Data 301 Quarter-long Project: A trading bot combining sentiment analysis on web-scraped articles and quantitative analysis of stock trends in order to enact stock trades.
The trading bot uses the results from both the quantitative and the sentiment models for a particular stock to make a decision of whether to buy or sell the stock each day. The bot has been trading on a paper account using the Alpaca API, and through its first 2 weeks trading, earned about a 4% ROI.
The data for the quantitative model is gathered using a wrapper for Twelve Data, an online purveyor of stock metrics. An important part of the data collected is the Moving Average Convergence Divergence (macd), along with percent change and volume.
The model consists of two input branches, a Long Short Term Memory (LSTM) network and a standard sequential neural network with one hidden layer. These two branches are then concatenated with a pooling layer. This model achieved F1 scores as high as 0.71 for certain stocks when evaluated on future stock prices.
Visuals were created for all 12 stocks being tracked. To see other prediction visualizations, see the visuals directory.
The Google News API is used to search and collect headlines for all of the companies being tracked. These headlines are filtered by company name and/or ticker, and then input into the model to get a sentiment prediction for each company.
The model is a Convolutional Neural Network (CNN) with an initial embedding layer followed by a single convolutional layer, which is then pooled and flattened for the two dense layers, the first using a rectified linear activation function and the final layer using a sigmoid activation function. The model was trained using a data set of 1,967 financial headlines with labeled positive or negative sentiment from an investor's perspective. When evaluated, the model achieved an F1 score of 0.87 for positive articles.
Given headlines that imply a positive outlook for a company, the model should output a prediction greater than 0.5. Likewise, headlines that imply a negative outlook should receive predictions less than 0.5:
| Input | Output |
|---|---|
| "Apple Is Reportedly Working on Custom Silicon for Apple Car" | 0.96578010 |
| "Tesla Beats on Profit, Reaffirms Goal of 500,000 Deliveries" | 0.98735297 |
| "Struggling Nikola Stock Still Has Further to Fall" | 0.06987518 |
| "Purdue Pharma to Plead Guilty, Pay $8.3 Billion Over Opiods" | 0.114830494 |