Numerai is an AI-run, crowd/open-sourced hedge fund/trading platform where competitors try to predict the movement of financial markets.
I used an LSTM-RNN for regression analysis on Numerai's proprietary data and I used the same model to predict the stock movement of Delphi Technologies and their split-off, Aptiv PLC.
Delphi's split-off[1] in December 2017 altered the company's original S & P 500 index. I recovered the data Delphi lost after the split and merged it into one dataset to predict the directional movement of their stock prices based on the Opening and High price of the original index.
On the Delphi/Aptiv data, the model had a final loss of 1e-2 at 140 epochs:
And with better parameter tuning the model was able to converge to the training data at 35 epochs with good results:
The model was validated on Delphi/Aptiv's Open-High price from 5/9/18 to 5/20/18:
The validation set converged with a final loss of 1e-2 at 60 epochs:
The model almost found an approximation, but it was ultimately unable to generalize to the test set.
