Fintech solution for automated Forex trading based on self-learning predictive ai models

Development with regional partners supported by HAMAG-BICRO is currently in the phase 4 of the project. The product launch is planned for October 2022.


FX-AI is a FINTECH software solution that enables users connected to international currency markets with market competitiveness based on the application of machine learning models for predicting trading conditions and implementing a smart automated trading strategy. Based on artificial intelligence technologies and methodologies (AI), this product enables the prediction of index and price trends and the optimization and automation of trading in capital markets.

Due to the volatility of price index movements within the time series of targeted capital markets for which we profile our “FX-AI” solution, so far developed 1st generation machine learning model, despite the application of extensive domain experience and conducted analysis and engineering features, can not achieve a stable level profitability. The reason for this is the latency in learning the characteristics of changing trends as well as changing trends of interdependence of different extrapolated features as well as the weight of their impact on future changes. Therefore, it was found that it is necessary to apply the models of deep enhanced learning, which began to be experimented with several years ago. By implementing project activities, we want to create a methodology for building models that will be much more sustainable, flexible, and in the application more robust and durable, ie more self-sustainable in production application compared to existing solutions.

Target customers: financial institutions such as banks, equity funds and brokerage houses, capital market advisory offices and other FINTECH companies.


Momentum strategy

  • Data Preprocessing includes 10 Gb dataset per currency pair
  • Implementation of Crossover Strategy
  • Calibrating Influence of λ parameter on indicator structure
  • Combining momentum change indicator with momentum direction indicator
  • Clustering of market periods (Positive & Negative trends and Extreme events)
  • Implementation of Optimal negative weight moving average (ONWMA)
  • Implementation of forexcalendar news for period 2007-2021

Trading  strategy

  • Risk management
  • Money management
  • Implementation of Imitation learning policies (DAGGER algorithm)

Feature pipeline

  • Constructed 364 features forming 11 group types:
    • n-minute returns
    • summed exponentially decayed returns
    • High − Low spread
    • Technical indicators: RSI, CCI, ADX, CMO, MACD, STOCH, ULTOSC, ATR
  • Features are calculated for both raw and filtered prices

Research and development activities for the 2nd generation of automated trading solutions in the foreign exchange markets began in March 2020.

Activities of the 3rd phase of research and development – October 2021 – January 2022

Research of FX-AI machine learning model for continuous trading

  • The model is trained on data of currency pairs (conversion of data from the limit order book (LOB) into open, high, low and closing minute prices (OHLC), interpolation of missing values, filtering unpredictable moments directly related to the Forex calendar, ie events announced and relevant to currency pair trading).
  • Algorithm for exchange rate marking has been made.
  • More than 360 features have been selected for the machine learning model.
  • Adequate machine learning models for currency pair trading have been selected: supervised learning, decision trees, random forests, deep neural networks, supported learning.

Research on a trading strategy based on the FX-AI model

  • Interpretation of the output of the machine learning model in the context of the probability of performing trading actions (buying, selling), defining possible positions (short, long, neutral), determining risk willingness (stop-loss limit) and profit-taking limit (take-profit) that insure us in highly volatile trading periods).
  • Research of different money management strategies, ie the amount of money invested depending on the output of the machine learning model.

Development of a production model for continuous testing and self-learning

  • Implementation of REST application software interface for FX-AI model
  • Defining architecture and endpoints.
  • Generate features from the real-time currency pair exchange rate reverse window.
  • Predicting the probability of future action in a trading strategy (introduction of pre-trained models and real-time evaluation).
  • Test the implementation of the FX-AI model and trading strategy in a real environment via the MetaTrader5 platform.

Development is currently in Phase 4 – Target Environment Demonstrations (TRL5).

  • Introduction of calibrated models and real-time evaluation

Continuation plan:

  • Phase 5 – Target Environment Testing (TRL6) – April-May 2022
  • Phase 6 – first implementation in production (TRL7) – May-June 2022

The commercialization of the solution is planned for the 2nd half of 2022.


Project team skills:

Expert knowledge and experience in developing deep neural networks, establishing self-learning systems, handling multi-dimensional financial time series, quantitative methods, feature engineering using technical indicators, statistical arbitrage, trend feature development and economic and financial data analysis .


To develop the project, the company applied the skills of traders in capital markets with more than 10 years of experience combined with the expertise of development teams that have been developing machine learning solutions for more than 10 years for clients in the financial sector (banks, brokerages, etc.). and To develop and train enhanced Deep Reinforcement Learning models on data sets larger than 1TB, the company has provided access to supercomputer resources.

  • IDEUS d.o.o. Slovenia – client “early-adopter”
  • Faculty of Electrical Engineering and Computing (FER), University of Zagreb – model research and development
The project was co-financed by the European Union from the European Regional Development Fund within the Operational Programme Competitiveness and cohesion

Project name:

Research and development of models for FX-AI solution – phase 3


Development of machine learning models for prediction of conditions and determination of strategy in automated trading on international capital markets with increased volatility.

Project goals:

Improve and refine existing models by applying innovative methods and technologies of machine learning, which will enable the launch of the FINTECH software solution “FX-AI”.

Overall funding:

19,760 EUR (148,210.55 HRK)  –  total EU funding: 10,000 EUR (75,000 HRK)

Implementation period:

21.10.2021 – 19.01.2022

Contact person:

Ivan Nikolić, project coordinator

What We Offer

We appreciate your trust greatly. Our clients choose us and our services because they know we are the best.

Our Customers

Trusted by leading global companies 48+ customers in 9+ countries

Ready to get started?

Don’t Hesitate to Contact Us

Please enter your name!
Please provide a valid email address!
Please enter your message!