US CPURNSA PCA Diagnostics

Reseed-aware rolling analysis. PCA uses a 60-day window of posterior curves. Projected nodes are aligned by absolute reference month, so a published front node is not compared with a forecast node after reseeding. Valid states before and after a reseed remain in the window; the reseed transition itself is not treated as a market move.

Current 60-day window: explained variance

Natural units use projected CPI index levels on the common reference-month set. Standardized units give each remaining node equal marginal volatility before PCA.

Window coverage across coordinate frames

The table shows which posterior states contribute to the rolling window. PCA percentages are calculated once on the aligned window, not independently on each reseed frame.

Method and limitations

PCA is a linear variance-concentration diagnostic, not a causal model and not posterior uncertainty. Independent-component analysis (ICA) is not used here: with short, correlated posterior paths it is unstable and does not provide a comparable “variance explained” measure.