The data feeds from dozens of cryptocurrency exchanges generate gigabytes of data every day, comprising trade executions, quotes (orders) entry and modification. This huge amount of data makes in-depth analysis impractical, beyond any professional chartist or quantitative trader capabilities. Fortunately, advances in artificial intelligence, machine learning in particular, and parallel computing give rise to new methods to extract information from big data sets, that human eyes cannot see.
Our experts in computer science, parallel computing and mathematics created a new proprietary approach for the identification of market patterns through so-called “markers”. These markers of different units and dimensions are analysed using deep learning methods inferring insight from markets.
The information derived by our tools is meant to be used in conjunction with customer's market view. From this starting point, we provide additional insight which can be used to support her initial idea.
To further simplify the usage of our tools, we standardized our model output introducing pre-defined parameters, such as timeframe, market view (bullish, bearish), and so on. Our deep learning models are regularly trained on new market data. We value their predictive nature. Therefore, we perform out-of-sample testing and monitor the performance on a regular basis as well.
The use of recent technologies, such as FPGAs, allows us to achieve seamless and near real time market insight. In contrast to conventional CPU based code execution, our approach guarantees market inference in near real time, enabling trading those insights.