This was sent as an email to all users, but we thought it could also make a reasonable blog post to let non-users know what we are currently doing.
1. Binance hacked. Action Required.
Binance’s hot wallet was hacked today in the morning, for a loss of about 7000 bitcoins. All Binance customers’ funds are safe as Binance will cover that with their own insurance (what they call SAFU). We still believe that Binance is the most secure exchange, but of course it raises questions about whether we should consider a connection with more exchanges.
Binance released this news earlier today: Security Update: API Key Reset
Apparently they are deleting all API keys (although it is very poorly written). This means the keys you are using with Airbag will be deleted too and you need to generate new ones.
Action required: You will need to repeat the process of creating a new API key in Binance and connecting it to Airbag. If you have trouble or questions, reach out.
You might also want to change your Binance 2FA. Meanwhile, for your peace of mind, we are taking the action of deleting your current keys on our end.
2. Users
We currently count 151 users, whose bots are managing a total of $31k. Most people started with small amounts as they gained confidence.
It might not sound like much, but each of you who has entrusted us with a single dollar really means a lot. It also proves our value proposition of simplicity and orientation to risk mitigation. Thanks to all of you.
3. Feedback
We have received several suggestions for improvements, including: Mobile App, better statistics, more graphs, more coins, etc. But really, 60% of you have said to us: “Just focus on improving the algorithms”, and that’s what we are doing, so don’t expect much visual improvement of the app for now (although later today we are releasing a new shiny landing page). Everything is happening behind the scenes to improve returns.
With regards to the mobile app, Airbag is a WebApp. At some point we will wrap it inside a native mobile app, but it is too time consuming right now. If you want a great mobile experience, we suggest you add the website to your home screen. It looks native removing the navigation bar (for iPhone ) (for Android )
4. Performance
In principle we are not going to speak about performance in future updates. There will be good months and bad months, and we think that performance can only be measured in the long term and with a lot of patience. We also think performance is for users to judge.
Nonetheless, because this is the first month of operation, we will share that we are quite pleased with the performance of the current bot. While it is true that Bitcoin had a higher total return per $, Airbag had a better return per unit of risk, or Sharpe, which is what we aim to optimize. The fair way to think about return/risk is that you could have invested more dollars with Airbag at the same risk as investing fewer dollars directly on Bitcoin, leading to more overall return. It is our objective to educate more people about risk.
Furthermore, the bot will shine in down-markets and sideway-markets, so it is quite good that it has kept reasonable pace also in up-markets.
However there are some things to improve too. The trades on May 3rd and 4th were too sensible and should have been more patient. We are reviewing the behavior.
5. Algorithm improvements
The algorithms are proprietary. This was a decision we made from the beginning. There are many platforms out there with open-source bots, but most of them are very weak (we tried many of them ourselves before deciding to build Airbag, and we realized we always wanted to modify them and make them much more intelligent), and we preferred to build one thing that works.
However it is only fair that even if proprietary, we share with you what the approach is. We will be sharing this as a technical blog post in the next few weeks. The blog post will share some of the problems we have faced when building a deep neural network with thousands of inputs, and some of the backtesting results.
As a sneak peak, this is the current backtesting for ETH/BTC, with an algorithm that has no single “If statements”. The algo learns to become better at winning, as we give it cookies when it makes good trades with Q-Learning (The pic below is performance on unseen data)
We can also tell you that we will move from 1-hour candles to 1-minute candles, which will increase the precision (you may have noticed that right now the bot was always buying at the same times of the day). The 1-hour candles were a proxy to get started fast.
Also, we have been backtesting several coins, so that we can build a diversified portfolio. While the AI should not care about which coins it trades, the reality is that BTC has very high liquidity while other coins have larger spreads and illiquidity, and so we need to do more testing.
We don’t have a final date for launch of the new algorithm. But we will notify it on the app.
6. Marketing
Many of you have come through friends, others through our tiny marketing campaign. We have had no PR, and we don’t know famous crypto bloggers or groups, and we are not on lists of “bot comparisons” yet.
If you think that what we are doing is worthwhile, we would appreciate:
- Any intros or comments on telegram or discord groups,
- recommendations or introductions to bloggers or cryptojournalists
- reviews on cryptocurrency communities, reddit, etc.
- or just invite your close friends.
Please help us grow so that we can invest more time building the best service and algorithms for you and your friends.
That’s it for now, thanks so much again. If you have any other suggestions just reach out or join our Telegram.
Your Airbag Team
https://airbag.ai
