Overview / Usage

This project is in the intersection of algorithms, quantitative finance, and machine learning. I will be making portfolio selections using Multiplicative Weights, an online learning algorithm. This approach can be adapted to an ensemble method by using various supervised learning methods as the 'experts', to which the algorithm assigns weights, based on their correctness.

Methodology / Approach

Hoping to build upon results from UC Santa Cruz, I am using the multiplicative weights algorithm. This selects from a number of 'experts', and assigns weights each day based on the correctness of each 'expert'. This is an online learning algorithm, and a major modification I've made is in selection of these 'experts'. I will aim to employ time series algorithms. So far, I've accumulated the following algorithms to be experts: LSTM, ARIMA, HMM, Linear Regression. These algorithms all have a core commonality: they all process sequential data to make a prediction.

There are various variables to be tuned, such as:Timescale, training time of mult weights algorithm before measuring success, amount of provided training data, experts, data format ($ prices, % increase, $ increase)

Technologies Used

For code, I am using Python with the following libraries: Keras, Scikit-learn, Tensorflow, Pandas, NumPy.