Deep Reinforcement Learning & Feature Extraction For Constructing Alpha Generating Equity Portfolios

November 8, 2021

Machine Learning & Portfolio Management

Deep Reinforcement Learning & Feature Extraction For Constructing Alpha Generating Equity Portfolios The ambition of this paper is to catch hidden information inside the Securities and Exchange Commission’s data in order to construct an equity portfolio that maximizes returns.

Moreover, our data will have a lack of knowledge about actual investment decisions and directions. As a result, or rather, to remedy this lack of knowledge. We will use feature extraction in order to filter out the best data through several criteria.

We propose a method employing powerful machine learning techniques (Deep Reinforcement Learning Machine Learning) to try to catch the missing pieces of information behind the decision process and use them as a prediction tool to construct quarterly equity portfolios.

In addition, this approach reached an annualized return of 21% with a sharpe ratio of 1.8 outperforming the S&P 500 both in returns and stability through historical backtesting.

After filtering data and extracting features, both the deep reinforcement learning and random forest machine learning models will be backtested with historical stock market data and evaluated by their rate of return (i.e. the parameter that determines the model’s degree of success in alpha cloning).

By incorporating extracted features into logistic regression and XGBoost machine learning models, stock price movement can be forecasted and from there, portfolios can be constructed.

So we will view the portfolio as the agent. Moreover, which undergoes the action of buying and selling stocks to achieve different states. Or stock pools with their respective features.

In conclusion, if the state accomplishes the reward of excess positive returns, the model “learns”. Which trades to execute given the existing portfolio to maximize total return. The second approach employs a random forest machine learning model. One which employs decision tree algorithms to predict the probability of positive returns for stock pools. Lastly, and from there, constructing a portfolio with the 50 stocks that will yield the highest return.