Hands-On Machine Learning for Algorithmic Trading

How to compute rolling window statistics

Given the sequential ordering of time series data, it is natural to compute familiar descriptive statistics for periods of a given length to detect stability or changes in behavior and obtain a smoothed representation that captures systematic aspects while filtering out the noise.

Rolling window statistics serve this process: they produce a new time series where each data point represents a summary statistic computed for a certain period of the original data. Moving averages are the most familiar example. The original data points can enter the computation with equal weights, or using weights to, for example, emphasize more recent data points. Exponential moving averages recursively compute weights ...