/InteractiveBrokers-PairsTrading-Algo

A high-frequency trading model using Interactive Brokers API with pairs and mean-reversion in Python

Primary LanguagePythonMIT LicenseMIT

Purpose

A simple trading equity trading model on Interactive Brokers' API dealing with (pseudo) high-frequency data studies.

Requirements

  • Python 3.7
  • IB Trader Workstation Build 973.2
  • IB paper or live trading account

What's new

14 Jun 2019

Special Thanks to chicago-joe for updating to work with Python 3.

19 Jun 2019

  • Version 3.0 released
  • ibpy library is dropped in favour of the newer ib_insync library.
  • The same code logic is ported over to use the features of ib_insync, compatible with Python 3.7. Includes various code cleanup.
  • Dropped matplotlib charting in favour of headless running inside Docker.

Setting up

Running on a local Python console

Steps to run the trading model on your command line:

  • Within a Python 3.7 environment, install the requirements:

      pip install -r requirements.txt
    
  • In IB Trader Workstation (TWS), go to Configuration > Api > Settings and:

    • enable ActiveX and Socket Clients
    • check the port number you will be using
    • If using Docker, uncheck Allow connections from localhost only and enter the machine IP running this model to Trusted IPs.
  • Update main.py with the required parameters and run the model with the command:

      python main.py
    

Running from a Docker container

This step is optional. You can choose to deploy one or several instances of these algos on a remote machine for execution using Docker.

A Docker container helps to automatically build your running environment and isolate changes, all in just a few simple commands!

To run this trading model in headless mode:

  • In TWS, ensure that remote API connections are accepted and the Docker machine's IP is added to Trusted IPs.

  • Ensure your machine has docker and docker-compose installed. Build the image with this command:

      docker-compose build
    
  • Update the parameters in docker-compose.yml. I've set the TWS_HOST value in my environment variables. This is the IP address of the remote machine running TWS. Or, you can just manually enter the IP address value directly. Then, run the image as a container instance:

      docker-compose up
    

    To run in headless mode, simply add the detached command -d, like this:

      docker-compose up -d
    

    In headless mode, you would have to start and stop the containers manually.

Key concepts

At the present moment, this model utilizes statistical arbitrage incorporating these methodologies:

  • Bootstrapping the model with historical data to derive usable strategy parameters
  • Resampling inhomogeneous time series to homogeneous time series
  • Selection of highly-correlated tradable pair
  • The ability to short one instrument and long the other.
  • Using volatility ratio to detect up or down trend.
  • Fair valuation of security using beta, or the mean over some past interval.
  • One pandas DataFrame to store historical prices

Other functions:

  • Generate trade signals and place buy/sell market orders based on every incoming tick data.
  • Re-evaluating beta every some interval in seconds.

And greatly inspired by these papers:

And book:

Step-by-step guide to more trading models

Mastering Python for Finance - Second Edition

I published a book titled 'Mastering Python for Finance - Second Edition', discussing additional algorithmic trading ideas, statistical analysis, machine learning and deep learning, which you might find it useful. It is available on major sales channels including Amazon, Safari Online and Barnes & Noble, in paperback, Kindle and ebook. Get it from:

Source codes and table of contents on GitHub:

Topics covered with source codes:

If you would like a FREE review copy, drop me an email at jamesmawm@gmail.com.

Suggested enhancements

Some ideas that you can extend this model for better results:

  • Extending to more than 2 securities and trade on optimum prices
  • Generate trade signals based on correlation and co-integration
  • Using PCA for next-period evaluation. In my book I've described the use of PCA to reconstruct the DOW index. Source codes here.
  • Include vector auto-regressions
  • Account for regime shifts (trending or mean-reverting states)
  • Account for structural breaks
  • Using EMA kernels instead of a rectangular one
  • Add in alphas(P/E, B/P ratios) and Kalman filter prediction

Disclaimer

  • Any securities listed is not a solicitation to trade.
  • This model has not been proven to be profitable in a live account.
  • I am not liable for any outcome of your trades.

Is this HFT?

Sure, I had some questions "how is this high-frequency" or "not for UHFT" or "this is not front-running". Let's take a closer look at these definitions:

  • High-frequency finance: the studying of incoming tick data arriving at high frequencies, say hundreds of ticks per second. High frequency finance aims to derive stylized facts from high frequency signals.
  • High-frequency trading: the turnover of positions at high frequencies; positions are typically held at most in seconds, which amounts to hundreds of trades per second.

This models aims to incorporate the above two functions and present a simplistic view to traders who wish to automate their trades, get started in Python trading or use a free trading platform.

Other software of interest

I write software in my free time. One of them for trading futures was simply called 'The Gateway'. It is a C# application that exposes a socket and public API method calls for interfacing Python with futures markets including CME, CBOT, NYSE, Eurex and ICE. Targets the T4 API.

More information on GitHub: https://github.com/hftstrat/The-Gateway-code-samples or view on the website.

Final notes

  • I haven't come across any complete high-frequency trading model lying around, so here's one to get started off the ground and running.
  • This model has never been used with a real account. All testing was done in demo account only.
  • The included strategy parameters are theoretical ideal conditions, which have not been adjusted for back-tested results.
  • This project is still a work in progress. A good model could take months or even years!