(3) Is the algorithm learning?
Yes, in some parts, but no on other parts. And it’s not all about the algorithm! So let’s break it down:
To start, you might know the saying “garbage in, garbage out”. That’s because as with most models, it’s only as good as the data you give it and, in this case, the coins it can choose from. That’s why our team reviews coins daily to see whether or not they are good enough to make it into the investable universe. We also impose the constraints (such as the maximum and minimum it can hold in any one coin). Our team is made of real people, and they learn.
Next is the “algorithm”. The portfolio construction Artificial Intelligence (A.I). We say it’s “Artificial Intelligence” because it’s made up of two subfields of AI. : Machine Learning (ML) and Operations Research (OR). OR-based techniques have demonstrated the ability to identify optimal and locally optimal solutions for well-defined problem spaces, and the ML part is ideal for learning with constraints. So it’s good at learning from the ever-increasing amount of data that it gets, but it benefits from the OR-side for portfolio construction, which is all about finding an “optimal” portfolio.
Then there is the aspect of translating all that into actions: how to trade, when, how much, where,... To make sure that the portfolio we’re holding is identical to the optimal one. And that is also up to the team, not the A.I. And careless execution would result in a worse portfolio than the optimal one recommended by the A.I., so we need to do a good job there too.
In conclusion: it learns some, but it isn’t alone. We learn too.
(4) Why don’t you have stablecoins in the universe so that the Autopilot can “retreat” when the market takes a dive like it did this month?
Indeed, we have no stablecoins, no fiat, no gold or commodities (which you can easily find tokenized), or anything other than true crypto projects in the investable universe.
The point here is that you can get exposed to the best of what crypto has to offer, but nothing else. And all positions are long so there are no shorts, no derivatives, no options, or anything else. Just plain crypto.
The reason is twofold:
1/ The algorithm seeks "a monotonic growth" which is done by minimizing length and depth of drawdowns and diversifying aggressively (which is the reason why it rarely holds ETH for example because ETH and BTC move in tango, but BTC has less downward volatility, so it sees it as better to satisfy the goals). Also, it invests for the long term whenever it makes a decision (2.5 to 5 years).
So, knowing this and taking into account the volatile nature of crypto, and we saw this in all our backtests, the result is that if you give it cash, it pretty much (not always, but still) takes the largest position in cash that it can.
It’s important to remember that there is no “momentum” built into this algorithm. It builds portfolios but doesn't trade (scalping, intraday or swing).
So it would not "try to retreat" very much to cash when seeing the market going down because it’s looking at the long term and long term the market is going up. With a great deal of volatility, but up nonetheless. So yes, if the market is down it follows. And when it's up it follows.
We chose this route because we think that time in the market is more important than timing the market. Morningstar proves this about once a year in their “active/passive” barometer. It’s also the reason why you can find a lot of literature about why investors have lower returns than investments, regardless of the asset class.