Threshold Signals from Normalized Price Changes and Activation Functions
Summary
This strategy transforms price changes across a selected timeframe into a normalized signal, then applies an activation function such as Swish, leaky ReLU, or a step function. It enters long when the processed value crosses above a positive threshold and short when it crosses below the corresponding negative threshold. The document also describes configurable price inputs, lookback and scaling parameters, optional volume use, and chart plots for the signal and thresholds.
The proposed benefit is that nonlinear transformations may smooth noise or change which movements pass the thresholds. However, the document supplies no measured performance results, and its claimed signal improvements are not supported by reported tests. Threshold and activation choices can discard useful information or miss trades. A further concern appears in the source: the default old system uses higher timeframe data with lookahead enabled, which can make historical signals differ from signals available in real time. Disabling repainting is discussed, but the default configuration makes careful validation essential.
Key ideas
- The method derives a signal from normalized price changes across a configurable timeframe.
- Long and short entries occur when the transformed signal crosses positive or negative thresholds.
- Swish, leaky ReLU, step, and no activation are available as signal transformations.
- Threshold selection and activation choice can alter signals and may filter out useful market information.
- The default old system requests higher timeframe data with lookahead enabled, creating a repainting concern.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.