Combining DeMarker and Envelopes Patterns with an RNN

Summary

The article describes custom Python implementations of DeMarker and price Envelopes, then turns combinations of their signals into features for a recurrent neural network using a white-noise kernel. DeMarker is calculated from smoothed increases in highs and decreases in lows, yielding an oscillator interpreted through overbought and oversold zones. Envelopes place percentage-offset bands around a moving average, with band contacts treated as possible breakout or reversion contexts. The author notes that rolling calculations create missing initial values and that the oscillator calculation should guard against division by zero.

Six patterns previously identified as suitable for forward walking are trained using data from 2023 at a four-hour interval, then evaluated over a period spanning 2023 to 2025. The reported results say that three patterns walk forward and three do not. The article presents this as an exploratory machine-learning workflow rather than conclusive evidence of a profitable strategy: it gives no performance metrics in the supplied text, and indicator thresholds, band settings, training choices, and validation design can affect outcomes.

Key ideas

• DeMarker compares smoothed upward high changes with combined upward and downward extremes to form a bounded momentum oscillator.
• Price Envelopes place fixed percentage bands around a moving average and can frame both breakout and reversion hypotheses.
• The article combines the two indicators into six patterns used as inputs to a recurrent neural network.
• Training uses 2023 four-hour data, and subsequent evaluation reports forward walking for three patterns and failure for three.
• The reported pattern results do not establish profitability, and the article provides no detailed performance metrics in the supplied text.

Tags

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