What is hftpy

This repository will include all kinds of tools and code I have built and optimized over the years. It is the tip of an icerberg, as this code is resulting from many iterations and algos built in high frequency trading/market-making.

Code speed and efficiency

Not everything needs to be C++, as I have seen many successful and profitable algos built in Python over the years. Optimizing code and hardware is a centralized game for the likes of Jump, Citadel, XTX etc...and you should get into it only after milking out all alpha and potential PnL from alpha/statistical model, algo and code design. Long story short, early in building the highest ROI on time comes from other pieces of your trading system, not code speed and gaining millis or nanos here and there with C++ or else.

A scalable implementation of online transforms and order book feature extraction

This is a simple and scalable implementation of 2 critical components of algo trading: the order book and exponential moving averages. This implementation illustrates the benefit for understanding of tech+quant+trading aspects while developing algo trading frameworks.

Features:

• CustomOrderBook class with utility quant functions

  • quant functions for alpha and trading execution

• ExponentialMovingAverage class

  • Same code for live and backtest
  • Memory efficient: no need to store an array of past data
  • Fast: no need to loop over past data

• Online compute

  • Offline moving statistics (like MA) have many downsides: need to store arrays, loop over them, don't represent full-history, high impact of outliers when they drop out of lookback
  • Online implementations have many benefits: only need to store 1 value due to recursive nature, always represent the full-history of data, impact of outliers is smoothed out

• Scalable architecture

  • Can be used in a live trading environment
  • Can be used in a backtest environment
  • Can be used in a distributed environment
  • Can be used in a single threaded environment

Running the script

Any recent Python+cryptofeed version would work. The only custom order book class included is gateio .

python run_data_listener.py -e GATEIO -s BTC-USDT ETH-USDT --key_id "<your_key_id>" --key_secret "<your_key_secret>"

Here an example of the ExponentialMovingAverage class. We set the required_n_warmups parameter to 100. The image below shows that at the 100th tick, the EMA finished warming up and turned from "None" to a value.

Cythonize the code

python setup.py build_ext --inplace