The main issue with supervised learning models like LSTMs or XGBoost is that they focus on point-in-time predictions(at time t).

This often leads to a significant discrepancy between backtesting and live trading results, primarily due to overfitting and the non-stationary nature of the market. 

This is why current research is shifting toward Reinforcement Learning (RL), which is better suited for non-linear market dynamics and optimizing for long-term cumulative returns rather than just immediate price direction. 

However, keep in mind that RL isn't a silver bullet. Achieving consistent profitability requires immense effort in environment modeling and reward function design. Simply switching to an RL model won’t guarantee stable profits without a very robust strategy and rigorous validation.


레딧에 누가 질문올렸길래 이렇게 답변해줬다 ㅇㅅㅇ