I am a trader first before I started learning programming (not too long ago tbh). I was satisfied to know that my usual strategy in day trading could be replicated in regular algorithm coding (i.e manually selecting stocks to backtest) with decent positive performance. However, this allows for human selection bias as I have to manually choose a stock to trade and was convinced using the algorithm framework can improve the performance.

I have since attempted to replicate my strategy within the algorithm framework using a dynamic universe and alpha model. The main issue now is that the performance yields a horrendously negative performance. I'm not sure where to pick this apart. 

Feedback, thoughts, suggestions, and questions are all welcome.