Combining a Multiscale High-Pass Filter with an Adaptive Classifier

Summary

This indicator combines a multiscale high-pass filter with a simple adaptive binary classifier to produce a smoothed directional signal. The filter compares fast and slow components, reduces noise, adjusts the result for ATR, and normalizes it over a rolling lookback. The classifier uses momentum, volatility, directional movement, oscillator position, price velocity, and a Supertrend-based resistance direction as binary features.

The classifier updates feature weights using prediction error against a target derived from standardized closing price, then fuses its output with the filter when both components are enabled. The final signal is smoothed and scaled using recent versus longer-term volatility; zero crossings generate directional markers and alerts. The source provides implementation details but no backtest, performance statistics, or evidence that the classifier predicts future returns. Its target is based on contemporaneous price standardization, and its hand-set feature weights and update rule warrant careful out-of-sample validation before use.

Key ideas

• A fast and slow high-pass component are differenced, smoothed, ATR-adjusted, and normalized into a bounded filter signal.
• The classifier combines six technical features and adapts feature weights using a prediction error update.
• The filter and classifier can be used separately or averaged, with a momentum fallback if both are disabled.
• The final output is smoothed and volatility-scaled, and signal direction changes trigger chart markers and alerts.
• The source gives no trading performance evidence, so predictive value and robustness remain unestablished.

Tags

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