UKF and Capsule Network Signals for Noisy Markets

Summary

The article describes an MQL5 Wizard signal class that combines an Unscented Kalman Filter (UKF) with a Capsule Network. The UKF estimates a hidden, potentially nonlinear price state from noisy observations, aiming to avoid some lag associated with conventional smoothing indicators. A CapsNet then acts as a structural validator, assessing whether the estimated state and proposed direction align with features such as RSI momentum and ATR volatility. The class can be configured to use the UKF alone or together with the CapsNet.

The intended setting is a noisy market where clean regime changes or repeated patterns are hard to identify, including short-timeframe trading. The article reports Strategy Tester examples, including a win rate above 90%, but also says the testing window and single-symbol scope are limited. The tests used no stop-loss, and the reported outcome was affected by one losing trade. The results are presented as a demonstration of the model and Wizard workflow, not evidence of a deployment-ready strategy; systematic risk controls and broader testing remain necessary.

Key ideas

• The UKF estimates a latent price state by accounting for nonlinear transitions and observation noise.
• The CapsNet is used to validate whether price state, momentum, and volatility features form a coherent setup.
• The signal class allows testing the UKF with or without the capsule network filter.
• The model is proposed for noisy environments where conventional indicators may lag or be whipsawed.
• The reported tests are limited to a short window and one symbol, and they omit stop-loss protection.

Tags

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