Hi Everyone,
I was wondering if anyone has tried and tested a system built for trading cryptocurrencies, I would like to know their approach and results,
Regards
1 Like
I’ve been working with hummingbot.org user scripts.
Not backtesting in in the usual sense, yet using paper trading and testnets till one is ready for live trading.
alan
2 Likes
I am working in the moment on a Bot. I have considerable experience in traditional trading (e.g. VIX-Futures). I am new to the crypto-market. My task is to develop the trading logic, my partner is responsible for the infrastructure (sending orders, receiving data…). The starting point for this part is Hummingbot. Hummingbot was initially developed for Market-Making. Usually Market-Making is the game of the big players.There are claims that crypto is different, but I am sceptical. My part, the trading logic, is currently developed as individual Notebooks. This shall be integrated into Hummingbot in the coming months. The MVP should be ready at 2023-10-05 (the birthday of bot developers). According my results so far is the Crypto Market less efficient than the traditional market. There is room to exploit this. I think the real problem is the counterparty-risk. I trust CBOE, but I don’t trust e.g. Binance.
But we are using currently the Binance API for downloading historic data. The models are based on traditional statistics like e.g. Random-Forests, Support-Vector-Machine and not Neural Networks. I think there are too less historic data for training a Neural-Net.
If you are interested, please contact me at c.donninger@wavenet.at
1 Like
My approach e.g. for VIX-Futures trading is first to backtest. If it does not work in backtesting, it definetly does not work in real trading live. The converse is not at all true. It is almost impossible to avoid overfitting. The next step is papertrading, followed by trading with small amounts of money. There are events which can’t happen in backtesting. E.g. short price-strikes, information delays … The bot has to handle this gracefully. Usually one detects in this phase also serious bugs. If this phase works for several months one can switch to full trading.
1 Like
Thanks a lot, I’d be reaching out to you, do you by any chance have a github test base code to help me begin my exploration, my experience in the crypto space is firm, but I’m never built a trade bot before, so I was wondering where to begin from, as this is quite different from designing a credit score card or fraud detection systems,
We have no github test base case. First of all is our project currently not mature enought. We are just at the beginning.
We also do not intend to make it public. I am interested in direct personal contact. “I tell you my tricks, you tell me yours”. It should be of mutual benefit. I am originally coming from computer-chess. There are open-source projects, but the top programmers cooperate in this way.
@rychrr @Chrilly @ajjcoppola I am a novice seeking to become a valuable member of a team specializing in the development of cryptocurrency trading bots. I am enthusiastic about contributing my effort to your endeavors. May I inquire if there are any current vacancies on your team that align with my interests?
Greeting i am sabih and i am working in some statergies to build a model of binance future testnet fetcing from secret api for trading and backtesting part for tp sl hit.So i was implemented so many ML as well as DL or ANN model for predicting the backtesting results for future trading and have some dataframe of 1minut interval for time-series data and 4hour interval trading result. i was using RSI indicator for making direction and trands on the bases of 1,-1 and 0 i have got a promising result after backtest my statergies but thier was a big way to achive this now i was implementing Some ML and DL model like SVM , Linear regression , KNN for neigouber trade time or tp sl hits, DL models like LSTM, GRU ,Diret , and read about opteuna (a python package for tuning parameters ) and i finetune 15 models based on ML and DL and got promising result.
Obtained 2 to 3 models that gave over 1000% my pnl on the year span of backtesting.
Overall currently it is on updating process.