How well does a Neural Network (NN) predict the closing price of Bitcoin (BTC) 30 days in advance, given five years worth of data about day to day price movements, certain technical indicators and the Google Trends number? I decided to find out.
I developed a linear model (lm) and a NN in R. Then, I created a Tableau Story to show the outcome, including the coding from R and predictions. You can view the Tableau Story here:
Out of curiosity, I also had the NN predict the closing price of Bitcoin on 11/25/19. The actual closing price was $7,122.14. The prediction was $8,881.25. Not great – the recent big drop and increase in volatility probably played a role in it being so far off.
In addition, I wanted a prediction 30 days out from 11/25/19, which is Christmas Day. The NN predicted a closing price of Bitcoin on Christmas of $7,988.26.
After checking the closing price of Bitcoin on Christmas Day, it looks like it came in at $7,195.79. A ways off from the Neural Network prediction of $7,988.26. A few days earlier it was a lot closer to that price, but took another tumble before Christmas. Bottom line, BTC can be so volatile that it’s very difficult to predict where it will be down the road accurately!