It is time for small investors and retail traders to leverage the state-of-the-art technology used in hedge funds. We offer a no-code platform for automatic trading system (ATS) designed with the algorithms used in most recent breakthroughs in AI (GO, DotA 2, Starcraft II etc). We first tackle the cryptocurrencies domain. Usually, such strategies are manually designed by experts with a set of intricate rules. Our technology, based on Reinforcement Learning (RL) - a subset of Machine Learning (ML) - derives those rules optimally from data with zero effort. Our platform facilitates the training and testing of the bots (or agents) in unknown conditions, certifying its performances, and because trading is not about creating a one-fits-all strategy, the user can also monitor and update his bots. Finally, the user can sell his bots or even scale his earnings using copy trading, or pooling.
We are engineers by training.
Nicolas Carrara is a postdoctoral researcher at the University of Toronto specialized in real life
including dialogue systems, autonomous driving, and traffic signal control. He is also specialized in
learning, safe RL, and multi-agents RL (MARL). He has published at international conferences including NeurIPS
UAI. He did several
industry including Google, Microsoft and Huawei.
Bachir Arif is tech lead at the Société Générale. Through his career, he has built a network of finance professionals including traders, portfolio managers and quants. At LIST he did research on blockchains, especially in cryptocurrencies. While being at Deloitte, he has developed multiples decentralized applications (Dapp).
We both bring to the company a combination of cutting edge technologies and domain expertise in order to design the perfect fintech product, with an extremely high barrier of entry (ML and blockchain).
We have identified three main problems in the retail trading space.
The majority of people is losing money trading (more than 80% on the forex). The reason is simple, unlike investing, trading is almost a zero-sum game, meaning if someone makes 5 dollars, another one loses 5 dollars. On this game, individuals have no edge against investment banks and hedge funds. Individuals usually trade manually, are subject to their own emotions, and it is a repetitive task. The trading frequency is also very slow. A lot of competitors offer automation solutions to design one's own trading strategy, but creating a trading strategy is hard, even assuming oracle-like directional information like "the price is going to go up".
All those questions can not be answered with classical ML. Most advanced hedge funds usually rely on ML models predicting the price, sometimes the volume, this is called forecasting; but they do not create automatically a trading strategy given this information, also called execution. It is usually the role of the trader, or the quant-trader, to find the trading strategy by combining a set of different indicators, ML models, and "if-then-else" rules. This is where RL shines, it can find the optimal actions to take in a given context, using past information, and anticipating the future, for optimal returns, solely using data. Simply put, RL will find the rules automatically, faster, and better than an expert.
If your trading strategy was optimal yesterday, it might be not the case today. Market adapts, it evolves, or more precisely, market participants update their strategies. To keep up, financial institutions hires hundreds of traders and quants to update theirs models on a regular basis. In the meantime, retail trades usually have a single strategy, never updated. Hence, it is impossible to compete. By nature, Reinforcement Learning solves this problem. Indeed, the more data the trading agent ingests, the more up to date and optimal the strategy will be.
In order to generate passive income, retail investors might be tempted to buy or rent a trading bot, or follow a trader on Telegram or Discord. It is simply impossible to discriminate between scams and legit services. In the telegram space, it is not uncommon that influencer traders encourage their followers to buy a particular stock for their own profit (see pump and dump). The bot market is saturated with offers promising high yield and unrealistic performances. Those bots might perform worse than the S&P500. This fallacy is particularly true in the crypto market, where the bullish trend (meaning the coin value is mostly increasing) gives the illusion that all strategies perform well. Trading is about alpha, meaning how much better the strategy performs against the overall market. At Myrill.io, we intend to certify bots' alpha and bots' origin using blockchain-related technologies.
We provide a no-code tool for ATS creation and certification. It is a platform based upon a modular pipeline.
The pipeline includes several modules, each of them can be customized, see the figure below:
Existing solutions in most recent fintech trading startups (including Y-Combinator backed startups) usually provides 5 and a limited version of 1 (no alternative data). The module 2 and 3 must be designed by hand using a set of rules and indicators. Some specialized companies provide 2, and we intend to integrate their services and offer our own. Nobody provides 3 and 4 as far as we know. Some companies offer black-box bots services, but are rule-based strategies under the hood (hence non-optimal).
Our architecture supports market-making (providing liquidity, i.e. enough instruments to be sold or bought on an exchange), arbitrage (buy 5 BTC on Binance and sell 5 BTC on Coinbase) and directional (price is going up, I buy).
Once a bot has been designed and trained, we associate a Non-Fungible Token (NFT) to it. We call it Myrmidon. We develop a Dapp to manage and store it. The Dapp interacts with the Solana blockchain, and we are working exclusively with smart contracts to guarantee verified data. The NFT certifies several properties of the bot:
This certification mechanism unlocks a market for true and unaltered trading bots. This way, users can rent, buy, or invest into myrmidons without having to worry about scams and performance issues.
Myrill is an ML as a service (MLaaS). User interact with a web interface or a mobile app in order to design, train and evaluate a trading bot. The figure bellow shows each step of the user story:
It goes as follows:
Myrill's Myrmidons is a collection of 100 uniquely generated NFT crypto trading bots stored on the Solana blockchain.
On top of a neat pixel artwork, each myrmidon unlocks several perks:
Each myrmidon has a unique investor profile. It is constructed following several parameters:
We are also selling 900 tokens "Myrill.io". As myrmidons, they give early-access to the Myrill's platform for at least a year before public opening. The token will also give privileged access to live events, learning contents, and merchandises.The auction starting price of a myrill.io token is 1 SOL.
Late 2021, we will run auctions for ten myrmidons each week. An auction lasts for one week. All myrmidons can be resold on the Myrill marketplace.
All Myrill.io tokens will be open for auctions at the same time.
Myrmidon#1 will be the first to be available for live trading in Q4 2022, then others will be release each month until Q4 2023. See the detailed roadmap here.