simply offered an alternative approach, that may or may not be better...

2 cents

Good news everybody...
I just invented an ice cream toaster oven.

On other news.
I updated the EA with a mas amount of improvements.
DO NOT USE ON LIVE ACCOUNT PLEASE.

Keep in mind I tested this on my MyFxChoice MT5 account using 1:200 leverage.
I also tested this on a FxPro account, but for some reason there is no commission and extremely high spreads on this fxpro account, which tells me I made a mistake in picking the wrong account there, but it still made a lot of profit so it is a good bad situation test.
I also used $1000 starting balances for the tests.
I do not suggest smaller.

I did not have time to test on other brokers or see why my fxpro account is problematic.
I also did not have time to convert to MT4 this time sorry, will do so much later as I get time.

Some of the things I put into the EA:
1) Base triangle strategy of buying at random, holding and selling at profit.

2) Grid strategy to increase the position size 3 times as loss increase over specific time periods with special calculations to check for spikes and zero point movement.

3) Stop loss at 25% drop during new stating position.

4) Use opening start position loss (commission+spread+whatever costs) as profit take point and grid level loss points.

5) Calculate average spread over a day and use spreads below as potential position opening points.
Start at beginning with no history with current spread as average.

6) Calculate equity low point since position start open over a 1 hour rolling period to calculate new grid entry points to be below it, else do not enter.
Done to detect spikes and market stability.

7) Calculate 4 grid full margin levels gap before opening a new position.
Will only use 3, 4th is to ensure gap for stopLoss and to prevent margin call on 100% margin accounts.
Also decreases drawdown to 7% and increases least free margin to 38% as seen over 2 years back test.

8) Increased processing speed by removing some testing calculations and decreasing amount of logs from "you are nuts" to "are you sure you aren't nuts?".

9) Take into account swap and ignore swap during profit take to stop swap bleed and take the knock during ultimate profit, which is minor.

10) A lot more :
I went through 10 versions since last week so a lot more was done, tested, removed, added.
Some small some large.
My coffee ran out so my brain stopped keeping track of all the changes ....
But WOW it is working.

6% growth forward test demo last week (5 days).
80% growth since last month 5/23 back test.
Then server crashed need new hard drive.
Will see when I can fix it later.

Sorry that the updates is taking so long.
Sleep is getting in the way, and work, and running out of funds.
Still not stopping

Good luck all, BUT DO NOT USE ON LIVE ACCOUNT.
Still too unstable.
My server crashed during forward test and it can not pick up where it left off YET.
I still need to build those things in as I get time, but it will be devastating to a LIVE account.

Still to do:
Recover from crashes, internet problems and anything I can think of.
1) Allow EA pause before opening new positions to allow for funds extract out of account, without affecting EA.
2) Maybe not pause, but something to indicate the amount of fund it must keep aside and to notify you when the funds are available for extraction.
3) Get history from broker history on new start and/or store history data in file that can survive crashes and other problems for a smooth restart and continuation of running period calculations.

Do not tell my wife.
If she know what I have been doing during my free time she will be extremely mad
Im afraid lol