In the problem, we have been asked by a trader to tell him when he should buy, hold, or sell the assets in his portfolio (Cash, Gold, Bitcoin). He only has $1000 cash on 9/11/2016, which is the time we start. We accomplished a prediction and planning model to run a five-year period to tell the trader when to buy, sell or hold the assets, resulting in earning $179,580 with the $1000 start cash.

I was in charge of the building of the planning model, the integration of the final model, and the visualization of the data.

Inspired by the fact that the cross point of moving averages (MA) can represent the trace of profit and loss, I build SLTMA by introducing 5-days-MA and 30-days-MA and do purchase/sell when they are crossed.

Drawback:

1. Severely affected by the short term volatility。

2. When the intersection between short-term and long-term moving average occurs, the local lowest or highest price point must already be passed. Even one day of acting ahead will result in a large improvement in return in the model.

Encouraged by the previous success, we further optimize the model by introducing more dependable economics like the Sharpe Ratio and the Credit of Risk Return. Upgrading those factors with LSTM, we calculate a final credit, Credit to Action, to determine if the trader needs to act, and the proportion of assets he will use to purchase/sell.