Anyone else here very seriously want to make a living out of algo trading crypto currencies? Passive income seems like the only true way to /make it/.
I played around with some trading bot ideas like inter-exchange arbitrage (too high risk for low reward imo) and some simple mean reversion trades but in real life trading it's completely different because your buys and sells impact the market.
So far I haven't been able to break above the magical break-even amount, you should see the work I did in pic related to calculate just how much wins you need to make (and what % the gains and losses should be) to at least break even (pic related excludes trading fees but you just simply include an extra term if you want)
Anyone make money consistently here? Even if you don't want to share anything valuable, I'd be up for a discussion.
Also, anyone here believe deep learning models could be used to predict if a trade is good to enter or not (not predicting price but rather a classification model that gives either a buy, short or do nothing signal)?
I've come to the realization that algorithm trading is not that hard if you have some bank and a decent bot. What's really hard is the amount fo data you have to manage in order to be profitable, which gets really large real fast. Assuming that you are reading trades, orderbooks and price while iterating through a bunch of coins requires substantial bandwidth and a lot of computer power, because the whole point of bots is to crunch all those numbers as fast as possible.
Jaxon Hughes
Not that much for a computer though. You underestimate how capable PCs are nowadays. If you set up a cheap server near the exchange it's more than enough to meet the needs of most bots
Xavier Campbell
Where do I learn this?
Aiden Anderson
deep learning is the best if you want pure math/patterns.
but the reality of markets is that they are inherently irrational and based on human psychology of where each person in the market predicts what the FUTURE fundamentals of an asset will be and how high they think it will go - each person's influence on the market is also weighted by how much capital they hold so that pretty much kills sentiment analysis. unless your doing HFT to exploit spreads or exchange arbitrage, any trading bot will not work unless also monitored by you.
since true fundamental analysis is impossible as well as creating an effective bot (unless you use HFT to exploit spreads or arbitrage) its near impossible to automate or predict with 100% consistency.
Your best bet is to see current sentiment on how supported TA is, and trade off that or do pseudo-fundamental analysis and take a best guess.
Gavin Evans
t. i've tried this before but then read a shitload of economics books and realised how impossible of a task it is
but what do i know.. o_o
Luke Murphy
You should read better books
Julian Thompson
Have someone tried to play with technical indicators/convergences? Do these even work with crypto?
Lucas Butler
Pro algo traders use lots of “other” data streams (sometimes proprietary sometimes purchased) to get potential buy signals apart from trade/OB dataset.
Also ML is pretty useless in trading for reasons that will take more words to explain. Signal Processing is considered better.
Again pro traders generate many buy signals from variety of data sources and then decide manually or semi-automatically by a rule based or weak-ensemble learner based system.
Lincoln Price
I am still waiting for usable exchange testnet where I could try to develop some simple bot
Thomas Williams
One application of ML would be to optimise various parameters such as which tick interval to consider, if you are using some MA based indicators what span MA should consider etc.
Brandon Torres
Good luck dealing with server latency and cpu overloads. I think you believe trading a single coin makes sense, but you should trade all of them to make profits, and every API in existence has limited quotas unless you are willing to pay for better access, and it's not cheap. Setting a server near the exchange? lmao sure, do you have a least 500k to start?
Bentley Gutierrez
so a coworker was boasting about having written primitive bot that works, he's super secretive about it, but apparently it's just looking for some human-found pattern in 2 dozens of coins, nothing spectacular, haven't heard from him ever since eth went below $200...
did you really do it on paper? why not use Mathematica or python or similar?
abut this "buy server near exchange" thing, I could understand it for real stock market but does it really matter in crypto? aren't current api servers of popular exchanges written in like ruby or javascript? I keep hearing they go down the moment something interesting happens to the market
Ian Moore
>algo trading >written papers you are joking right?
Ethan Rodriguez
just put the VWAP and its std bands on a tick chart, throw on some short term fib ma's and watch it for a month, how fucking hard do you want to make this. oh you don't want to make the scary scalps on your own, you want your tame kindly bot to do it for you, tough fucking shit it isn't going to work, use your eyes/brain and hit it.
100% consistency is not a problem and as long you got yourself a super-martingale as opposed sub-martingale.
Dont read economics books, at least no the obvious ones. Look up algo research funds and look at their public and strategies. Get your books from their footnotes.
Brandon Nelson
I arrived at the analytical solution on paper and included some examples for demonstrative purposes, while keeping the work still on paper instead of transitioning to a monitor screencap. Not sure what the big issue is?
Angel Perez
it's not that there is a problem with what you did so far, but if go further in, models will become bigger and you would need to work with data then doing it on paper will become just silly, building models in software also enables new ways to explore them, not to mention huge boost in iteration speed
Anthony Powell
I've tested things algorithmically too, I just thought I'd share some pen and paper work with Jow Forums. Anyone here can stick that formula into a basic Python script and use it for setting targets for comparison. I thought a mathematical proof would be a nice addition instead of just a basic graph or some shit
Liam Morales
Sounds prohibitively expensive to set up. No retail trader can compete against that
David Perez
that's great to hear (actually you can do the proof in python too, there is sagemath, that's like mathematica, but in python)
anyway going back to the topic, have you perhaps downloaded any historical data? I know there are several providers out there - and even some free datasets - so I was wondering which one to choose
Jonathan Ward
it is extremely difficult to algo trade crypto due to the relatively small amount of money needed to move and influence the market, combined with the heavy wash trading on every exchange (orders can't be trusted, even executed orders may not be real)
combined with the fact that top wall street traders fail to beat the 7% growth of the market, you're probably better off doing some active real estate investing like selling storage space
Charles Robinson
Links to tutorials user? I understand the math and i can code python already
Owen Wood
I’m learning python, wanted to build with python from scratch, ended up using talib and catalyst, both are great to just program what you need. I’m looking for a scalping bot and trying to write a code with Ema‘s, %-stop losses and profit takings. Closest I got from backtestig sofar is 1.2% profit, which is probably not good enough for really start trading. For me the pinnacle is neural networks and machine learning but I have a loooooooooong way to go.
Jayden Campbell
Quantopian.
Kayden Long
PyAlgoTrade is a “near-pro” kind of framework.
Nicholas Brooks
I've been collecting my own every minute for the past few months with a $5/m server. Used cryptocompare's API
Jacob Carter
With this R script you can create backtest data (from Poloneix) for a given crypto pair (ETH_BTC here)
