Wavelets Reveal Changes: Causal Forecasting with Haar Features

Day 2 of the Short Series: Signal Processing Tools for Time Series

Fourier methods are useful when a time series contains stable cycles. But many real signals are not mainly cyclical. They jump, drift, break, spike, and change regimes. Wavelets are designed for this kind of structure.

This article introduces a simple causal wavelet-style forecasting method using Haar features. The method detects changes at several time …