Predictive Models for Crypto using Machine Learning

Piet_Pompies said:
Sure if it is 1.2% 100% of the time. 1 week testing is too short. It can all be wiped out with a -30% day.
Click to expand...

Just have fun. Get a demo account that does realtime fake trades and see what it does. In my experience using historical testing doesn't really tell you much.
 
I once dabbled in this. About R220 million in crypto traded (at the time, around 2018) and I managed to over 10x my initial investment (got the binance txs to prove it). That said, it was a specific strategy capitalizing on a specific, time limited market condition.

Would not recommend this to anyone to be honest. There are firms dedicated to capitalizing on any gap in the markets with budgets and intelligence the average man could only dream of.
 
B-1 said:
Just have fun. Get a demo account that does realtime fake trades and see what it does. In my experience using historical testing doesn't really tell you much.
Click to expand...
Absolutely. Very easy to over fit the test data. Also your model most likely doesn't have nearly enough variables to model, at even a rough approximation, the "randomness" in the markets
 
Harmonic said:
Absolutely. Very easy to over fit the test data. Also your model most likely doesn't have nearly enough variables to model, at even a rough approximation, the "randomness" in the markets
Click to expand...
Spot on gold/fx all decades all markets etc

Crypto is a whole different story what you can maybe do is do a trend spotter and manually trade it. Elon farts and doge moons crypto is just way to volatile for this imo.
 
Harmonic said:
Absolutely. Very easy to over fit the test data. Also your model most likely doesn't have nearly enough variables to model, at even a rough approximation, the "randomness" in the markets
Click to expand...
Yup, most common mistake - that was why I suggested he fixes his back testing methodology first. :)
 
Piet_Pompies said:
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...
Click to expand...
What do you want to do with the FinTech service? Sell it to them? Have them capitalize it?
 
ML and doing back testing yielded the following system that I am testing with an auto trading bot:

1) I only buy and sell DYDX on Binance.
2) Use 1 Minute data.
3) Set RSI to 180.
4) Setup 2 EMAs. EMA 20 and EMA 40.
5) Use Donchian Channels for 60min MIN and MAX.

Setup for entries/buys:
A) Only ever do purchases on Mondays 4am-10am or Fridays after 12:00-16:00.
B) Purchase when the following conditions are met:
I) RSI pulled back to below 47.
II) After the pull-back the EMA 20 crossed above EMA 40.
III) RSI is rising (I.e. The current RSI is greater than the average of the past 10 minutes RSI).

Setup for exists:
1) Hard Stop is 35% loss.
2) Profit Stops:
I) The MAX RSI of the past 60 mins is above 53.
II) The current RSI is >0.8 less than the MAX RSI (in I) above).
III) You are in profit.

The 35% hard stop makes this a high risk-high reward system. Definitely not for people that cannot handle a 34.9% pullback :p.

Happy with it so far.
 
Last edited:
Piet_Pompies said:
ML and doing back testing yielded the following system that I am testing with an auto trading bot:

1) I only buy and sell DYDX on Binance.
2) Use 1 Minute data.
3) Set RSI to 180.
4) Setup 2 EMAs. EMA 20 and EMA 40.
5) Use Donchian Channels for 60min MIN and MAX.

Setup for entries/buys:
A) Only ever do purchases on Mondays 4am-10am or Fridays after 12:00-16:00.
B) Purchase when the following conditions are met:
I) RSI pulled back to below 47.
II) After the pull-back the EMA 20 crossed above EMA 40.
III) RSI is rising (I.e. The current RSI is greater than the average of the past 10 minutes RSI).

Setup for exists:
1) Hard Stop is 35% loss.
2) Profit Stops:
I) The MAX RSI of the past 60 mins is above 53.
II) The current RSI is >0.8 less than the MAX RSI (in I) above).
III) You are in profit.

The 35% hard stop makes this a high risk-high reward system. Definitely not for people that cannot handle a 34.9% pullback :p.

Happy with it so far.
Click to expand...
Yea these kind of setups can work imo if you are small trader in the big scheme of things

The stop loss does help to make sure you don't lose it all

Since most exchanges aren't properly regulated , (tin foil hat theory) i am of the opinion they (exchanges) probably run the data and know where all the stop losses are, so know what they would have to do to trigger a cascade effect to make your coins their coins

Charlie Lee made a killing by wash trading a nothing coin into a something coin and then telling his loyal followers to hodl while he is dumping on them

So yea if getting into trading not being a loyal follower is probably paramount

Now yes i would imagine they won't play this game, if the taking won't be big enough

Naturally easy if you have tons of coins to play with (which exchanges have)

So to a certain degree i feel the system is rigged once the kitty in orders that can be cascaded becomes too much the market can be manipulated and probably will
 
Last edited:
leon.davibe said:
Yea these kind of setups can work imo if you are small trader in the big scheme of things

The stop loss does help to make sure you don't lose it all

Since most exchanges aren't properly regulated , (tin foil hat theory) i am of the opinion they (exchanges) probably run the data and know where all the stop losses are, so know what they would have to do to trigger a cascade effect to make your coins their coins

Charlie Lee made a killing by wash trading a nothing coin into a something coin and then telling his loyal followers to hodl while he is dumping on them

So yea if getting into trading not being a loyal follower is probably paramount

Now yes i would imagine they won't play this game, if the taking won't be big enough

Naturally easy if you have tons of coins to play with (which exchanges have)

So to a certain degree i feel the system is rigged once the kitty in orders that can be cascaded becomes too much the market can be manipulated
Click to expand...
Yeah I agree, the exchanges know how to make money. My opinion is that the 10% drops and gains in alt-coins are just exchanges making money from people leveraging 10x or more. Checking open interest is also a useful indicator and can help with identifying which way these big changes should go.
 
Piet_Pompies said:
Yeah I agree, the exchanges know how to make money. My opinion is that the 10% drops and gains in alt-coins are just exchanges making money from people leveraging 10x or more. Checking open interest is also a useful indicator and can help with identifying which way these big changes should go.
Click to expand...
It is all a numbers game , you just have to win more than you lose

And compound effects of the wins will pay of

But yea that is just like saying buy low sell high ,how hard can this be :)

This isn't for everybody you need to have the right temperament
, Have rules and a tested system that works which sounds like you are aiming for

The problem though depends on what you are trading

ie 35% sounds like you won't get washed out with cascades/manipulation, but if trading higher leverage the 35% becomes a small market movement

And then the other is

Back testing does it take order books into account? Or just the spot value

Slippage ? Can affect stuff a lot
Coin dependant naturally some more resilient than others
 
I wrote a little bot (based on back testing and machine learngin identifying the most predictive parameters) and it is doing the trades for me and I am happy with it. Started with $100 and now at $190 since Oct 2023.
Here is what it is based on for trading ONLY DYDX on Binance.
1) 1 Minute data
2) EMA_1, n=27
3) EMA_2, n=43
4) RSI, n=180
5) Donchain channel of RSI (nr 4 above), n=60.

Entry conditions are as follows:
1) Low of the Donchian channel for RSI (number 5) must be below 46.
2) For the previous minute, EMA1 must be less than EMA2
3) For the current minute, the EMA1 must be greater than EMA2.
4) For the current minute the RSI must be greater than RSI of the previous minute.
5) Entry can only be during the following hours (0 Hour = 00:00 Sunday, 168 hour = 23:00 Saturday):
<19, >26 and <76, >125 and <146, > 162.

Exit conditions:
1) When in Profit (no matter how small or big) (Profit is calculated with assumption of 0.3% trading fee included).
2) High of the Donchian channel for RSI > 53.
3) RSI is 0.4 < than High of the Donchian channel for RSI.

Stop loss is when loss is greater than 35%.

Haven't had one stop loss reached so far out of 22 trades . Although the current open position is down +- 19% (bought when I had $225). Back testing of 2-year data (with major up and down swings in market) shows a PPV of 97% so I am expecting a stop loss to be hit.

Some other rules.
Trade your whole account that you started with (Never all your money, only the starting money you can afford to lose).
 
with backtesting we assume that in every tick our order won the competition and we got filled at the desired price.
when an opportunity that we studied presents itself in the market, our algo triggers and posts an order. but in live trading, most of these opportunities will be snatched by faster algos, or the market maker, so our order either ends up unfilled, or we pick up stale prices (adverse selection).
in other words, we most likely overfit our model to the past price movements, rather than predicting it.

it makes sense too:
a) everyone in the world is seeing the same data, probably uses similar indicators and price triggers. there will be someone faster, more skilled and with more information than us in the market.
b) rsi, macd, ema, et al are lagging indicators. probably our competition uses a forward looking indicator

forward looking indicator?
AhgHtnP.jpg

insider trading, market manipulation and good old pump and dumps are very very common in this industry.
crypto is the game of information asymmetry.
we can beat the market if we replace our look-back indicators with some sort of “information trigger”.

in other words we use algo trading just for execution, we construct the trade idea elsewhere.
8fiw4d.jpg
 
CryptosEra said:
View attachment 1660442
Click to expand...

to clarify I will ape into the TWAP of an insider trader any time of the day
but I don’t agree with this view that whales move the price, or market is manipulated.
its absurdly childish. How about we stop projecting our under performance to criminal whales and just accept other market participants are better than us?!

no one is forced to trade. you deposited, and you lost, don’t hate the game.
take responsibility of your wallet and your loss.
yes, I know we have HFTs and 8figure hedge funds trading against the $500 networth plebs
that’s criminal!
but where else the plebian can buy $100 doge and retire?
don’t hate the game; change the difficulty in the setting!
 
You must log in or register to reply here.
Back
Top
Sign up to the MyBroadband newsletter
X