Stock Predictor and Portfolio Optimizer

A novice's attempt for (weekly) stock prices prediction and portfolio optimization.

Technologies used -

• Python 3
• Scipy
• Scikit-learn
• matplotlib

Data Source

• Quandl Data platform

Major Features

• Prediction of stock prices on weekly basis (predicting next week's prices for a given stock)
• Optimizing portfolio distributions ( optimum ratio of each stock in a portfolio to potentialy maximize profits )

Algorithms and implementation mechanisms

• Weekly prediction is being done using KNN Regressor using "Bollinger Band Value" and "Simple Moving Average" as input features.
• Stock porfolio optimizer is done for maximizing the "Sharpe Ratio" or "culumative returns", using scipy minimizer (minimizing for -1 * value)

How to run?

• predictorWeekly.py is a weekly stock prices predictor.
• optimizer.py is a portfolio optimizer.
• To run, install python 3 with pip, and then simply install all the libraries mentioned in the script file that you are trying to run, for e.g. pip install pandas
• Once done, you can just run these like any other python script.
• Data is already added in the repository under backend/data folder.

Testing and results

• This application has been tested for National Stock Exchange, India. The weekly predictions have upto 75 % corelation with the actual results, for the leading (largest market capitalization) stocks.

TODO

• Add all dependancies in requirements.txt
• Creating web services
• Creating Web-based front-end.
• Improving accuracy by adding more Fundamental and Technical features

Notes

• As I've mentioned, this implementation is at early stage. If you are an Machine Learning or Stock Market Enthusiast / Expert, feel free to suggest improvements / corrections by creating an issue (or you contact me at vishwajeetvatharkar@gmail.com )
• As I have beginner-level skillset in Python programming language, I might have missed many of the best practices and architectural patterns specific to Python ecosystem, please feel free to suggest some improvements.
• This application was part of my academic project coursework (Major Project, Engineering Final Year, Information Technology)

Credits

• The algorithmic implementations are inspired by the suggestions given by Prof. Tucker Bach in his MOO Course series on Udacity - Machine Learning for Trading