R中13ARIMA-SEATS季节性调整问题：序列未变及检测方法

Hey there, let's tackle your two questions about seasonal adjustment of German inflation data—super common pitfalls, so great you're digging into this!

There are several reliable ways to check for seasonality in a time series, split between visual and statistical approaches:

Visual Methods #

• Line Plot: Plot the raw time series and look for repeating peaks/troughs that align with seasonal cycles (e.g., every 12 months for monthly inflation data).
• Seasonal Subseries Plot: Group observations by their season (e.g., all January values, all February values) and plot them side-by-side. This makes it easy to spot consistent seasonal patterns. In R, use forecast::ggseasonplot() for a polished version:

  library(forecast)
  ggseasonplot(your_inflation_ts) # your_inflation_ts is your ts object with frequency=12
  

• Seasonal Boxplots: Create boxplots grouped by season (e.g., month) to compare distributions across seasons. Significant differences in median/quartiles indicate seasonality.

Statistical Tests & Diagnostics #

• KPSS Seasonality Test: A formal test to detect seasonal unit roots (a sign of strong seasonality). Use tseries::kpss.test() with the seasonal null hypothesis:

  library(tseries)
  kpss.test(your_inflation_ts, null = "seasonal")
  

• ACF/PACF Analysis: Check the Autocorrelation Function (ACF) plot for significant spikes at seasonal lags (e.g., lag 12, 24 for monthly data). A significant spike at lag 12 means monthly seasonality is present.
• seasonal Package Diagnostics: When you run seasonal::seas() on your data, the output includes automated seasonality checks. The summary will explicitly state if significant seasonality was detected.

If your X-13ARIMA-SEATS adjustment (I assume you meant the standard X-13ARIMA-SEATS method, not 13ARIMA-SEATS) returns the exact same sequence, here are the most likely culprits:

• No Significant Seasonality in the Raw Data: The X-13ARIMA-SEATS algorithm first tests for seasonality. If it doesn't detect a statistically significant seasonal pattern, it will return the original series without adjustments. Use the diagnostics from seas() to confirm this.
• Incorrect Seasonal Frequency: German inflation is monthly, so the correct frequency is 12. If you accidentally set frequency=13 when creating your ts object, the method can't identify the true seasonal cycle, so it won't adjust the data. Double-check your time series definition:

  # Correct monthly ts definition
  inflation_ts <- ts(raw_data, frequency = 12, start = c(YYYY, MM))
  

• Misconfigured seas() Parameters: If you passed arguments that disable seasonal adjustment (e.g., seasonal="none"), the method will skip the adjustment step. Review your function call to ensure you're not overriding the default behavior.
• Data Was Already Seasonally Adjusted: If your "raw" data is actually already a seasonally adjusted series (common with some economic datasets), re-running the adjustment will leave it unchanged. Check the data source documentation to confirm.
• Negligible Adjustment Magnitude: In rare cases, the detected seasonal component might be extremely small—so small that it doesn't show up in the rounded values of your series. You can inspect the seasonal component directly with seas_diagnosis$series$sa to see if tiny adjustments were applied.

内容的提问来源于stack exchange，提问作者Sean