Testing and Transforming Time Series for Stationarity
Summary
The article explains stationarity as stability in a time series’ statistical behavior over time, distinguishing strict, first-order, weak, trend, and difference stationarity. It describes why the assumption matters: many forecasting procedures rely on past behavior being informative about future behavior, while trends, seasonality, or changing variance can make estimates unreliable. Financial price levels are presented as a common non-stationary example.
Suggested diagnostics include plotting the series, comparing summary statistics across segments, inspecting autocorrelation patterns, and applying formal tests such as ADF and KPSS. The article also mentions differencing and logarithms as ways to address trend or variance changes. It illustrates the discussion with stock price and daily price-change plots and reports that price levels have different behavior across time windows. Visual and partition-based checks are preliminary; the excerpt provides no complete test results or trading strategy, and stationarity does not itself establish predictive value.
Key ideas
- Stationarity means a series’ statistical properties remain stable over time, though definitions differ in which properties must be stable.
- Trends, seasonality, or changing variance can make a series non-stationary and weaken model inference.
- Plots, segmented summary statistics, autocorrelation, ADF, and KPSS can help assess stationarity.
- Differencing can remove trends, while logarithms may help stabilize variance.
- Stationarity checks are diagnostics and do not establish that a series is useful for forecasting.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.