• Our group had two main goals for this project.
  • Our first goal was to find out which combination of moving averages produced the best trading signals and portfolio performance.
  • The second goal was to figure out which type of model would best predict the price of Bitcoin.
• Going into this we made no assumptions about which type of models to use, however we researched some of the most used models in crypto trading and tested them.
• Models that we tested include LSTM, SVR, SVR-Tuned, Ridge and Lasso.

• We reviewed the entire Bitcoin dataset to find optimum moving average combinations.
• We settled on using the Short Moving Average of 20 days and the Long Moving Average of 80 days.
• Trading signals produced when Moving Averages cross these thresholds would generate a portfolio of ~$10K with an initial investment of $3K.
• Our signals were simply buy when the short crosses over the long moving average and sell when the short crosses below the long moving average.
• Volume or other inputs were not used in the signals, simply just the cross over points.

Here is a graph showing the portfolio value over time with buy signals in purple and sell signals in yellow:

Here is a graph of those same triggers with the moving averages plotted all on a logarithmic scale:

We ran some manual calculations of the RMSE and R2, which were based on the test data set.

Results:

RMSE_LSTM = 0.280881 R2_LSTM = 0.051536