Wouldn’t you like to know where Bitcoin is going? I sure would. The prices of Bitcoin and other cryptocurrencies seem to be all over the place. They go up and down much more than regular financial assets like stocks and bonds. And even for those traditional markets, predicting future price movements is one of the hardest prediction problems there is.
On the other hand, the rewards could be huge: if you can manage to be 51% right and 49% wrong on average, you will make great money long-term and the sky is the limit. That’s why I’m creating a system for automated prediction of crypto prices.
This is the approach:
- Once a day, predict prices 24 hours ahead.
- Only tell me whether they’ll be UP or DOWN, not by how much.
- Whether we get one big move right or many small moves, the system is aimed at maximizing overall returns on investment.
- Transaction costs are included, because they can add up quickly.
The system is focused on rewards only. It does not try to avoid volatility of returns, nor occasional very bad returns, nor prolonged periods of negative returns. Instead, I manage risk by:
- Not investing all my savings in this.
- Diversifying among cryptocurrencies (although many are correlated with Bitcoin).
- Only switching between “long” and “neutral” positions, not taking short positions.
So far, I have 3 separate models running in production mode. Below you can see the results for a simulated trader using the predictions (solid line), and compare it to the price index of the cryptocurrency (dashed line).
The background colour indicates whether the model said BUY (green) or SELL (red).
- Although the models have lost out on recent gains, I’m happy they avoided the worst of the drop when the coronavirus hit.
- The graph represents a very short time period, way too short to draw any conclusions about the validity of the models. The only reason I’m showing this is to give you a taste of what I’m working on.
- Overall portfolio results would depend heavily on how you allocate funds among the models.
- Data are automatically updated daily.
28 Apr 2020 update
- Added historic predictions, going back to 18 Feb 2020.
- Added green/red bars to indicate BUY/SELL actions.
- Models for REP, ADA, DASH, XLM, etc.
- A more thorough evaluation of historic predictions, quarter by quarter.