Obsidian: Social Analytic Alpha Generator

Inspiration

Consumers are now directly influenced by social media when making purchasing decisions. These can directly influence how well a product does and can have a major impact on how much the stock price of a company changes. For example: if Apple release a new iPhone, consumers typically will make purchasing decision using reviews seen on youtube

Using new technology it is possible to capture this analytic data in the form of data structures contains vital data that can be used by asset managers to beat the market.

What it does

Our system aims to provide a prediction of future asset prices based on social analytic data in real time. Due to the limited time frame, we chose one asset and started processing data. At the time of writing, we were able to manipulate the data structure but were unable to analyse it to produce the future predictions as intended using Deep Neural Networks. We chose one stock (AAPL) to focus on to demonstrate the focus

How we built it

We used the Numpy and Pandas library to group the unstructured data into something useful. For example, Facebook has different types of likes, we organised it to get a useful ratio stat to see the value. We wanted to get some time-series data so that we predict the future price using the price from the day before.

Challenges we ran into

Due to not having any prior experience in Pandas, we spent a lot of time organising and finding the right functions to organise the data.

Accomplishments that we're proud of

Gaining an understanding of Quandl, in particular using the API to import data to manipulate it Using Pandas to produce data frames to produce more structured data for our uses

What we learned

Presentations from Imperial professors and lecturers gave us an insight into how machine learning can be used in testing products in different environments and the importance of behavioural finance to help asset managers make the best investment decisions How pandas and Numpy can be used to producing data in a format to allow convoluted networks to predict the future of data