A very bad stock prediction model. Goals of the project include
- Making a trillion dollars
- Implementing prediction and trading models, using recurrent neural networks and reinforcement learning, respectively
- Building web scrapers (possibly passing headlines through a CNN) to generate more input data
- The worst
- Input n-day moving average, output of the next k-days moving average prices
- Tries to predict absolute values (emphasis on tries)
- Generally terrible all around
- Aim to predict single day price changes instead of absolute values
- Can call model repeatedly, inputting the last predicted price, to generate a sequence of future prices
- Associate a confidence metric that can compound over time, resulting in longer predictions being less likely
- Add in supplemental information - news headlines, Reddit comments, etc
- Either hand-selected features (polarity), or pass through another network for processing