Simple High Profit low Drawdown Triangular Arbitrage

I will always update this POST #1 with the latest information.
So if you do not want to read all the posts then just read this one.

I will also keep the EAs up to date here.

The rest of the posts are interesting and I love flame wars


*******************************
Very basic summary
*******************************
1) Buy 1 unit GBPJPY
2) Sell 1 unit GBPUSD
3) Sell ?? units using the USD in point (2) in USDJPY to balance everything out

Buy low and sell high.
Rinse and repeat.

Can this work or what is the pitfalls with this idea ?


*******************************
Detail
*******************************
OK I am still new in this, but must say triangular arbitrage looks interesting.
I made an EA to trade GBPUSB, GBPJPY and USDJPY

Basically
1) Buy 1 unit GBPJPY
2) Sell 1 unit GBPUSD
3) Sell ?? units using the USD in point (2) in USDJPY to balance everything out

This basically ZEROs your position minus spread, commission and swap.
Just hold on to this until in profit.
It will oscillate around the ZERO point.

Once I make profit I swap the above 123 points to Sell, Buy, Buy.

My entry can be random and then go from there, but since I started I changed it so that I calculate what the low and high is and use those calculations as entry points.

Now sometimes you do accidentally get in at just the wrong time.
Basically buying at the highest point and it never gets back to that point.

In 2 years time I found it about 9 times.
You can see these in the load spikes at the bottom of the chart.

This problem is solved by a minor grid.
Basically once you unluckily buy at the highest point then you will be stuck, but no worries, because it can only fall so far before coming back, but never past that point again so just wait a bit and let it fall and once it fell enough then Buy again the same amount and leave it for a guaranteed profit a bit more than half way back so just wait a bit.

Now as I said My speciality is maths and programming, but I am new to the forex market.

The demo EA I wrote I ran from 2015-06-23 till 2017-06-16.
I started with $1000 and ended with $2,725,839.38
Maximum drawdown during this time was 25% of the account.

I did have a safety catch in the system to prevent losses from going beyond 25% which was hit 2 times during the 2 year run, which does not affect the EA seeing that it will adapt itself to the next balance as needed.
The cause for the 25% drawdown spikes was weekend gaps on the Monday.
I tried preventing Friday trades which caused the maximum drawdown to drop to 15%, but the ending profit also dropped to about $1,800,000.00

Other things to note was that I was maxing out my Metatrader 5 back test lot size at 99 lots after the 1 year mark at about 2016-06-28 and from there it stopped growing exponentially and started a slower sequential growth.
In the live environment you can split your account into multiple accounts long before it reaches the limit for even more profit.

The 25% drawdown hits will probably also increase if you can keep the exponential growth growing from 2 hits to about 4 hits, maybe 5 at most, but the EA is designed for it and can handle about 20 times 25% hist before starting to show a loss, because the growth rate is about 100% every 2 months and at worst it hit the 25% drawdown 2 times in one year..

I do not have funds to test this live, but demo testing seem quite good.

Any advice will be great.

I even found this to be profitable using a starting capitol of only $100, but at this price all 3 pairs will start at the same 0.01 lots, which is a bit off in balance seeing that the USDJPY can not be the same as the other, but once it breaks the $1000 profit mark it works good and the $10,000 mark is just rocket fuel for it.

Ignore the upward spikes in the charts.
The EA is told to start selling the highest profitable once first.
What matters to me is the equity and how far it falls after opening a 3 pair group.

The 2 pair 25% drawdown happened in the beginning of the chart and therefore is almost invisible when compared to the profit of the rest of the chart.


*******************************************************************************
RULE UPDATE : 2017-09-23
*******************************************************************************
Ok this is a MAJOR update.

Rules changed a bit.
Still triangular on the 3 symbols.

But.
I now calculate what the average profit would be if I did a BUY, SELL, SELL and also SELL,BUY,BUY.
This gives me 2 lines.
Call the higher line the ASK line and the lower line the BUY line.

Between the 2 lines is the average line calculated over some time, which if mostly a straight line as the buy and sell lines basically cancels each other out there.

Now from time to time you will notice the ASK line cross the average line and this is the time you must open a Buy(Sell, Buy, Buy) to profit from the inevitable up movement.
Close the open buy once the SELL line cross the average line.
Do the opposite for opening a sell(Buy,Sell,Sell).

One more thing is that I do not just open when it crosses the middle average line, but I only open once it crosses the average line for some distance before opening.

Ok there is a lot more checks and stuff I put in the EA, but that is the basics.

I tested this on MyFxchoice with a leverage of 1:200 so try and keep to those settings.
I might work on other settings and I did try and build in intelligence to adapt to other brokers, but I had limited time and it might not work properly.
Backtest to make sure it works properly before a forward test.

DO NOT USE ON LIVE ACCOUNT.

Other things in the EA are :
1) Do not trade weekend gaps for the Monday slippage can be big.
2) Do not trade between midnight and 2 am for the same reason as (1).
3) Move all display and log coding after the system calculations to increase trade calculation performance.
4) Use only 50% of ac
count for trading, which only mean that 50% of the account must be converted to used margin.
Draw-down is about 5% then, but I can not increase it for fear of getting a margin call.
20% might be better, but I will test it one day.

5) Check for failed trades and kill them to start over.
6) Check for weekend and inactive markets to prevent system failure.
7) Check commission using single 0.01 lot trade.
8) Massive amounts of logs and graph code to display what is happening.
This uses a lot of memory so I must probably move it into another EA for display reasons only seeing that it is not needed to do trading with, but looks nice and you can see why the EA does what it does.
Then there can be one EA to do trading with and another display EA which you can load somewhere else to show the workings, without the display EA affecting the trade processing.

Problems:
1) Slippage and trade delays.
I am getting close to 500 milliseconds delay on opening 3 trades for the 3 symbols at a time.
This can cause some major slippage, especially on the 3rd trade, but luckily I still make profit on most trades seeing that the EA does cater for some slippage.
I'm still trying to find ways to decrease the time delay, but it seem to be a limitation on how metatrader works.

1.1) I might be able to improve the trade speed using 4 metatrader terminals and I have already created programs to allow EAs communicate with the outside world within less than 1 millisecond, but there is still a delay of about 300 milliseconds for a single trade to start.

On the bright side I have now developed a driver that can connect an EA to anything else from other EAs to other terminals or even outside programs like java and even to other PCs in other countries or any other device for that matter or even making a metatrader EA use things like the FIX api.
The possibilities are big for this driver.

1.2) Delayed trade starts seem to be caused by the terminal logging out if there is no trade action for some time and when a new trade is needed then the terminal for have to take some time to log back in before the trade can be opened or closed.
I could not find a way to prevent logout, EXCEPT by sending dummy trades or impossible pending trades or trade modifications every 20 to 30 seconds.
Problem with this might be that broker will not appreciate receiving so many trade signals.

1.3) FIX API.
I have been playing around with FIX api and it might be a solution to this trade delay, but I need to test it first and .... well... time is not on my side



*******************************************************************************
RUNNING THE EA
*******************************************************************************
Do not run on live account.

Just attach the EA to a chart ....... That is it

Some more details.
The EA is made to run with symbols GBPUSD, GBPJPY, USDJPY
Exactly and it is case sensitive.
If those symbols is not in your market watch or trade able then it will not work, UNLESS...

There are a couple of parameters in the EA you can set.
Mostly the trade symbols.
Keep in mind the EA is only made for GBPUSD, GBPJPY, USDJPY
BUT if your broker contains symbols like GBPUSD#e ex.. then you can use the parameters to change the EA to use your broker's symbols.
Keep in mind it is case sensitive and therefore if your broker lists GBPUSD#e and you put in GBPUSD#E then the EA will not work.

To view the chart drawn by the EA I suggest making the chart window background BLACK and zoom the chart out by dragging down on the right hand vertical price bar till u can see all the lines the EA draws on the chart.
I'll even suggest making everything black as the chart price gives no useful information.

thanks for sharing your system , but haw u calculate USDJPY lot ?

looks impressive.
one thing i can think of on real account are issues with execution times of your orders.
In demo they execute instantly, on real accounts you may experience drawdowns due to the lag.
I guess maybe forward test this strategy also on live demo account first?

Hi Jacques, I can test it on a live account if you share the EA.

@Waddah
I might calculate it wrong, but I do the following :
GBPJPY = lotSize
GBPUSD = lotSize
USDJPY = lotSize * GBPUSD

@HurryK
I know yes and would love to test it.
I am busy testing it on a demo account, but I have got a 400ms lag
Still making a profit (just a bit less), but lots of requotes and slippage.
I do have plans to cater for slippages and requotes, which should be simple programming, but my problem is time and funds to do the development an maybe get a VPS to get a better forex connection which will cost me about $40 a month ..... That is a lot of money in South African rands and beyond my financial abilities.
Took me a few months to get it where it is now and to make it bomb proof probably another 2 months.

@shiva AND @HurryK
I have attached the EA to this message, BUT PLEASE DO NOT USE ON LIVE.
Rather on a faster than mine demo account.
Reasoning is the stuff I have not build into it yet.
As stated by @HurryK there are things on a live account that is not in back tests.
To list a few :
1) Requotes.
EA should will still make a profit, but with further development it will make a lot closer to back test profit if it knows to look for it and cater for it, which I know how to program.... Will just take some time.... and at the moment I can not afford taking more time of work.
2) Slippage.
EA should will still make a profit, but with further development it will make a lot closer to back test profit if it knows to look for it and cater for it, which I know how to program.... Will just take some time.... and at the moment I can not afford taking more time of work.
3) CRASH
At the moment the EA lives in a fantasy perfect word of perfect internet and perfect metatrader programs.
Isn't that nice.
3.1) At the moment the EA does not check for existing positions when it starts up and therefore if there are it will mess with its head and calculations.
3.2) If Metatrader restarts then it will also not look for trades left open since last time it ran and therefore miscalculate the future.
3.3) EA Freeze.
I have not seen this, but it is a possibility for which I would like to build in a remote connection to my cellphone or email or something to notify me of any problems so I can jump in and save it from gremlins.
3.4) Check for existing positions and process them or warn the user and stop if not part of the system.
3.5) Check for multiple EAs which can interfere with it.
3.6) A Metatrader 4 version of the EA.(Already done, but not tested.)
3.7) I still need a way to extract profits, without affecting the EA.
This can be done using a user settings that can be activated mid run to tell the EA to leave alone a user selected amount and then notify user once this amount is lose so she can extract it without affecting the EA.
You can extract any amount at the moment in-between trades, basically if there is no open position, which is difficult seeing that the EA can immediately open another trade.
3.8) Quite a few other possible problems also need to be looked into....

All of these CRASH points are easily solved in the program.... Will just take some time.... and at the moment I can not afford taking more time of work.

To run the EA.
1) Make sure you have got no open positions on you WHOLE account.
2) Attach it to a GBPUSB or GBPJPY or USDJPY chart.
ONLY ONE CHART and not multiple charts.
Any time frame.
It just need to be active and receive tick signals from anywhere.
You can actually attach it to any chart, but the tick signal timing will differ so a one of the top 3 pairs will probably work best.
3) I suggest a starting account with $10,000, but it should work with less.
As stated it even worked on a $100 account, but it is risky with whiskey seeing that the USDJPY pair is not balanced correctly then.
4) Thats it.
No settings or anything is needed.
It does check spreads and will not open trades if the spread is too high.
I suggest a ENC/commission broker with low spreads.
It will take about 6 months for it to average out and show real profits.
Or sooner if you are lucky.
It does some trades once every 3 days on average, but can do a few per minute if the maket moves a lot.
Keep in mind that this version of the EA is set to EXTREMELY aggressive.
I can set it to safer settings which will still make about 500% profit a year, but hey I'M A DAREDEVIL Or I was dropped on my head to much.

if u r not adding or closing trade sizes during positions floating, then u basicly trading the price changes n this can go positive or negative.

the initial lot size calculation seem correct so far but 4 the actual prices. when u open arbitrage trio n let it run 4 days or weeks then the pricing will change n so lot sizes should be fixed by adding or closing partially. if not then u basicly trading the distortion of price change n this can go as said positive or negative

n when trades floating swaps will kill, if u wanna test on real find a swap free account, just sayin

@mr.brown
Thanks for the input
Not sure I understand correctly, but here goes.

I actually bargain on the price changes.
If the price never changed then I would make no profit or loss for that matter, except for the spread, swap and commission.

I call the point at which I open the 3 pair the ZERO point.
The price normally oscillate up and down (profit and loss) around this price.
BUT once in a while(once a year or something) the ZERO point moves.
The ZERO point can not move to much or else the whole financial world system will fall apart, but still with leverage it can affect you.
But for this instance I do have another catch.
If I open a position and I see that this position stays open for more than 3 days then I will check for the lowest point over the last floating 3 days time and if this low point is hit again I will open another position in the same direction to counter the ZERO point move.
This way the price only have to move half way back to get into profit and close the whole position.

You will see this happening in the cart I added above on the bottom DEPOSIT LOAD.
There are spikes where my position doubled.
I never go more than 2 full positions in and if it does not recover by then then I have one final stop loss of 25%, which was hit 2 times in 2 years, which is more than acceptable with this growth rate.

My tests include SWAP so it already caters for swap.
Most of my trades happen to fast in one day to incur any swap.
I will add an image above showing swap loss on some trades and the profit made even with swap.

Does someone know how to get the full trade history in text format so that I can attach it to this forum in stead of screen prints.

if u use initial price in your lot size calculation which u r, "26.03" "26.03" n "37.01" then when price changes u should change the lot size too by partially closing or adding. clear ?

PS : is weekly double triple swaps added into account ?

btw, Am not offensive, have tried triangular arbitrage years ago
I can give u n advise, use dll n use 3 platforms. connect them by using shared dll while when same instance opens 3 trades at the same time it puts them in a queue, if u use 3 patforms connected to same broker server then 3 will be opened at the same time [milliseconds]

hire a vps from the mean broker data center so u can get < 2ms latency
good luck

@mr.brown
Yes I suspect double and triple is included, but swap is a minor amount to the profit.

As for the size calculation.
I leave the calculation the same for the run of the position, but I do recalculate once a position is closed and a new one is opened up.

The size calculation change very little over a long period of time and therefore it does not affect the profit much as seen over the 2 year run backtest and most positions are closed the same day or hour.

I try to keep things as simple as possible so I keep away from position sizing as that might eat a little into the profit.

n when u put all together code, vps, dll, deposit, time, when its time to withdrawal, when broker check ur accıount detail n finds u simply trading gbpusd gbpjpy usdjpy or any other triangular arbitrage, it will probably freeze ur account n u will forget the deposit

during bt u have no latency n slippage, when u run it on demo it can make profit too, but when on real u will see the ugly truth
good luck anyway

@mr.brown
I really appreciate your input.

I was thinking the same thing about 3 programs running the same time, but I will do that at a later stage seeing that there is a lot of other potential problem that need programming first.

As for the VPS.
Yes that is a dream of mine, but $40 a month is a bit beyond my financial means
But if it ever goes live then a VPS is a must for the speed.

At the moment my poor test server here at home has got a 400ms delay and all the problems that come with it, but still making profit
Basically I adjusted the profit take to take into account the worst possible slippage and still some extra profit.
So maybe my poor test server is a good stress test.

Here is the cheapest VPS. I am using more then 5 years. Take a 1GB RAM, Metered Bandwidth Windows VPS for $10/month
https://goo.gl/wTLezS

vps should be placed at the same data center where mean broker servers located. random vps location wont work

Thanks Jacques, as you wisely sugsted I will test it on demo and will also place it on myfxbook so that all of us can monitor how this goes

@skg3007
Thanks for the advice, but @mr.brown is right in that you need a fast connection to the broker you use.
I used to use a vps connected directly to myfxchoice and it was fantastic, but it bled me dry financially.($40)
One day maybe I can do it again.

@shiva
This will sound stupid, but how do I also see the results of this myfxbook ?
Is it edible ?

I'm a master builder, programmer, hardware and software automation in high security mining, tagging, gas, vehicle automation, camera body and facial recognition, architectural automatic build fenestration scanning and calculation automation, lamp room controls, and bla bla bla bla , BUT I know very little about forex.
I do love mathematics and after a few months of dabbling in the forex I started seeing some patterns like this and applied what I know about A.I and automation and it seem to work.
But how fxbook works is still beyond me

Monitor this forum or mail me from time to time and I will give you updates to the EA.
Keep in mind what I said about the shortfalls of the EA.
If you want to stop and start it then it must be done inbetween trades, so be quick or take the knock on a new trade and just close it after closing the EA.
Do not start the EA on a account with already open position as it will not know about them and it will mess things up.
Even trades opened by this EA itself...

I will insert an update to stop more gracefully and do take profit better.
One day.....

I mostly tested this using a ENC broker account.
I did test a normal non commission account also and it did make profit, but less.
I tested the metatrader 4 version even less, but it worked thus far.

Hi Jacques, Thanks for sharing. one question, this system is not suitable on Friday?

Hi Jacques, Myfxbook.com is a website to publish trading results. Just like forexfactory's trade explorer.
Since your EA is MT5 and not MT4, I chose to use myfxbook instead of forexfactory explorer. Unfortunately forexfactory does not have a provision for publishing MT5 explorers yet.
I will post the link here for all to see. It updates in real time.