Nice post. 60% accuracy is really good. Is that on your test data? It would be interesting to run a live test of your model (even if it is just paper trading) to collect true out of sample performance data.