Predictive Models for Crypto using Machine Learning

Changed the eval_metric, added additional parameters and changed the accuracy measurement.
AAVE:
AAVE.png

I am happy with these models, I will invest $50 in the next predicted upswing after the pull back.
 
A downturn within 24hrs is predicted by the latest models.
AAVE:
AAVE.png

Scores on the left represent PPV/Precision (top) and sensitivity (bottom) for prediction on the verification set (28 days/672hours).
 
Some of the models suggested SOL would be a good chance for a profit within 24h so I got some last night.
Let's see where the predicted stop kicks in.
TEST1.png
 
The AAVE model looks decent.
AAVE_M.png
The 1st blue line represents predicted entries.
The red line is the exit signal.
The stats on the figure are as follows:
1) Matthews correlation coefficient (zero is random, > zero is better than random, 1 is perfect)
2) PPV
3) Sensitivity


Plan is to put in $50 for every trade and tally up profit/losses over a month (the 1st post in this thread is updated after every round)
Bought $50 of AAVE.
Stop loss set at $89
TEST.png

EDIT:
The stop loss was reached. Out: $47.51664577 (5% loss)

EDIT:
SOL looked good:
SOL_M.png

Bought $50 SOL at 37.0654. Stop loss at 36.3

TEST1.png
 
Last edited:
I have now played around with the ML and also completed a script to automate trades. Initial testing shows an average of 1.2% profit per day in the last week. Some room for improvement. My question is, what if I want to commercialise this?

Does anybody have experience?
 
Piet_Pompies said:
I have now played around with the ML and also completed a script to automate trades. Initial testing shows an average of 1.2% profit per day in the last week. Some room for improvement. My question is, what if I want to commercialise this?

Does anybody have experience?
Click to expand...
You can use any algo trading platform out there (eg, quantconnect and such) that supports your instruments to trade it.

I suggest looking for the error in your back testing methodology first though.
 
Piet_Pompies said:
I have now played around with the ML and also completed a script to automate trades. Initial testing shows an average of 1.2% profit per day in the last week. Some room for improvement. My question is, what if I want to commercialise this?

Does anybody have experience?
Click to expand...
Why commercialize? At 1.2% a day cash in everything you can and let it run.
 
Snyper564 said:
Why commercialize? At 1.2% a day cash in everything you can and let it run.
Click to expand...
^^^ this

if anyone ever managed to conjure a proper automated successful trading bot or some such ... they would use it themselves to make money, they would not sell it to other people to make money

refer to MTI threads for example
 
NarrowBandFtw said:
^^^ this

if anyone ever managed to conjure a proper automated successful trading bot or some such ... they would use it themselves to make money, they would not sell it to other people to make money

refer to MTI threads for example
Click to expand...
Spot on! You need no ones money but your own.
 
Piet_Pompies said:
I have now played around with the ML and also completed a script to automate trades. Initial testing shows an average of 1.2% profit per day in the last week. Some room for improvement. My question is, what if I want to commercialise this?

Does anybody have experience?
Click to expand...

If you manage to get together 50k just yourself this what you could look at in 1 year


1676139294056.png

100k this


1676139318173.png

You need no one else if you actually have this working...

Mortgage your home take out a personal loan you will be a millionaire on less than R50k if you can get that.
 
Snyper564 said:
If you manage to get together 50k just yourself this what you could look at in 1 year


View attachment 1474063

100k this


View attachment 1474065

You need no one else if you actually have this working...

Mortgage your home take out a personal loan you will be a millionaire on less than R50k if you can get that.
Click to expand...
I know right. I'll need at least 1 month of live testing to confirm this. 1 week and an uptrend won't do it. I suspect the next 2 weeks there will be some pull back in the market so it will be a good test to see how robust the system is.

Still interested in some form of FinTech business startup.
 
Piet_Pompies said:
I know right. I'll need at least 1 month of live testing to confirm this. 1 week and an uptrend won't do it. I suspect the next 2 weeks there will be some pull back in the market so it will be a good test to see how robust the system is.

Still interested in some form of FinTech business startup.
Click to expand...
Dude if your thing works...

YOU WILL NEED NO ONE.

Also a month of testing live even with years of back testing can only do so much.

But once again if it works you need no one.
 
Snyper564 said:
Dude if your thing works...

YOU WILL NEED NO ONE.

Also a month of testing live even with years of back testing can only do so much.

But once again if it works you need no one.
Click to expand...
I hear you. It depends though. If I get 0.2% p/d instead of 1.2% in the next 2 weeks then some sort of Fintech service might still be an option. Oh well, let's see. Back to coding and live testing...
 
1.2% per day compounded is a lot of money even if you invest R5.
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X