Reinforcement Learning Trader with GMDH Agents and Normalized Prices

Summary

This document describes a trading system built around reinforcement-learning agents implemented in an MQL library. An agent accepts a vector of predictor values and contains an ensemble of trees; the example supplies normalized closing prices as inputs and obtains a trade signal from the agent. It says training is performed in a single tester pass with learning enabled, then learning is disabled for subsequent operation.

The page mentions a demonstration on training and test samples, but gives no performance figures, validation design, trading rules, or details about reward construction and data handling. It points to a separate article for the underlying idea and simplest algorithm, so the method is only outlined here. The description does not establish that the approach generalizes beyond the samples or accounts for transaction costs and overfitting.

Key ideas

• The system uses reinforcement-learning agents that process predictor vectors and return trade signals.
• The example fills agent inputs with normalized historical closing prices.
• Training is described as a single tester pass with learning enabled, followed by disabling learning.
• The document gives no metrics or validation details to assess out-of-sample performance.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.