Portfolio Analysis by Machine Learning

Note: This article was originally published on December 21, 2017 and has been migrated from our previous blog. Some details — tools, libraries, benchmarks, industry context — may be outdated. For our latest perspective, see our recent posts.

The stock market has always been a core interest in predictive analytics. Thus it came as no surprise when one of our clients sought our assistance in his investment strategy.

We designed an intelligent asset allocation system consisting of a Long Short-term Memory Recurrent Neural Network (LSTM), which predicted the expected return for every asset in a portfolio. In order to estimate an asset’s risk, a novel technique called Variational Dropout was used. Another neural network was used to determine the optimal set of weights for the portfolio’s assets.

These combined algorithms provided an investment strategy with returns comparable to Wealthfront.