Price-Based vs Event-Based BTC Forecasting

One reason I got out of stock trading and into crypto trading years ago is because it was clear that stocks were being driven by events and news much more than fundamentals, and it was nearly impossible for me to keep up with the events and news. I think many people shared this concern, and I think the technology of the day was simply not up to the task.

coinpredictor.io now has two forecasts for bitcoin price. One is price-based and one is event-based. Let me emphasize that simply because event-based prediction, or price-based with event data augmentation, is superior in theory, it does not follow that actual implementations are very good. Indeed, we might expect as these models are developed they are rather awful at first. In any case it's interesting to compare.

So how would we know which is more accurate? Forwarding tests are the best tests of models like these, and the two models have significantly different predictions in the near future. Both models predict an uptick in bitcoin price this month of February, although the event-based model is more optimistic on magnitude.

Price-based predicts an average price decline from March 3 to April 3, and a stronger decline from April 3 to May 3. Event-based predicts an increase throughout March which is roughly in-line with the predicted increase in February, on average. Prediction is only given through the first month of April, but it indicates a roughly sideways move from the last week of March, at a much higher average level compared to March's average price.

Specifically, the event-based model predicts a bitcoin price of about $6000 during the first week of April, while the price-based model is calling for a price of $3838 on April 2. That's a huge difference! Let's see who's closer when we get there. If the price falls anywhere in between I'd be happy to use coinpredictor.io as an estimating factor going forward. They estimate a number of other coins as well. For reference, today's bitcoin price is about $3400.