fincl

Introduction

After reading Advances in Financial Machine Learning by Marcos Lopez de Prado I was incredibly motivated to further explore machine learning applications in finance. I built fincl to encapsulate some of those learnings and buildings.

Automating the entire pipeline like I've done here is likely not the way an actual trading strategy would be implemented (unless attempting to build an automated ML platform like H2O Driverless AI) but I chose the mantra of "build to learn" for the techniques described in and learn I did while combining the various disparate techniques described in the book.

I've also included a historical backtest and tearsheet for people not familiar with ML classification reports (despite the many problems with such an evaluation method).

The frontend and backend are fully capable of both classical labeling (a ML classifier generates a trading signal) and meta-labeling (a ML classifier filters a trading signal) and while both methods managed to produce similar prediction scores, inconsistent results in the walk-forward PnL simulation of meta-labeling combined with a self-imposed deadline have made me include only the classical labeling results herein for the time being.

Installation

Backend

I recommend installing via anaconda, I used anaconda 4.8.2 on Windows 10 to generate and test the enclosed environment.yml:

To create a Python 3.6 environment and install the required dependencies:

conda env create -n fincl -f backend/environment.yml

I used nbdev to develop the backend, which means code lives in .ipynb files certain cells are marked to be included in the final .py package. I.e. you develop your functions inside jupyter notebook and then use nbdev_build_lib to build your library.

Frontend

I used yarn version 1.12.1 to manage & develop the frontend, installation and running the dev server should be as simple as

cd frontend/
yarn
yarn start

Inspiration & Thanks