Inspiration

In recent years there has been tremendous growth in readily available news related to traded assets in international financial markets. This financial news is now available through real-time online sources such as Internet news and social media sources. The increase in the availability of financial news and investor’s ease of access to it has a potentially significant impact on market stock price movement as these news items are swiftly transformed into investors' sentiment which in turn drives prices.

In this project, we use the Thompson Reuters News Analytics data set to construct a series of daily sentiment scores for Technology Stock listed on Nasdaq and NYSE. We use these daily market sentiment scores to study the influence of financial news sentiment scores on the stock returns of these constituents using Machine Learning Model

Relevance Vector Machine (RVM)

RVM is a machine learning technique that uses Bayesian inference to obtain parsimonious solutions for regression and probability classification.

We applied RVM due to the fact that this algorithm could automatically embark on filter selection, kernel method, and density estimation.

In our project, we use this method to forecast the probability density function of the status of single stock (go up or down in the fixed time window). We planned to used different quantiles of prob density function as the scores of over 500 stocks. The strategy could adjust with market-neutral and we can apply different decay and delay period to enhance the sharpe ratio and decrease the max drawdown to a large extent.

BNN (Bayesian Neural Network)

Deep Neural network usually has a high model-capacity which makes it easily overfitted to training data. Applying the overfitted NN to unseen test data could result in over-confident prediction and risky decisions. BNN combines the Bayesian method with a neural network, using variational inference to estimate posterior parameter distribution. It also gives a better density estimation which can help with safe decision making.

Compared with RVM, BNN has more model-capacity and has the ability to model a more complicated distribution of stock status in the real world.