Why do you think there is even that?
-
Join the MyBroadband community
South Africa’s biggest forum. Discuss, discover, and connect with thousands of members.
-
Question of the Day: What's the most interesting small town you've visited in South Africa?
Predictive Models for Crypto using Machine Learning
- Thread starter Piet_Pompies
- Start date
Your opposition is using ML too, and using it with far more resources and Far more expertise.
? That’s a pretty non sequitur response.
Piet_Pompies
Active Member
- Joined
- Aug 5, 2022
- Messages
- 37
- Reaction score
- 21
It is a case of chucking mud and see what sticks. I am trying various indicators at various time frames and then selecting the optimal parameters that yield the smallest error and best PPV.
So far the following parameters appear to be good for buy signals for the BTC set:
atr, evwma, BBands, Keltner, ADX, RSI, week hour, macd and EMA's
For selling signals:
OBV, ADX, VWMA, macd, week hour, EMAs and BBands.
Playing aroung with the hyperparameters also helps, escpecially "eta" and "max.depth".
BTC is still predicted to be quiet now.
Solana is predicted to trend upwards. Model accuracy for buy signals is 76%. Let's see what happens.
ETC is predicted to trend downwards. Sell accuracy is >95%
Last edited:
Piet_Pompies
Active Member
- Joined
- Aug 5, 2022
- Messages
- 37
- Reaction score
- 21
Viewed? I have viewed some of these videos to learn about some trading systems and indicators.
And I try to read up on some of the value propositions of the various coins.
Participated? I am about to put in $50 if the models show some potential.
Omw...
How many days months years etc.
Have you actually watched the crypto market?
I am not asking how many years data you have I'm asking how many years etc you have watched it.
Your response indicates that you are very very new to crypto.
Last edited:
s0lar
Executive Member
- Joined
- Sep 22, 2009
- Messages
- 7,336
- Reaction score
- 5,941
They just gate keeping the local Mybb crypto cool kids club. Don’t expect much positivity to come from the grumpy old men here.
Invest your $50 and let’s track that against what you have modelled. Would be informative for many.
You serious?
We are talking abt someone that appears to have zero crypto knowledge. Crypto trading is very very different to your "traditional trading". But I mean I'm just a mean old cool kid.
Also another reason you don't wanna do this is in extreme periods of volatility exchanges "crash".
It's all good to say I have this bot and it's awesome. And first major rush on the exchanges cause another outage and there is zero you can do cause you have never seen this happen before...cause your crypto experience is limited to watching the odd trading vid..
I said don't waste your time with something like this and crypto due to its EXTREME VOLATILITY.
Rather do this with fx or gold. I'm not discounting the AI bot I'm just saying stay away from crypto
Last edited:
s0lar
Executive Member
- Joined
- Sep 22, 2009
- Messages
- 7,336
- Reaction score
- 5,941
Gotcha.
That said the speed of fx and gold is slow as mud to train and improve in comparison to crypto. Large dataset of historical data has a negligible influence to future trades. So I don’t see value in training against gold. FX possibly maybe.
Crypto more fun and as somebody said driven by solving media and the herd mentality if you can pic up of those trends early you may just be into something. I.e track the sentiment indicators.
Piet_Pompies
Active Member
- Joined
- Aug 5, 2022
- Messages
- 37
- Reaction score
- 21
I have watched it since 2018. But that means nothing.
Piet_Pompies
Active Member
- Joined
- Aug 5, 2022
- Messages
- 37
- Reaction score
- 21
I am looking for the most volatile coins. The ones with the biggest up and down swings. ALICE is an example. But none of my models yield any kind of decent accuracy for it so I will stay away from it. However, other coins with decent volatility such as MANA, Sandbox, Aragon, FTM and Solana yield decent models. These are the ones I think I will try.
I have about $100 to burn for this little pet project.
Last edited:
Piet_Pompies
Active Member
- Joined
- Aug 5, 2022
- Messages
- 37
- Reaction score
- 21
Some models to see how they peform in the next 2 days.
Top graph is buy signals. (Blue line predicts at least 9% increase peak within next 48h)
Middle graph is sell signals. (Red line predicts at least >5% decrease within next 48h) (Accuracies for verification set are top left)
Bottom graph is Buy minus sell to remove overlaps to give best buy signals.
Everything right of the black vertical line is the last 48h predicted by the model (data not part of training or veification set).
Everything to the left of the black line is the 2 weeks prior verification dataset.
Data before the graph up to 20 Jan 2022 was used to train the models.
AAVE:
ANT:
View attachment 1361081
AVAX:
View attachment 1361083
BTC:
View attachment 1361085
ETH:
AAVE:
MATIC:
SAND:
Top graph is buy signals. (Blue line predicts at least 9% increase peak within next 48h)
Middle graph is sell signals. (Red line predicts at least >5% decrease within next 48h) (Accuracies for verification set are top left)
Bottom graph is Buy minus sell to remove overlaps to give best buy signals.
Everything right of the black vertical line is the last 48h predicted by the model (data not part of training or veification set).
Everything to the left of the black line is the 2 weeks prior verification dataset.
Data before the graph up to 20 Jan 2022 was used to train the models.
AAVE:
ANT:
View attachment 1361081
AVAX:
View attachment 1361083
BTC:
View attachment 1361085
ETH:
AAVE:
MATIC:
SAND:
Attachments
Piet_Pompies
Active Member
- Joined
- Aug 5, 2022
- Messages
- 37
- Reaction score
- 21
The AAVE model at 15:30 today predicted an uptrend. It was hovering at 100.5.
Now it is 104.5.
I'll take this as a lucky guess and wait for other models to show potential.
Latest model:
Now it is 104.5.
I'll take this as a lucky guess and wait for other models to show potential.
Latest model:
Last edited:
Piet_Pompies
Active Member
- Joined
- Aug 5, 2022
- Messages
- 37
- Reaction score
- 21
An example of ML failure: SOL
AAVE:
Predicts an upturn in a >5% drop. Either an ML failure or there is a major spike coming.
AAVE:
Predicts an upturn in a >5% drop. Either an ML failure or there is a major spike coming.
Last edited:
Piet_Pompies
Active Member
- Joined
- Aug 5, 2022
- Messages
- 37
- Reaction score
- 21
We will see if it is a failure.
No it does not set off red flags. All it means is that the ML could successfully predict 95% of future downtrends in a specific verification dataset. That does not imply that those patterns will translate to future time points.
That verification dateset was 1 week. Accuracy drops as I increase the length.
If it is an ML failure, so what? I don't expect 100% success. I am hoping for > 50% success. I am trying a few more parameter changes and settings before putting some money in with this system.
Last edited:
Anything that can be predicted as high as 95% or more in finance is either a known quantity that doesn’t need to be predicted or wrong.
So, this is a key point. Anything directional with more than 50% plus epsilon is either arbitrage (and you will lose the race), or incredibly likely to be a conceptual error.
Non-directional statistics such as volatility could let you infer the probability of the price going over 106%, without any ML. Why this still isn’t going to help you trade, is a conceptual issue: You’re effectively leaning towards a market making type of strategy, but the volatility is already traded to exactly the point that the risk of further trading (which is a swing in one direction) outweighs the benefit of capturing the spread.
Piet_Pompies
Active Member
- Joined
- Aug 5, 2022
- Messages
- 37
- Reaction score
- 21
The most likely option can be that the verification set was just too small and just happens to contain simpler patterns that are picked up by ML.
Here is ETC with a longer verificaton set.
