Machine Learning Process
The following are some of the steps involved in building our machine learning-based trading strategies.
We are a team of data scientists and traders who believe that machine learning can help us make more informed decisions.
We have developed 7 steps for building our AI-driven strategies, which you will find outlined below.
Data is the backbone of our investing process. We use a variety of sources to help us begin to make informed decisions. These sources include order book and exchange information like funding rates, macroeconomic trends that can impact the global markets, social media sentiment, and crypto on-chain analytics.
We summarize and clean the data to remove noise while identifying key features or trends. During this step, we seek exploitable edges in the market that can be used for strategy formulation.
We use historical data to validate the effectiveness of our strategies thesis. If it looks promising, then optimization and fine-tuning can begin in order for us to find out which features are useful by running backtests on them throughout time periods that span from months all the way up to years.
The signals from our strategy thesis are combined with the results of a backtest and input into machine learning models that determine whether or not to act on future opportunities.
Using machine learning techniques, we are able to determine how much money should optimally be bet on each trade. Our risk management model draws from the Kelly criterion for betting and order flow toxicity as well other factors such as on-chain movements or general liquidity conditions in markets at large when building our "emergency trigger".
Once we have optimized a strategy for success, the next step is to run a live test with some capital. We monitor how well our algorithm performs in real-time on live market data and then pick out strategy sets of winners based on that performance. This process allows us to get an idea if there are any areas where improvements need to be made before going full steam ahead.
Once we deploy our strategies, our quantitative research team is on the lookout for any sign that the algorithms may not be working as they should. If we see something, then we immediately put into action solutions so that trading can continue smoothly.