AI trading differs from forex robots or similar automated trading systems. A key feature of automated AI trading, compared to merely automating something, is its ability to understand which are the best and worst decisions to be made, for a given task and data set. The Alpha, however, is not available for purchase as it is an exclusive AI system designed for the Alpha Union internal use.
Using technical analysis, our algorithm not only considers the patterns of various trends, including support and resistance levels, and cross-indicators, but our AI trading system is also able to create its own real-time index for each currency pair that is used to identify the best direction and target.
Concerning fundamental analysis, the AI makes a correlation within the economic calendar to find news or data that can affect specific currency pairs. In doing so, the algorithm can identify the following scenarios:
Market inactivity to not trade or send signals.
Avoidance of certain risky entries.
Measure favourable patterns after-events.
Our algorithm is a single multi-layer system integrated with some of the largest forex brokers via rest API and FIX protocol, and its parameters are optimised continuously to reach the best APY, Sharpe Ratio, and Drawdown. However, it does not change its core rules over time. We take extra precaution to avoid overfitting our models to any price data.
We use two main approaches to AI. The first is a search of maximum return and minimum drawdown using Multi-Objective Particle Swarm Optimisation, which can validate our indicators using a sliding window method to guarantee the best parameters for a certain encoding of indicators.
In this way, we selected the set of indicators to use in our second AI, settling a time series classification via Convolutional Neural Network, which the architecture is similar to an AlexNet with data augmentation, based on our own math concepts. This second AI indicates if it is the buy, sell, or hold, for given security.
Our most effective indicators are made by ourselves, primarily developed in Python as a real-time algorithm and based on out-of-sample back-testing. To evaluate our model, we chose the MQL language to demonstrate our results on the MetaTrader 5 (MT5) platform. Our main libraries are Matplotlib, NumPy, Pandas, Parallelism with ray remote, and Talib.
We are a technology service provider that provides trading solutions. You can reach us by sending an email if you have any inquiry and we will respond to you as soon as possible.
20 Leslie St Toronto, ON,
Canada M4M3L4