predicting-stock-prices

Using an LSTM to predict stock prices

0. Download data from https://www.kaggle.com/borismarjanovic/price-volume-data-for-all-us-stocks-etfs/data and place it in a data/ folder in the project root directory such that you have a "data/Data", "data/ETFs", and "data/Stocks" folder
1. Run python3 train.py
2. (optional) Run python3 test_ood_stocks.py to test your trained model on OOD (out-of-distribution) stocks, which are stocks your model was not trained on
3. (optional) Run python3 test_buy_seller.py to simulate a buying and selling bot in an auction house
4. (optional) Run python3 test_live_data.py to pull live prices off of a stock exchange site and compare the success of the model on a daily basis

TODO:

• Investigate common fault signals in stock prediction ( https://www.investopedia.com/terms/f/false-signal.asp and https://www.investopedia.com/articles/active-trading/052014/how-use-moving-average-buy-stocks.asp)
• Instead of directly passing EMA/SMA into the LSTM, pass raw data into LSTM1, then pass smoothing data into LSTM2, concatenate their outputs, and pass it into a FCN

Possible analysis:

• Plot effect of simple moving average
• Plot effect of exponential moving average
• Show results on predicting of one stock
• Show results on predicting of multiple stocks
• Investigate use of L1 vs. MSE loss
• Write study based around performance of different window sizes (or even hidden state sizes, batch sizes, etc.)
• Investigate whether there's a difference in fluctuations/spikes between popular stocks and non-popular stocks