[scatter-lag] Lag Plot for Time Series Autocorrelation Diagnosis

[MarkusNeusinger]

Description

A lag plot is a scatter plot of a time series against a lagged version of itself (y(t) vs y(t-k)). If the data is random, points scatter uniformly; if autocorrelated, they form patterns (linear for AR processes, elliptical for seasonal). This is a quick visual diagnostic tool complementing the existing acf-pacf spec.

Applications

• Checking for autocorrelation before applying regression models
• Diagnosing stationarity in financial time series
• Identifying seasonal patterns in sensor data
• Validating residual independence after model fitting

Data

• value (float) — time series values
• lag (int) — lag order k (default 1)
• Size: 100–5000 observations

Notes

• Default lag = 1, but should support configurable lag
• Add diagonal reference line (y = x) for visual comparison
• Points colored by time for temporal context (optional)
• Strong linear pattern = high autocorrelation at that lag

[claude]

✅ Spec ID: scatter-lag

Rationale: A lag plot is a scatter plot of y(t) vs y(t-k) used to diagnose autocorrelation — the scatter type with lag variant captures this well. This complements the existing acf-pacf spec (which shows correlation coefficients as bars) by providing the raw point-level diagnostic view. No duplicate or near-duplicate exists.

Add label approved to begin generation.

[github-actions]

✅ Specification Ready: scatter-lag

PR: #5264

---

specification.md

scatter-lag: Lag Plot for Time Series Autocorrelation Diagnosis

Description

A lag plot is a scatter plot of a time series against a lagged version of itself, plotting y(t) on the x-axis versus y(t+k) on the y-axis for a given lag order k. If the data is purely random, points scatter uniformly with no visible structure; if autocorrelation is present, distinctive patterns emerge — linear clusters for autoregressive processes, elliptical shapes for seasonal data. This provides a quick visual diagnostic for time series dependence, complementing numerical tools like ACF/PACF.

Applications

• Checking for autocorrelation before applying regression models that assume independent residuals
• Diagnosing stationarity and serial dependence in financial return series
• Identifying seasonal or cyclical patterns in sensor and environmental monitoring data
• Validating residual independence after fitting ARIMA or other time series models

Data

• value (float) — time series observations in chronological order
• lag (int) — lag order k, default 1 (plot y(t) vs y(t+k))
• Size: 100–5000 observations
• Example: daily stock returns, hourly temperature readings, or synthetic AR(1) process data

Notes

• Default lag = 1, but the implementation should support configurable lag values (e.g., 1, 7, 12)
• Include a diagonal reference line (y = x) to help assess whether the series is uncorrelated
• Optionally color points by their time index to reveal temporal structure within the scatter
• Strong linear pattern along the diagonal indicates high positive autocorrelation at the given lag; perpendicular spread indicates negative autocorrelation
• Consider adding a correlation coefficient annotation (r value) to quantify the visual pattern

---

Next: Add approved label to merge this specification to main.

---

🤖 spec-create workflow

[github-actions]

🚀 Specification Merged!

Specification: scatter-lag
PR: #5264 (merged)

The specification is now in main and synced to PostgreSQL.

---

Next: Add generate:{library} labels to start generating implementations.

Available libraries: matplotlib, seaborn, plotly, bokeh, altair, plotnine, pygal, highcharts, letsplot

Or use workflow dispatch for bulk operations.

---

🤖 spec-create workflow