GLASS BOXES FOR THE FOREIGN EXCHANGE MARKET: ARCHITECTURE AND PROPERTIES OF ADDITIVE INTERPRETED MODELS IN AUTOMATED TRADING

Klimin Pavel Yurievich

The paper considers the application of glass (intrinsically interpretable) models to the tasks of developing automated trading systems in the foreign exchange market. An architecture for building a trading robot is proposed, in which the core is a generalized additive model with tree boosting (Explicable Boosting Machine, EBM). The model is trained in Python, exported to ONNX format, and then used in the MQL5 Expert Advisor to generate trading signals in real time. It is shown that the additive representation It allows you to simultaneously ensure high-quality forecasting of the direction of price movement and a transparent decomposition of the contribution of features to each trading decision. Interpretability metrics specialized for financial time series are discussed: the stability of the forms of functions, the consistency of the signs of effects with economic intuition, and the "turnover" of factors that determine transactions. The final section connects the theoretical constructions with the practical results of testing the adviser on historical data.

Keywords:explicable AI, glass model, Explicable Boosting Machine, ONNX, MQL5, foreign exchange market, algorithmic trading, automated trading systems.