Demo of an RNN trained on NASDAQ:GOOG. Trained on GOOG pricing history from 2012-2016. Predicts full month TREND* for January 2017. Can be applied to other stocks with minimal tuning and .csv file
*Due to Brownian Motion, stock prices are independent of previous data. Thus the exact price cannot be identified, however, the Neural Network is able to identify patters and trends. This means the predicted graph will look very similar to the real graph, but price points may differ.
Todo:
- try different timesteps, currently uses 3 months of previous data to make next days prediction each iteration.
- trying different number of neurons per layer
- adding higher dimensionality thorough layerization and/or 3rd dimension indicators.
- changing batch size, learning rate, optimizers and loss function
- deploying program with newer data (up to June 2019) and testing performance
- deploying on smaller priced stocks with less volatility (FB, Sprint, Tesla)
- changing epochs (currently 100 iterations until full train)