Portfolio project with Reinforcement Learning

1) Data Collection

refer Open source
with preprocessor.py split data into 3:1 (train:test).

2) Momentum-based Investment Strategies

I implemented these strategies based on this paper: Online Portfolio Selection: A Survey

• BAH (Buy and Hold)
  Equally invest m assets at once

• Best
  Choose the best profitable asset in a hindsight

• CRP (Constant Rebalanced Portfolio)
  Rebalance assets to a fixed ratio every period

• EG (Exponential Gradient)
  It is based on "Follow-the-Winner" approach
  It aims to maximize log-return with little change in portfolio value

• Anticor (Anti correlation)
  It is based on "Follow-the-Looser" approach
  It assumes mean-reversion considering cross-correlation and auto-correlation

• OLMAR (Online Moving Average Reversion)
  It predicts future price with moving average
  This method minimizes the change of portfolio value which yields profit more than certain value (epsilon)

3) Simple RL Application

Coming soon !

4) Results

visualize.show_allocation_ratio function

Toy example