Using Simplified Wavelet Filters to Smooth Crypto Price Trends
Summary
The document introduces wavelets as localized, multi-scale filters for separating slower price movement from faster fluctuations. It demonstrates a simplified approach that applies selected coefficient sets as weighted smoothing filters, varies the sampling step or repeated smoothing level, and uses the resulting curve’s direction as a trend signal. Seven named wavelet families are compared conceptually by window length, symmetry, negative weights, smoothness, and lag. The examples use cryptocurrency prices and describe choosing smoothing levels to suit the intended holding period.
The article explicitly says its method omits key parts of formal wavelet analysis, including multi-level decomposition, threshold denoising, inverse reconstruction, and careful boundary treatment. It presents illustrative examples and code but does not provide quantified performance evidence or a robust out-of-sample evaluation. Smoothing can delay signals, and the simple rule of going long or short based on the latest smoothed-price change is not shown to overcome costs, whipsaws, or overfitting; the author frames the method as exploratory rather than research-grade.
Key ideas
- Wavelet filters can smooth price series at different time scales, with stronger smoothing generally introducing more lag.
- The article’s implementation uses weighted convolutions as a simplified trend extractor rather than a complete wavelet decomposition and reconstruction.
- Filter choice, symmetry, coefficient signs, and smoothing level affect responsiveness and the shapes of the resulting signals.
- The suggested trend rule takes positions according to the direction of the smoothed price, but the document does not establish its profitability.
- Boundary effects and omitted denoising steps limit the method’s suitability for rigorous research without further validation.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.