We inspired ourselves from the overgrowing influence of social media on the prices of stocks. When Kylie Jenner posted about snapchat the company lost 1.3 billion dollars. We wanted to make simple models and combine them together to try to predict stock prices.
Another thing we tried to implement was GUI for traders to manage their portfolio more easily, and for them to understand the quantitative analysis in a more efficient and understandable way.
What it does
The main objective of the first model is to use Facebook's posts (to company) sentiments (positive neutral undefined negative) and engagement to predict the change in the company's stock value. In the project, data is organised as pandas dataframe.
Multi-Linear Regression Model is used to predict the effect of different sentiments to tock value. It turns out to be [ 0.00014944 0.00063176 -0.0006409 -0.00067026]. In other words, per positive post increase the company's stock value by 0.00014944.
Additionally, a simple linear regression model is used to test and to predict the future.
Also py.plot exists in the project to visualize data and make things clear.
The historical data compares the moving average of the stock for the previous 3 days with the future price of the stock, training a linear regression model on 80% of the data and testing it on another 20%. Then we did the same for competitors stock change and the subsequent change in the company in questions stock. Lastly we computed the quarterly revenue vs the subsequent average stock price over the next 3 months.
Eventually once our models are more accurate we expect to combine them in a weighted average, based on their accuracy scores.
We also built a GUI using tkinter where the trader can see his stocks, see how much money he's made and search for stocks and their future predictions, but as this has not been able to run we have not uploaded it to depots.
Challenges we ran into
The availability of data is quite restricted, most of the pages required a premium subscription. We do not have a lot of finance knowledge, and we are not computing students which made it difficult to implement some machine learning.
What's next for RayRox
Future Development: It is based on MCD now, but can be generalized to all reach, reply and other features can be included. Improve our model, look at other social media platforms and indicators of a company's performance. Improve the GUI. And hopefully expand it to be able to work with any other stock. We would also love to get our new scraper to run efficiently.