Machine Learning Picked My Crypto Portfolio For 100 Days — These Are The Results
Over the course of the last 100 days, I have meticulously tracked the performance of my portfolio. While my portfolio itself is nothing spectacular, there has been widespread interest in the machine learning strategy I’m using to pick the assets that go into my portfolio.
The idea struck me 105 days ago. Wouldn’t it be cool to build a complete cryptocurrency portfolio strategy entirely based on machine-learning predictions?
I set out to automate my entire portfolio using predictions generated by machine learning models.
So, that’s what I did. After a few days to get everything set up, I was able to launch a portfolio strategy that only relied on machine learning predictions to decide what cryptocurrencies I would buy each week.
Over the course of the last 100 days, I have continuously updated the portfolio based on weekly predictions, tracked the progress, and ultimately inspired thousands of other investors and traders to consider similar alternative strategies when building a crypto portfolio.
How does the portfolio strategy work?
Once a week on Monday, I use the Nomics 7 day predictions to update the assets that should be included in my portfolio. Using 100% real funds, I manage the portfolio through Shrimpy.
Managing the portfolio through Shrimpy means I can be a social leader and allow other machine learning lovers to follow the portfolio updates in real-time. Shrimpy also has powerful trading tools, so I’m never required to log into my exchange account to manage the portfolio.
Results
The results so far in the study have been astonishing. Not only has the machine learning portfolio strategy outperformed Bitcoin, the market, and other popular strategies, but it’s done so in an exceptional fashion.
Over the course of the last 100 days, the machine learning strategy has already peaked at a nearly 250% performance increase. This performance is a 4–5x increase over Bitcoin and the rest of the market which has only seen a 50–75% increase over the same period.
Interestingly, we see a subtle pattern in the results. The performance of the machine learning strategy rises aggressively for several weeks, then experiences a down period, then rises aggressively again for another few weeks.
There is no saying whether this behavior will continue to repeat itself, but it demonstrates the instability of the predictions at some times.
Additional information about the latest results can be found in the week 15 study update. Every week, an update is published to review the results from the previous week, layout the predictions for the next week, and summarize any thoughts.
Discussion
The following discussion will be based on my opinions and insights into the strategy. Take these thoughts with a grain of salt.
In my opinion, Nomics appears to be good at identifying assets that have been winners and betting on those assets to continue winning. I have seen fewer instances where Nomics is able to predict a breakout before it happens.
Essentially, if an asset has been flatlined for a few weeks, Nomics seems to predict that the asset will continue to be flatlined for the next week. Logically, this makes sense because that is the most probable result.
However, once an asset experiences a breakout, Nomics feels like a good tool to decide whether that breakout will continue or reverse.
Many of the selections by the machine learning strategy have been highly volatile assets. This results in large swings from week to week. One week the strategy could be up 70%, then the next week it could be down 20%.
The high risk and high reward nature of this strategy does not make it suitable for many investors. It would be sensible to carefully think about how this strategy could impact your portfolio if utilized.
Identifying areas of risk and mitigating those risks is an important part of portfolio management. This is why we’ve built Shrimpy in a way that we can divide funds into multiple portfolios, allocate multiple strategies at a time, and take a long-term view of the market.
Conclusion
The first 100 days of the Machine Learning Case Study have produced exceptional results. As always, however, that does not guarantee future performance.
Although 100 days might have been enough time to convince some people that machine learning can have a place in portfolio management, it’s not time to close the study down yet.
I encourage everyone to join in on the study by signing up for Shrimpy, favoriting the MLCaseStudy leader, and monitoring the performance over the next 100 days.
See you on Shrimpy!
