Using EWMA to Balance Responsiveness and Stability in Market Estimates
Summary
The article explains how exponentially weighted moving averages (EWMAs) give more influence to recent observations while retaining a diminishing contribution from older data. It motivates the method with changing correlations between SPY and TLT: full-history averages suggest negative correlation, while rolling 90-day and 500-day estimates vary over time and differ in responsiveness. Short windows react quickly but can be noisy; longer windows are steadier but may lag. The author demonstrates EWMA calculations for prices and correlations, including separate decay factors for covariance and variance estimates, and discusses a warmup period because early values rely on little history. An Rcpp implementation is offered for faster computation when processing higher-frequency data or many estimates. The examples use historical market data, but they do not establish that one decay factor or estimation method is universally best. The choice depends on the task and the balance required between reacting to changes and smoothing noise.
Key ideas
- An EWMA gives recent observations more weight while the influence of older observations decays over time.
- Short rolling windows react quickly to changing relationships but can produce unstable estimates.
- Longer windows smooth noise but may respond too slowly to market changes.
- A warmup period can reduce the influence of unreliable early EWMA values.
- Exponentially weighted covariance and variance estimates can be combined to estimate changing correlations.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.