Python implementation for "Deep learning with long short-term memory networks for financial market predictions" working paper by Thomas Fischer and Christopher Krauss. https://www.econstor.eu/bitstream/10419/157808/1/886576210.pdf
Simply collect and load three years of data for the S&P 500 constituents. You can do this easily using https://github.com/nicolasvianavega/stock_price_time_series.
The output consists of:
- Trained Keras model (json)
- Model weights (h5)
- Pandas data frame consisting of normalized stock returns and up/down probabilities for each stock at any given day for a test dataset
# Collect S&P 500 Constituents data using stock_price_time_series
symbol_list = ['AAPL','GOOGL','TSLA']
since = '2015-01-01'
until = '2018-01-05'
value = 'adj_close' #recommended
results = data_collect(symbol_list, since, until, value)
results
date | AAPL | GOOGL | TSLA |
---|---|---|---|
2015-01-02 | 170.9074 | 1073.2100 | 320.5300 |
2015-01-03 | 170.8776 | 1091.5200 | 317.2500 |
... | ... | ... | ... |
2018-01-05 | 173.6259 | 1110.2900 | 316.5800 |
After collecting, just load the data into the script