Peer evaluation: CooperNederhood

[CooperNederhood]

Cool combination of NLP with financial data!

My main suggestion is related to your overfitting woes, as I had a similar issue. You could look into either ridge regression and lasso regression which are linear regression approaches that penalize for model complexity, thereby reducing the likelihood of overfitting. You can also use a cross-validation approach to train the model which should get you better predictive power.