← Back to portfolio

SK C&C · Jul 2018 – Feb 2019

Portfolio Robo-advisor

Role · Reinforcement-learning models, time-series analysis

01Background & Goals

  • A system that manages a portfolio of user-chosen stocks and rebalances it automatically, considering price changes and trading costs.

02Contributions

Reinforcement-learning trading
  1. Built an ensemble deep Q-learning model that accounts for the portfolio, price changes and trading costs
Stock recommendation
  1. Embedded the behavior of co-moving stocks into vectors with Word2Vec (Stock2vec)
  2. Clustered the vectors with K-means to recommend stocks that move alike

03Tech Stack

Framework / Platform
PyTorch, scikit-learn
Methodology
Deep Q-learning, Word2Vec, K-means

04Results

  • Applied to a paper-trading system