iyedexe / trippybot

Trippy Bot :

Presentation :

• This is a start of a framework/codebase for creating strategies and backtesting them.
• The reason for this rogue framework (and not using backtrader, vectorbt or even quantconnect)
  • Working with asynchronous market data signals
  • Flexibility with market data fields and frequency
  • Simplicity and ease of mind of being batman.

HOW to :

{code} Run a strategy live: live_trader.py --configfile live_config.ini --strategy <STRAT_NAME>

Backtest a strategy: backtester.py --configfile backtest_config.ini --strategy <STRAT_NAME>

Record market data: recorder.py --configfile recorder_config.ini --symbols <SYMBOLS_LIST> --type <TICK|CANDLE> {code}

Strategies :

Triangular arbitrage:

Triangular arbitrage (also referred to as cross currency arbitrage or three-point arbitrage) is the act of exploiting an arbitrage opportunity resulting from a pricing discrepancy among three different currencies in the foreign exchange market.
.
Using a starting coin, a path is a combination of BUY and SELL transactions performed simultaniously (here done in sequence because of illiquid portfolio).
These transactions result in a higher balance of the starting coin.

• Example :
  The simple arbitage cycle based on BNB, BTC and ETH in that order :
  BNB -> BTC -> ETH -> BNB One possible path is :
  SELL BNB/BTC
  • unitary cost : 1 BNB
  • get : Y1 * (1 - Fee1%/100) BTC
    BUY ETH/BTC
  • unitairy cost : 1 BTC
  • get : 1/X2 *(1 - Fee2%/100) ETH
    SELL ETH/BNB
  • unitairy cost : 1 ETH
  • get : Y3 *(1 - Fee3%/100) BNB

==> FinalBNB = StartingBNB * (100 - Fee1%) * Y1 * (100 - Fee2%) * 1/X2 * (100 - Fee3%) * Y3
r is the profitabilty of a cycle :
**r = (1 - Fee1%/100) * (1 - Fee2%/100) * (1 - Fee3%/100) * Y1 * Y3 * 1/X2 **
Startegy is profitable is made if r>1

The implementation of this strategy is initialized using one starting currency and computes all possible arbitrage paths. It later performs a profitability check on every tick and if an arbitrage opportunity is detected it trades it.

Reinforcement learning:

Supertrend:

Implementation choices :

Order size:

when we say we BUY(<-) a PAIR USD/EUR we are buying the base, ie USD and selling the quote, ie EUR.
==> we are spending EUR.
when we say we SELL(->) a PAIR USD/EUR we are selling the base, ie USD and buying the quote, ie EUR.
==> we are spending USD

this resulted in the choice where
- buy orders are sized in quote currency
- sell orders are sized in base currency

Infra

After some research and experimentation, latency can be critical in such strategy and any way to minimize it is welcome.
as per this awesome work : https://docs.google.com/spreadsheets/d/1W345Qgp1QdcKdyg_hRkzl79OTkGVtxMJJZnJGjtdSp4/edit#gid=0

US East N. Virginia (us-east-1)	US West N. California (us-west-1)	Frankfurt (eu-central-1)	London (eu-west-1)	Hong Kong (ap-east-1)	Seoul (ap-northeast-2)	Singapore (ap-southeast-1)	Tokyo (ap-northeast-1)
0.424	0.16	0.268	0.269	0.075	0.072	0.206	0.023

Each value is the round trip latency in seconds for a REST fetch_ticker() call via the Python ccxt library (average of 10 serial iterations) from the specified server location. the lowest latency is if the app is hosted on aws Tokyo. (ap-northeast-1)

References :

APIs used :

OIHTTP for async WS :
https://docs.aiohttp.org/en/stable/client_reference.html
ASYNCIO for general async:
https://docs.python.org/3/library/asyncio-sync.html#asyncio.Event
BINANCE WebSockets API:
https://binance-docs.github.io/apidocs/websocket_api/en/#signed-request-example-rsa
BINANCE Streams Websockets API:
https://github.com/binance/binance-spot-api-docs/blob/master/web-socket-streams.md

Inspiration:

client webstream implementation:

https://github.com/binance/binance-signature-examples/blob/master/python/websocket-api/websocket_api_client.py

good inspiration:

https://dev.to/ken_mwaura1/crypto-data-bot-using-python-binance-websockets-and-postgresql-db-5fnd

potential improvement:

https://www.thealgorists.com/Algo/ShortestPaths/Arbitrage

Market data ressource :

https://polygon.io/flat-files/crypto-trades
https://www.cryptoarchive.com.au/downloads

How to setup :

>python -m venv venv       
windows :  
>venv\Scripts\activate.bat  
linux :    
>source venv/bin/activate   
>pip install -r requirements.txt   

TODO :

• Strategy runner:

  • Strategy evaluate if quantity on book is enough.
  • Handle and process exec quantity
  • After processing signal, use only last update

• DevOps:

  • Automatic Jenkins deployment on aws zone
  • Grafana monitoring
    • Technical metrics on process health, network traffic, server cpu and mem
    • Functional metrics on current balance, list of trades, PNL per trade and global

• Strategies :

  • Create book recorder
  • Create strategy backtester
  • ML reinforcement learning
  • Arbitrage strategy using bellmanford