Causal Online Neural Network for Daily Long-or-Cash Signals

Summary

This indicator implements a small online neural network that produces long-or-cash signals from six daily market features. The features describe relative moving-average level, RSI-based slope, distance from a regression mean, directional efficiency, volatility relative to its average, and candle pressure adjusted for relative volume. Once the forecast horizon has elapsed and the outcome is known, the model trains on the earlier feature vector. Adjustable response and selectivity settings change its learning horizon, adaptation speed, and signal confirmation requirements. The script is intended for confirmed daily bars and can reject other chart intervals.

It also includes a structural comparison, a live skill audit, and a dashboard comparing model equity and drawdown with buy-and-hold from a shared start date, with a user-set transaction cost. These are features of the indicator, not reported findings: the supplied excerpt contains no measured results or independent validation. Its claims about causal updating and confirmed states describe the implementation approach, but do not establish predictive value or robustness across assets and market regimes.

Key ideas

• The model uses six normalized price, trend, volatility, and volume-related features to generate a daily signal.
• It trains on an earlier feature vector only after the selected forecast horizon has elapsed.
• Response settings adjust learning behavior, while selectivity settings adjust transition thresholds and confirmations.
• The indicator includes a model-versus-buy-and-hold audit with configurable transaction costs.
• The document provides no empirical results establishing predictive performance.

Tags

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