I’m happy to begin my official algo (machine learning strategy) journal.
For those of you who aren’t familiar with machine learning It is a statistical robot that predicts the future. HA! With an accepted amount of error of course.
There are two predictive models which I engineered. 1 is for the long side and the other is for the short side.
Here is a general summary from an operational stand point of how this works.
Summary:
Each model generates a score from 0 to 1 where 0 is the least likely of an occurrence and a score of 1 is the most likely. For example if the short side model generated a score of 0.90 and the long model generated a score of 0.01 then there is a heavy bias towards the short side.
That being said I don’t conduct any analysis as the robot does it on its own. I am merely the clerk for a supposed more intelligent entity.
The robot has spoken and I am to initiate a short.
Open Positions:
Sell signal was generated for the EUR/USD on yesterdays opening @ 1.1233 however my funding hadn’t gone through.
Now my funding went through.
Short 1000 EUR/USD @ 1.12970
Woke up at 12:30am EST to run my code but due to a time delay in the data I needed to wait until 1:30am to get the February 3rd time stamp. I didnt want to wait so I went back to bed. Then BAM the market went up against me. Today I’ll be adding a dummy row to my data that says “if the previous day is complete but there is no time stamp yet for today add a present day dummy time stamp”. My code is looking for todays date to generate a score so a dummy date will do fine as long as I have all of yesterdays data.
My score says to be neutral today but I was way to late to act on that. I’ll take a loss today. While I could look at technicals and attempt to look for a more opportune exit that would defeat the purpose of the models I built.
Trade 1 update:
entry Short 1000 EUR/USD @ 1.12970
exit Buy 1000 EUR/USD @ 1.13947
-97.7 pips
$9.77 loss
Cool journal. Are you going to share what your algo looks at to make the score?
Once you get the model down you can add a bit to automatically open and close trades. Then you can relax on the beach while the code does the work 
Does your code tell you what to trade or just that you need to look for long or short positions?
Hi Sam,
Its a machine learning classification model for both the long and short side models. Its a theoretical probability score that the market will move in a given direction. The secret sauce isnt in the model though its in the data engineering that gets fed into the models.
Hi Ig,
The model outputs are being used for both entry and exit as well as trade duration. As long as the score says to be long then I’m long. If it switches to either a neutral or a short score then I’m out of that long and waiting for the next trade or switching my position to a short side trade.
Because there are two models one for long side and one for short side there are three outcomes when those scores are blended. Those outcomes are long, short, or neutral.
1 Like
models are showing to be long the EUR/USD
open positions:
long 1000 EUR/USD @ 1.14632
So how many trades have you taken based of the algo recommendations? And are you making any pips?
Hi Sam - Ive only placed the 2 trades. I had some hardware issues with my new setup earlier this week so I haven’t placed any new trades. The extra sticks of memory I bought to add to the new build made my primary system unstable.
I need to update this journal with the outcome of the above long trade. Both trades were in the red. Still too soon for me to know. I’m also trying to understand the discipline and routine I need to keep.
Right now I’m building a USD/CAD model so I can get the trade volume up. Since I only use the daily candlestick data the model puts out 1 to 2 trades a week for the EUR/USD so adding another currency should give me a total of 2 to 4 trades a week.
Trade 2 update from February 4th:
entry long 1000 EUR/USD @ 1.14632
exit sell1000 EUR/USD @ 1.14446
-18.6 pips
$1.86 loss
Hi All,
I have been here but waiting for the next trade. I also had some hardware issues where the sticks of memory I added made my machine instable. While that was going on I missed a trade to be long the EUR/USD. That would have been a loss. While I am working on a new model I have also been working on my discipline. Here are the rules.
- No entry or exit unless the model quantitatively says to do so. In other words “no chart analysis” and no “hunting for better entry”.
1.2. May cover half the spread.
- On Sundays the model is run when the market opens and on all other days it’s run at the end of the day.
Trade 3, 2022-02-16 00:41 :
entry long 1000 EUR/USD @ 1.13543
Blue arrow and the blue horizontal line is the trade entry
Hi,
Since the Russian invasion I have taken a break from the markets. During the break I reengineered some of my code to mainly optimize my models based on revenue and accuracy where revenue takes a front seat. I should be back in the mix in the next few weeks. However this time I will also be generating scores for the USD/CAD and the EUR/USD. I’m hoping my volume will double so I can speed up this experiment.
I also need to update the last trade I entered. It was a 50 pip loss.
See you all soon.
Best,
Ryan
