Optimizing portfolios (linear combinations of shares) based on stochastic continuous-time models is notoriously difficult. We architect a collection of neural networks and regression estimates to approximate the optimal portfolio allocation of any number of shares based on perceived fair price-informed by current sentiment, historical performance, and competitor performance. We also architect a means to divide the market and train on smaller segments whilst retaining a characteristically high degree of accuracy. This model is of particular significance by virtue of its characteristically low training time and nano-second scale queries.

Many quantitative funds rely on data collection from buying user trades from brokerages while other funds attempt to make money off of small differences in price over minuscule amounts of time. This paper attempts to architect a python[1] system by which we analyze signals from Markov Chains[2], Polynomial Regression, and Sentiment Analysis latter passed into Decision Trees[3] for a holistic evaluation based solely on publicly available information.

A document by Siddharth Mohanty. Click on the document to view its contents.