Using a Beta-VAE to Refine Stochastic and FrAMA Signals

Summary

The article applies a beta variational autoencoder to binary event features built from the Stochastic Oscillator and Fractal Adaptive Moving Average (FrAMA). It focuses on three patterns that had underperformed in earlier tests, encoding indicator conditions as yes-or-no inputs, training models on historical data for gold, the S&P 500, and USD/JPY, then exporting them for use in an MQL5 Expert Advisor. The approach aims to learn useful combinations of events in a compressed latent representation while retaining explicit indicator logic in preprocessing.

The reported forward-walk results show a modest improvement for gold, while the S&P 500 and USD/JPY remained below breakeven. The discussion contrasts binary features with continuous inputs, noting that binary encoding can clarify event definitions but still depends on sound feature construction. The author stresses that the test window is limited and that apparent gains may not persist across market regimes. The results are exploratory and require independent, longer-term evaluation before deployment.

Key ideas

• The model learns from binary event vectors derived from Stochastic and FrAMA conditions.
• The experiment targets three indicator patterns that had lagged in earlier walk-forward tests.
• Reported gains were modest and limited to gold among the three tested markets.
• Binary inputs make event logic explicit but do not remove the need for careful feature design.
• The short test window makes the results preliminary rather than evidence of durable profitability.

Tags

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